@dreb/ai 2.70.1 → 2.72.0

package/dist/models.js

@@ -116,7 +116,7 @@ export function calculateCost(model, usage) {
 }
 /** Parse Claude family versions without mistaking long date suffixes for versions. */
 function claudeFamilyVersion(modelId) {
-    const match = modelId.match(/(opus|sonnet)-(\d{1,2})(?:[.-](\d{1,2}))?(?!\d)/);
+    const match = modelId.match(/(opus|sonnet|haiku)-(\d{1,2})(?:[.-](\d{1,2}))?(?!\d)/);
     if (!match)
         return undefined;
     return {
@@ -125,6 +125,25 @@ function claudeFamilyVersion(modelId) {
         minor: match[3] === undefined ? undefined : Number.parseInt(match[3], 10),
     };
 }
+/** True when a parsed Claude version is at least `major.minor`. */
+function claudeAtLeast(claude, major, minor = 0) {
+    return claude.major > major || (claude.major === major && (claude.minor ?? 0) >= minor);
+}
+/**
+ * Anthropic-native transports, where dreb controls the Anthropic `thinking` /
+ * `output_config.effort` request fields directly: the Anthropic Messages API
+ * (Anthropic, GitHub Copilot's `/v1/messages`, opencode, custom Anthropic-compatible
+ * endpoints) and Amazon Bedrock. Claude served through OpenAI-compatible APIs
+ * (OpenRouter, Copilot Fable, ...) keeps the legacy capability rules because those
+ * transports map effort themselves.
+ */
+function isAnthropicNativeTransport(model) {
+    return model.api === "anthropic-messages" || model.api === "bedrock-converse-stream";
+}
+/** Claude model on an Anthropic-native transport, or undefined. */
+function nativeClaude(model) {
+    return isAnthropicNativeTransport(model) ? claudeFamilyVersion(model.id) : undefined;
+}
 /**
  * Parse Qwen family versions without mistaking parameter counts for minor versions.
  * Only a dot separates major from minor (qwen3.8-27b, Qwen-3.8); a hyphen after the
@@ -158,12 +177,21 @@ export function isQwen38OrLater(modelId) {
  * Supported today:
  * - GPT-5.6 model families (Sol, Terra, Luna, and the alias)
  * - GPT-6 model families (Astra and future variants)
+ * - On Anthropic-native transports (Anthropic Messages, including GitHub Copilot, and
+ *   Bedrock): Claude Opus/Sonnet 4.6+ (including the Claude 5 families) and Haiku 5.5+
  */
 export function supportsMax(model) {
     const id = model.id.toLowerCase();
     // Anchored to the id start or a gateway "/" prefix so unrelated ids
     // (e.g. "gpt-60") do not match.
-    return /(?:^|\/)gpt-5\.6(?:$|[-.])/.test(id) || /(?:^|\/)gpt-6(?:$|[-.])/.test(id);
+    if (/(?:^|\/)gpt-5\.6(?:$|[-.])/.test(id) || /(?:^|\/)gpt-6(?:$|[-.])/.test(id))
+        return true;
+    const claude = nativeClaude(model);
+    if (!claude)
+        return false;
+    if (claude.family === "haiku")
+        return claudeAtLeast(claude, 5, 5);
+    return claudeAtLeast(claude, 4, 6);
 }
 /**
  * Check if a model supports xhigh thinking level.
@@ -171,7 +199,10 @@ export function supportsMax(model) {
  * Supported today:
  * - GPT-5.2 through GPT-5.6 model families
  * - GPT-6 model families (Astra and future variants)
- * - Claude Opus 4.6–4.x and Claude 5 model families
+ * - Claude on Anthropic-native transports: Opus 4.7+, Sonnet 5+, Haiku 5.5+
+ *   (Opus 4.6 and Sonnet 4.6 have `max` but no native `xhigh`)
+ * - Claude on OpenAI-compatible transports (OpenRouter, ...): Opus 4.6+ and Sonnet 5+
+ *   (the legacy rule; those transports map effort themselves)
  * - Kimi Code K3 (xhigh maps to its advertised "max" effort)
  * - Qwen 3.8+ model families (xhigh is their top native effort tier)
  */
@@ -180,30 +211,130 @@ export function supportsXhigh(model) {
         return true;
     if (isQwen38OrLater(model.id))
         return true;
-    if (model.id.includes("gpt-5.2") ||
-        model.id.includes("gpt-5.3") ||
-        model.id.includes("gpt-5.4") ||
-        model.id.includes("gpt-5.5") ||
-        model.id.includes("gpt-5.6") ||
-        model.id.includes("gpt-6")) {
+    if (/gpt-(?:5\.[2-6]|6)/.test(model.id))
         return true;
-    }
     const claude = claudeFamilyVersion(model.id);
     if (!claude)
         return false;
-    if (claude.family === "opus") {
-        return claude.major >= 5 || (claude.major === 4 && (claude.minor ?? 0) >= 6);
+    if (isAnthropicNativeTransport(model)) {
+        if (claude.family === "opus")
+            return claudeAtLeast(claude, 4, 7);
+        if (claude.family === "sonnet")
+            return claudeAtLeast(claude, 5);
+        return claudeAtLeast(claude, 5, 5);
     }
-    return claude.major >= 5;
+    // Legacy rule for proxied Claude transports (unchanged; Haiku never qualified).
+    if (claude.family === "opus")
+        return claudeAtLeast(claude, 4, 6);
+    if (claude.family === "sonnet")
+        return claudeAtLeast(claude, 5);
+    return false;
 }
 /**
- * Check if a model uses adaptive thinking (Opus/Sonnet 4.6–4.x and Claude 5
- * families), where the `thinkingDisplay` option is honored. Mirrors the
- * per-provider internal checks.
+ * Check if a model uses adaptive thinking (Opus/Sonnet 4.6+, Haiku 5+, including the
+ * Claude 5 families), where `output_config.effort` and the `thinkingDisplay` option
+ * are honored. Mirrors the per-provider internal checks.
  */
 export function supportsAdaptiveThinking(model) {
     const claude = claudeFamilyVersion(model.id);
-    return claude != null && (claude.major >= 5 || (claude.major === 4 && (claude.minor ?? 0) >= 6));
+    if (!claude)
+        return false;
+    if (claude.family === "haiku")
+        return claudeAtLeast(claude, 5);
+    return claudeAtLeast(claude, 4, 6);
+}
+/**
+ * Check whether thinking can be turned off for a model (dreb level `off`).
+ *
+ * False only for Claude Opus 5.5+ on Anthropic-native transports, where adaptive
+ * thinking is always on and `thinking: {type: "disabled"}` is rejected. Callers that
+ * offer selectable levels should hide `off` for such models; {@link clampThinkingLevel}
+ * maps a requested `off` to `minimal`.
+ */
+export function supportsThinkingOff(model) {
+    return getClaudeThinkingOffMode(model) !== "omit";
+}
+/**
+ * Select the request shape that turns thinking off for a Claude model:
+ * - `"omit"`: send neither a `thinking` field nor an effort (Opus 5.5+, where thinking
+ *   cannot be disabled; the model default effort applies)
+ * - `"between_tools"`: `thinking: {type: "between_tools"}` (Sonnet 5.5+, where
+ *   `disabled` is rejected)
+ * - `"disabled"`: `thinking: {type: "disabled"}` (every other model, and every model on
+ *   non-native transports)
+ */
+export function getClaudeThinkingOffMode(model) {
+    const claude = nativeClaude(model);
+    if (!claude)
+        return "disabled";
+    if (claude.family === "opus" && claudeAtLeast(claude, 5, 5))
+        return "omit";
+    if (claude.family === "sonnet" && claudeAtLeast(claude, 5, 5))
+        return "between_tools";
+    return "disabled";
+}
+/**
+ * Check whether a Claude model thinks by default when a request omits the `thinking`
+ * field (the Claude 5 generation). Transports that otherwise omit the field to turn
+ * thinking off (Bedrock) must send an explicit off config for these models.
+ */
+export function claudeThinksByDefault(model) {
+    const claude = claudeFamilyVersion(model.id);
+    return claude !== undefined && claude.major >= 5;
+}
+/** Every dreb thinking level, in ascending order. */
+const THINKING_LEVEL_ORDER = [
+    "off",
+    "minimal",
+    "low",
+    "medium",
+    "high",
+    "xhigh",
+    "max",
+];
+/**
+ * Thinking levels a model supports, in ascending order: `["off"]` for non-reasoning
+ * models; otherwise the full scale minus `off` when thinking cannot be turned off
+ * (Claude Opus 5.5+), `xhigh` without {@link supportsXhigh}, and `max` without
+ * {@link supportsMax}.
+ */
+export function getAvailableThinkingLevels(model) {
+    if (!model?.reasoning)
+        return ["off"];
+    return THINKING_LEVEL_ORDER.filter((level) => {
+        if (level === "off")
+            return supportsThinkingOff(model);
+        if (level === "xhigh")
+            return supportsXhigh(model);
+        if (level === "max")
+            return supportsMax(model);
+        return true;
+    });
+}
+/**
+ * Clamp a requested thinking level to what a model supports. This is the single
+ * source of truth for thinking-level fallback across dreb: the nearest available
+ * level at or above the request, else the highest available level. So:
+ * - non-reasoning models → `off`
+ * - `off` on models where thinking cannot be turned off → `minimal`
+ * - unsupported `xhigh` → `max` when the model has `max` (Opus/Sonnet 4.6), else `high`
+ * - unsupported `max` → `xhigh` when supported, else `high`
+ * - an unknown level → the lowest available level
+ */
+export function clampThinkingLevel(model, level) {
+    const available = getAvailableThinkingLevels(model);
+    const requested = THINKING_LEVEL_ORDER.indexOf(level);
+    return (available.find((candidate) => THINKING_LEVEL_ORDER.indexOf(candidate) >= requested) ??
+        available[available.length - 1]);
+}
+/**
+ * Anthropic adaptive-thinking effort (`output_config.effort`) for a dreb thinking level,
+ * after clamping it to the model. `minimal` → `low`; `xhigh` and `max` are sent as
+ * themselves; an unset level (or `off` on a non-reasoning model) → `high`.
+ */
+export function getClaudeEffort(model, level) {
+    const clamped = level === undefined ? "high" : clampThinkingLevel(model, level);
+    return clamped === "minimal" ? "low" : clamped === "off" ? "high" : clamped;
 }
 /**
  * Check if two models are equal by comparing both their id and provider.

package/dist/models.js.map

@@ -1 +1 @@
-{"version":3,"file":"models.js","sourceRoot":"","sources":["../src/models.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,MAAM,EAAE,MAAM,uBAAuB,CAAC;AAG/C,MAAM,aAAa,GAAyC,IAAI,GAAG,EAAE,CAAC;AAEtE,iDAAiD;AACjD,KAAK,MAAM,CAAC,QAAQ,EAAE,MAAM,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,EAAE,CAAC;IACzD,MAAM,cAAc,GAAG,IAAI,GAAG,EAAsB,CAAC;IACrD,KAAK,MAAM,CAAC,EAAE,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,EAAE,CAAC;QAClD,cAAc,CAAC,GAAG,CAAC,EAAE,EAAE,KAAmB,CAAC,CAAC;IAC7C,CAAC;IACD,aAAa,CAAC,GAAG,CAAC,QAAQ,EAAE,cAAc,CAAC,CAAC;AAC7C,CAAC;AAOD,MAAM,UAAU,QAAQ,CACvB,QAAmB,EACnB,OAAiB,EACsB;IACvC,MAAM,cAAc,GAAG,aAAa,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC;IACnD,OAAO,cAAc,EAAE,GAAG,CAAC,OAAiB,CAAyC,CAAC;AAAA,CACtF;AAED;;;;;;GAMG;AACH,MAAM,UAAU,YAAY,CAAC,EAAU,EAAW;IACjD,IAAI,EAAE,CAAC,QAAQ,CAAC,SAAS,CAAC;QAAE,OAAO,IAAI,CAAC;IACxC,OAAO,CAAC,SAAS,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;AAAA,CAC3B;AAED;;;;;;;;;;;;;;;;GAgBG;AACH,MAAM,UAAU,SAAS,CAAC,QAAgB,EAAE,OAAe,EAA0B;IACpF,MAAM,cAAc,GAAG,aAAa,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC;IACnD,IAAI,CAAC,cAAc;QAAE,OAAO,SAAS,CAAC;IAEtC,wBAAwB;IACxB,MAAM,KAAK,GAAG,cAAc,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;IAC1C,IAAI,KAAK;QAAE,OAAO,KAAK,CAAC;IAExB,qCAAqC;IACrC,MAAM,iBAAiB,GAAG,OAAO,CAAC,WAAW,EAAE,CAAC;IAChD,MAAM,OAAO,GAAG,KAAK,CAAC,IAAI,CAAC,cAAc,CAAC,MAAM,EAAE,CAAC,CAAC,MAAM,CACzD,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,WAAW,EAAE,CAAC,QAAQ,CAAC,iBAAiB,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,WAAW,EAAE,CAAC,QAAQ,CAAC,iBAAiB,CAAC,CAC1G,CAAC;IAEF,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,SAAS,CAAC;IAC3C,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,OAAO,CAAC,CAAC,CAAC,CAAC;IAE5C,gEAA8D;IAC9D,MAAM,OAAO,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,YAAY,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;IAC1D,MAAM,aAAa,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;IAEjE,IAAI,OAAO,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACxB,mEAAiE;QACjE,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,aAAa,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;QACjD,OAAO,OAAO,CAAC,CAAC,CAAC,CAAC;IACnB,CAAC;IAED,kCAAgC;IAChC,aAAa,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,aAAa,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;IACvD,OAAO,aAAa,CAAC,CAAC,CAAC,CAAC;AAAA,CACxB;AAED;;;;;;GAMG;AACH,MAAM,UAAU,eAAe,CAAC,OAAe,EAAE,MAAoB,EAA0B;IAC9F,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,SAAS,CAAC;IAE1C,MAAM,iBAAiB,GAAG,OAAO,CAAC,WAAW,EAAE,CAAC;IAEhD,oCAAoC;IACpC,MAAM,SAAS,GAAG,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,WAAW,EAAE,KAAK,iBAAiB,CAAC,CAAC;IAC/E,IAAI,SAAS;QAAE,OAAO,SAAS,CAAC;IAEhC,qCAAqC;IACrC,MAAM,OAAO,GAAG,MAAM,CAAC,MAAM,CAC5B,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,WAAW,EAAE,CAAC,QAAQ,CAAC,iBAAiB,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,WAAW,EAAE,CAAC,QAAQ,CAAC,iBAAiB,CAAC,CAC1G,CAAC;IAEF,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,SAAS,CAAC;IAC3C,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,OAAO,CAAC,CAAC,CAAC,CAAC;IAE5C,gEAA8D;IAC9D,MAAM,OAAO,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,YAAY,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;IAC1D,MAAM,aAAa,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;IAEjE,IAAI,OAAO,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACxB,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,aAAa,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;QACjD,OAAO,OAAO,CAAC,CAAC,CAAC,CAAC;IACnB,CAAC;IAED,aAAa,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,aAAa,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;IACvD,OAAO,aAAa,CAAC,CAAC,CAAC,CAAC;AAAA,CACxB;AAED,MAAM,UAAU,YAAY,GAAoB;IAC/C,OAAO,KAAK,CAAC,IAAI,CAAC,aAAa,CAAC,IAAI,EAAE,CAAoB,CAAC;AAAA,CAC3D;AAED,MAAM,UAAU,SAAS,CACxB,QAAmB,EAC8C;IACjE,MAAM,MAAM,GAAG,aAAa,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC;IAC3C,OAAO,MAAM,CAAC,CAAC,CAAE,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,MAAM,EAAE,CAAoE,CAAC,CAAC,CAAC,EAAE,CAAC;AAAA,CACrH;AAED,MAAM,UAAU,aAAa,CAAmB,KAAkB,EAAE,KAAY,EAAiB;IAChG,KAAK,CAAC,IAAI,CAAC,KAAK,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,GAAG,OAAO,CAAC,GAAG,KAAK,CAAC,KAAK,CAAC;IAC9D,KAAK,CAAC,IAAI,CAAC,MAAM,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,MAAM,GAAG,OAAO,CAAC,GAAG,KAAK,CAAC,MAAM,CAAC;IACjE,KAAK,CAAC,IAAI,CAAC,SAAS,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,SAAS,GAAG,OAAO,CAAC,GAAG,KAAK,CAAC,SAAS,CAAC;IAC1E,KAAK,CAAC,IAAI,CAAC,UAAU,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,UAAU,GAAG,OAAO,CAAC,GAAG,KAAK,CAAC,UAAU,CAAC;IAC7E,KAAK,CAAC,IAAI,CAAC,KAAK,GAAG,KAAK,CAAC,IAAI,CAAC,KAAK,GAAG,KAAK,CAAC,IAAI,CAAC,MAAM,GAAG,KAAK,CAAC,IAAI,CAAC,SAAS,GAAG,KAAK,CAAC,IAAI,CAAC,UAAU,CAAC;IACvG,OAAO,KAAK,CAAC,IAAI,CAAC;AAAA,CAClB;AAQD,sFAAsF;AACtF,SAAS,mBAAmB,CAAC,OAAe,EAAmC;IAC9E,MAAM,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC,iDAAiD,CAAC,CAAC;IAC/E,IAAI,CAAC,KAAK;QAAE,OAAO,SAAS,CAAC;IAC7B,OAAO;QACN,MAAM,EAAE,KAAK,CAAC,CAAC,CAAkC;QACjD,KAAK,EAAE,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC;QACpC,KAAK,EAAE,KAAK,CAAC,CAAC,CAAC,KAAK,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC;KACzE,CAAC;AAAA,CACF;AAED;;;;;;;GAOG;AACH,SAAS,iBAAiB,CAAC,OAAe,EAAgD;IACzF,MAAM,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC,+CAA+C,CAAC,CAAC;IAC7E,IAAI,CAAC,KAAK;QAAE,OAAO,SAAS,CAAC;IAC7B,OAAO;QACN,KAAK,EAAE,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC;QACpC,KAAK,EAAE,KAAK,CAAC,CAAC,CAAC,KAAK,SAAS,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC;KACjE,CAAC;AAAA,CACF;AAED;;;GAGG;AACH,MAAM,UAAU,eAAe,CAAC,OAAe,EAAW;IACzD,MAAM,IAAI,GAAG,iBAAiB,CAAC,OAAO,CAAC,CAAC;IACxC,IAAI,CAAC,IAAI;QAAE,OAAO,KAAK,CAAC;IACxB,OAAO,IAAI,CAAC,KAAK,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,KAAK,CAAC,IAAI,IAAI,CAAC,KAAK,IAAI,CAAC,CAAC,CAAC;AAAA,CAC/D;AAED;;;;;;GAMG;AACH,MAAM,UAAU,WAAW,CAAmB,KAAkB,EAAW;IAC1E,MAAM,EAAE,GAAG,KAAK,CAAC,EAAE,CAAC,WAAW,EAAE,CAAC;IAClC,oEAAoE;IACpE,gCAAgC;IAChC,OAAO,4BAA4B,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,yBAAyB,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;AAAA,CACnF;AAED;;;;;;;;;GASG;AACH,MAAM,UAAU,aAAa,CAAmB,KAAkB,EAAW;IAC5E,IAAI,KAAK,CAAC,QAAQ,KAAK,mBAAmB,IAAI,KAAK,CAAC,EAAE,KAAK,IAAI;QAAE,OAAO,IAAI,CAAC;IAC7E,IAAI,eAAe,CAAC,KAAK,CAAC,EAAE,CAAC;QAAE,OAAO,IAAI,CAAC;IAC3C,IACC,KAAK,CAAC,EAAE,CAAC,QAAQ,CAAC,SAAS,CAAC;QAC5B,KAAK,CAAC,EAAE,CAAC,QAAQ,CAAC,SAAS,CAAC;QAC5B,KAAK,CAAC,EAAE,CAAC,QAAQ,CAAC,SAAS,CAAC;QAC5B,KAAK,CAAC,EAAE,CAAC,QAAQ,CAAC,SAAS,CAAC;QAC5B,KAAK,CAAC,EAAE,CAAC,QAAQ,CAAC,SAAS,CAAC;QAC5B,KAAK,CAAC,EAAE,CAAC,QAAQ,CAAC,OAAO,CAAC,EACzB,CAAC;QACF,OAAO,IAAI,CAAC;IACb,CAAC;IAED,MAAM,MAAM,GAAG,mBAAmB,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;IAC7C,IAAI,CAAC,MAAM;QAAE,OAAO,KAAK,CAAC;IAC1B,IAAI,MAAM,CAAC,MAAM,KAAK,MAAM,EAAE,CAAC;QAC9B,OAAO,MAAM,CAAC,KAAK,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,KAAK,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,KAAK,IAAI,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC;IAC9E,CAAC;IACD,OAAO,MAAM,CAAC,KAAK,IAAI,CAAC,CAAC;AAAA,CACzB;AAED;;;;GAIG;AACH,MAAM,UAAU,wBAAwB,CAAmB,KAAkB,EAAW;IACvF,MAAM,MAAM,GAAG,mBAAmB,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;IAC7C,OAAO,MAAM,IAAI,IAAI,IAAI,CAAC,MAAM,CAAC,KAAK,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,KAAK,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,KAAK,IAAI,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;AAAA,CACjG;AAED;;;GAGG;AACH,MAAM,UAAU,cAAc,CAC7B,CAAiC,EACjC,CAAiC,EACvB;IACV,IAAI,CAAC,CAAC,IAAI,CAAC,CAAC;QAAE,OAAO,KAAK,CAAC;IAC3B,OAAO,CAAC,CAAC,EAAE,KAAK,CAAC,CAAC,EAAE,IAAI,CAAC,CAAC,QAAQ,KAAK,CAAC,CAAC,QAAQ,CAAC;AAAA,CAClD","sourcesContent":["import { MODELS } from \"./models.generated.js\";\nimport type { Api, KnownProvider, Model, Usage } from \"./types.js\";\n\nconst modelRegistry: Map<string, Map<string, Model<Api>>> = new Map();\n\n// Initialize registry from MODELS on module load\nfor (const [provider, models] of Object.entries(MODELS)) {\n\tconst providerModels = new Map<string, Model<Api>>();\n\tfor (const [id, model] of Object.entries(models)) {\n\t\tproviderModels.set(id, model as Model<Api>);\n\t}\n\tmodelRegistry.set(provider, providerModels);\n}\n\ntype ModelApi<\n\tTProvider extends KnownProvider,\n\tTModelId extends keyof (typeof MODELS)[TProvider],\n> = (typeof MODELS)[TProvider][TModelId] extends { api: infer TApi } ? (TApi extends Api ? TApi : never) : never;\n\nexport function getModel<TProvider extends KnownProvider, TModelId extends keyof (typeof MODELS)[TProvider]>(\n\tprovider: TProvider,\n\tmodelId: TModelId,\n): Model<ModelApi<TProvider, TModelId>> {\n\tconst providerModels = modelRegistry.get(provider);\n\treturn providerModels?.get(modelId as string) as Model<ModelApi<TProvider, TModelId>>;\n}\n\n/**\n * Check if a model ID looks like an alias (no date suffix).\n * Aliases are preferred over dated versions when fuzzy matching.\n *\n * IDs ending with `-latest` are treated as aliases.\n * IDs ending with a date pattern (`-YYYYMMDD`) are treated as dated versions.\n */\nexport function isModelAlias(id: string): boolean {\n\tif (id.endsWith(\"-latest\")) return true;\n\treturn !/-\\d{8}$/.test(id);\n}\n\n/**\n * Find a model by fuzzy matching against the provider's registered models.\n *\n * Resolution order:\n * 1. Exact match by provider + model ID (via registry Map.get)\n * 2. Case-insensitive substring match against model ID and display name\n * 3. Among matches, prefer aliases (non-dated IDs) over dated versions\n * 4. Among ties, pick the lexicographically highest (latest) ID\n *\n * This is the same matching logic used by the CLI, subagent model resolution,\n * and interactive mode — centralised here so tests can exercise the real path.\n *\n * @example\n * findModel(\"anthropic\", \"sonnet\")  // → latest claude-sonnet alias\n * findModel(\"anthropic\", \"haiku\")   // → latest claude-haiku alias\n * findModel(\"openai\", \"gpt-5\")     // → latest gpt-5 alias\n */\nexport function findModel(provider: string, pattern: string): Model<Api> | undefined {\n\tconst providerModels = modelRegistry.get(provider);\n\tif (!providerModels) return undefined;\n\n\t// Try exact match first\n\tconst exact = providerModels.get(pattern);\n\tif (exact) return exact;\n\n\t// Substring match (case-insensitive)\n\tconst normalizedPattern = pattern.toLowerCase();\n\tconst matches = Array.from(providerModels.values()).filter(\n\t\t(m) => m.id.toLowerCase().includes(normalizedPattern) || m.name?.toLowerCase().includes(normalizedPattern),\n\t);\n\n\tif (matches.length === 0) return undefined;\n\tif (matches.length === 1) return matches[0];\n\n\t// Multiple matches — separate into aliases and dated versions\n\tconst aliases = matches.filter((m) => isModelAlias(m.id));\n\tconst datedVersions = matches.filter((m) => !isModelAlias(m.id));\n\n\tif (aliases.length > 0) {\n\t\t// Prefer alias — if multiple, pick the lexicographically highest\n\t\taliases.sort((a, b) => b.id.localeCompare(a.id));\n\t\treturn aliases[0];\n\t}\n\n\t// All dated — prefer the latest\n\tdatedVersions.sort((a, b) => b.id.localeCompare(a.id));\n\treturn datedVersions[0];\n}\n\n/**\n * Find a model by fuzzy matching against a flat array of models.\n * Same algorithm as findModel() but operates on an arbitrary model list\n * instead of the built-in registry.\n *\n * Used by model-resolver.ts and other code that manages its own model lists.\n */\nexport function findModelInList(pattern: string, models: Model<Api>[]): Model<Api> | undefined {\n\tif (models.length === 0) return undefined;\n\n\tconst normalizedPattern = pattern.toLowerCase();\n\n\t// Exact ID match (case-insensitive)\n\tconst exactById = models.find((m) => m.id.toLowerCase() === normalizedPattern);\n\tif (exactById) return exactById;\n\n\t// Substring match (case-insensitive)\n\tconst matches = models.filter(\n\t\t(m) => m.id.toLowerCase().includes(normalizedPattern) || m.name?.toLowerCase().includes(normalizedPattern),\n\t);\n\n\tif (matches.length === 0) return undefined;\n\tif (matches.length === 1) return matches[0];\n\n\t// Multiple matches — separate into aliases and dated versions\n\tconst aliases = matches.filter((m) => isModelAlias(m.id));\n\tconst datedVersions = matches.filter((m) => !isModelAlias(m.id));\n\n\tif (aliases.length > 0) {\n\t\taliases.sort((a, b) => b.id.localeCompare(a.id));\n\t\treturn aliases[0];\n\t}\n\n\tdatedVersions.sort((a, b) => b.id.localeCompare(a.id));\n\treturn datedVersions[0];\n}\n\nexport function getProviders(): KnownProvider[] {\n\treturn Array.from(modelRegistry.keys()) as KnownProvider[];\n}\n\nexport function getModels<TProvider extends KnownProvider>(\n\tprovider: TProvider,\n): Model<ModelApi<TProvider, keyof (typeof MODELS)[TProvider]>>[] {\n\tconst models = modelRegistry.get(provider);\n\treturn models ? (Array.from(models.values()) as Model<ModelApi<TProvider, keyof (typeof MODELS)[TProvider]>>[]) : [];\n}\n\nexport function calculateCost<TApi extends Api>(model: Model<TApi>, usage: Usage): Usage[\"cost\"] {\n\tusage.cost.input = (model.cost.input / 1000000) * usage.input;\n\tusage.cost.output = (model.cost.output / 1000000) * usage.output;\n\tusage.cost.cacheRead = (model.cost.cacheRead / 1000000) * usage.cacheRead;\n\tusage.cost.cacheWrite = (model.cost.cacheWrite / 1000000) * usage.cacheWrite;\n\tusage.cost.total = usage.cost.input + usage.cost.output + usage.cost.cacheRead + usage.cost.cacheWrite;\n\treturn usage.cost;\n}\n\ninterface ClaudeFamilyVersion {\n\tfamily: \"opus\" | \"sonnet\";\n\tmajor: number;\n\tminor?: number;\n}\n\n/** Parse Claude family versions without mistaking long date suffixes for versions. */\nfunction claudeFamilyVersion(modelId: string): ClaudeFamilyVersion | undefined {\n\tconst match = modelId.match(/(opus|sonnet)-(\\d{1,2})(?:[.-](\\d{1,2}))?(?!\\d)/);\n\tif (!match) return undefined;\n\treturn {\n\t\tfamily: match[1] as ClaudeFamilyVersion[\"family\"],\n\t\tmajor: Number.parseInt(match[2], 10),\n\t\tminor: match[3] === undefined ? undefined : Number.parseInt(match[3], 10),\n\t};\n}\n\n/**\n * Parse Qwen family versions without mistaking parameter counts for minor versions.\n * Only a dot separates major from minor (qwen3.8-27b, Qwen-3.8); a hyphen after the\n * major version introduces the parameter size (qwen3-32b → 3.0, not 3.32). A capture\n * directly followed by a `b`-suffixed size token is a parameter count, not a version\n * (qwen-32b, Qwen-7B-Chat, deepseek-r1-distill-qwen-32b), and Qwen majors never reach\n * two digits, so only a single digit is accepted.\n */\nfunction qwenFamilyVersion(modelId: string): { major: number; minor: number } | undefined {\n\tconst match = modelId.match(/\\bqwen[\\s_-]*v?(\\d)(?:\\.(\\d+))?(?![\\d.]*b\\b)/i);\n\tif (!match) return undefined;\n\treturn {\n\t\tmajor: Number.parseInt(match[1], 10),\n\t\tminor: match[2] === undefined ? 0 : Number.parseInt(match[2], 10),\n\t};\n}\n\n/**\n * Check whether a model id belongs to Qwen 3.8 or a later Qwen generation — the first\n * Qwen family with graded `reasoning_effort` tiers (low/medium/xhigh, default xhigh).\n */\nexport function isQwen38OrLater(modelId: string): boolean {\n\tconst qwen = qwenFamilyVersion(modelId);\n\tif (!qwen) return false;\n\treturn qwen.major > 3 || (qwen.major === 3 && qwen.minor >= 8);\n}\n\n/**\n * Check if a model supports the native `max` reasoning tier.\n *\n * Supported today:\n * - GPT-5.6 model families (Sol, Terra, Luna, and the alias)\n * - GPT-6 model families (Astra and future variants)\n */\nexport function supportsMax<TApi extends Api>(model: Model<TApi>): boolean {\n\tconst id = model.id.toLowerCase();\n\t// Anchored to the id start or a gateway \"/\" prefix so unrelated ids\n\t// (e.g. \"gpt-60\") do not match.\n\treturn /(?:^|\\/)gpt-5\\.6(?:$|[-.])/.test(id) || /(?:^|\\/)gpt-6(?:$|[-.])/.test(id);\n}\n\n/**\n * Check if a model supports xhigh thinking level.\n *\n * Supported today:\n * - GPT-5.2 through GPT-5.6 model families\n * - GPT-6 model families (Astra and future variants)\n * - Claude Opus 4.6–4.x and Claude 5 model families\n * - Kimi Code K3 (xhigh maps to its advertised \"max\" effort)\n * - Qwen 3.8+ model families (xhigh is their top native effort tier)\n */\nexport function supportsXhigh<TApi extends Api>(model: Model<TApi>): boolean {\n\tif (model.provider === \"kimi-coding-oauth\" && model.id === \"k3\") return true;\n\tif (isQwen38OrLater(model.id)) return true;\n\tif (\n\t\tmodel.id.includes(\"gpt-5.2\") ||\n\t\tmodel.id.includes(\"gpt-5.3\") ||\n\t\tmodel.id.includes(\"gpt-5.4\") ||\n\t\tmodel.id.includes(\"gpt-5.5\") ||\n\t\tmodel.id.includes(\"gpt-5.6\") ||\n\t\tmodel.id.includes(\"gpt-6\")\n\t) {\n\t\treturn true;\n\t}\n\n\tconst claude = claudeFamilyVersion(model.id);\n\tif (!claude) return false;\n\tif (claude.family === \"opus\") {\n\t\treturn claude.major >= 5 || (claude.major === 4 && (claude.minor ?? 0) >= 6);\n\t}\n\treturn claude.major >= 5;\n}\n\n/**\n * Check if a model uses adaptive thinking (Opus/Sonnet 4.6–4.x and Claude 5\n * families), where the `thinkingDisplay` option is honored. Mirrors the\n * per-provider internal checks.\n */\nexport function supportsAdaptiveThinking<TApi extends Api>(model: Model<TApi>): boolean {\n\tconst claude = claudeFamilyVersion(model.id);\n\treturn claude != null && (claude.major >= 5 || (claude.major === 4 && (claude.minor ?? 0) >= 6));\n}\n\n/**\n * Check if two models are equal by comparing both their id and provider.\n * Returns false if either model is null or undefined.\n */\nexport function modelsAreEqual<TApi extends Api>(\n\ta: Model<TApi> | null | undefined,\n\tb: Model<TApi> | null | undefined,\n): boolean {\n\tif (!a || !b) return false;\n\treturn a.id === b.id && a.provider === b.provider;\n}\n"]}
+{"version":3,"file":"models.js","sourceRoot":"","sources":["../src/models.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,MAAM,EAAE,MAAM,uBAAuB,CAAC;AAG/C,MAAM,aAAa,GAAyC,IAAI,GAAG,EAAE,CAAC;AAEtE,iDAAiD;AACjD,KAAK,MAAM,CAAC,QAAQ,EAAE,MAAM,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,EAAE,CAAC;IACzD,MAAM,cAAc,GAAG,IAAI,GAAG,EAAsB,CAAC;IACrD,KAAK,MAAM,CAAC,EAAE,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,EAAE,CAAC;QAClD,cAAc,CAAC,GAAG,CAAC,EAAE,EAAE,KAAmB,CAAC,CAAC;IAC7C,CAAC;IACD,aAAa,CAAC,GAAG,CAAC,QAAQ,EAAE,cAAc,CAAC,CAAC;AAC7C,CAAC;AAOD,MAAM,UAAU,QAAQ,CACvB,QAAmB,EACnB,OAAiB,EACsB;IACvC,MAAM,cAAc,GAAG,aAAa,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC;IACnD,OAAO,cAAc,EAAE,GAAG,CAAC,OAAiB,CAAyC,CAAC;AAAA,CACtF;AAED;;;;;;GAMG;AACH,MAAM,UAAU,YAAY,CAAC,EAAU,EAAW;IACjD,IAAI,EAAE,CAAC,QAAQ,CAAC,SAAS,CAAC;QAAE,OAAO,IAAI,CAAC;IACxC,OAAO,CAAC,SAAS,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;AAAA,CAC3B;AAED;;;;;;;;;;;;;;;;GAgBG;AACH,MAAM,UAAU,SAAS,CAAC,QAAgB,EAAE,OAAe,EAA0B;IACpF,MAAM,cAAc,GAAG,aAAa,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC;IACnD,IAAI,CAAC,cAAc;QAAE,OAAO,SAAS,CAAC;IAEtC,wBAAwB;IACxB,MAAM,KAAK,GAAG,cAAc,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;IAC1C,IAAI,KAAK;QAAE,OAAO,KAAK,CAAC;IAExB,qCAAqC;IACrC,MAAM,iBAAiB,GAAG,OAAO,CAAC,WAAW,EAAE,CAAC;IAChD,MAAM,OAAO,GAAG,KAAK,CAAC,IAAI,CAAC,cAAc,CAAC,MAAM,EAAE,CAAC,CAAC,MAAM,CACzD,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,WAAW,EAAE,CAAC,QAAQ,CAAC,iBAAiB,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,WAAW,EAAE,CAAC,QAAQ,CAAC,iBAAiB,CAAC,CAC1G,CAAC;IAEF,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,SAAS,CAAC;IAC3C,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,OAAO,CAAC,CAAC,CAAC,CAAC;IAE5C,gEAA8D;IAC9D,MAAM,OAAO,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,YAAY,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;IAC1D,MAAM,aAAa,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;IAEjE,IAAI,OAAO,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACxB,mEAAiE;QACjE,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,aAAa,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;QACjD,OAAO,OAAO,CAAC,CAAC,CAAC,CAAC;IACnB,CAAC;IAED,kCAAgC;IAChC,aAAa,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,aAAa,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;IACvD,OAAO,aAAa,CAAC,CAAC,CAAC,CAAC;AAAA,CACxB;AAED;;;;;;GAMG;AACH,MAAM,UAAU,eAAe,CAAC,OAAe,EAAE,MAAoB,EAA0B;IAC9F,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,SAAS,CAAC;IAE1C,MAAM,iBAAiB,GAAG,OAAO,CAAC,WAAW,EAAE,CAAC;IAEhD,oCAAoC;IACpC,MAAM,SAAS,GAAG,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,WAAW,EAAE,KAAK,iBAAiB,CAAC,CAAC;IAC/E,IAAI,SAAS;QAAE,OAAO,SAAS,CAAC;IAEhC,qCAAqC;IACrC,MAAM,OAAO,GAAG,MAAM,CAAC,MAAM,CAC5B,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,WAAW,EAAE,CAAC,QAAQ,CAAC,iBAAiB,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,WAAW,EAAE,CAAC,QAAQ,CAAC,iBAAiB,CAAC,CAC1G,CAAC;IAEF,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,SAAS,CAAC;IAC3C,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,OAAO,CAAC,CAAC,CAAC,CAAC;IAE5C,gEAA8D;IAC9D,MAAM,OAAO,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,YAAY,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;IAC1D,MAAM,aAAa,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;IAEjE,IAAI,OAAO,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACxB,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,aAAa,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;QACjD,OAAO,OAAO,CAAC,CAAC,CAAC,CAAC;IACnB,CAAC;IAED,aAAa,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,aAAa,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;IACvD,OAAO,aAAa,CAAC,CAAC,CAAC,CAAC;AAAA,CACxB;AAED,MAAM,UAAU,YAAY,GAAoB;IAC/C,OAAO,KAAK,CAAC,IAAI,CAAC,aAAa,CAAC,IAAI,EAAE,CAAoB,CAAC;AAAA,CAC3D;AAED,MAAM,UAAU,SAAS,CACxB,QAAmB,EAC8C;IACjE,MAAM,MAAM,GAAG,aAAa,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC;IAC3C,OAAO,MAAM,CAAC,CAAC,CAAE,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,MAAM,EAAE,CAAoE,CAAC,CAAC,CAAC,EAAE,CAAC;AAAA,CACrH;AAED,MAAM,UAAU,aAAa,CAAmB,KAAkB,EAAE,KAAY,EAAiB;IAChG,KAAK,CAAC,IAAI,CAAC,KAAK,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,GAAG,OAAO,CAAC,GAAG,KAAK,CAAC,KAAK,CAAC;IAC9D,KAAK,CAAC,IAAI,CAAC,MAAM,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,MAAM,GAAG,OAAO,CAAC,GAAG,KAAK,CAAC,MAAM,CAAC;IACjE,KAAK,CAAC,IAAI,CAAC,SAAS,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,SAAS,GAAG,OAAO,CAAC,GAAG,KAAK,CAAC,SAAS,CAAC;IAC1E,KAAK,CAAC,IAAI,CAAC,UAAU,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,UAAU,GAAG,OAAO,CAAC,GAAG,KAAK,CAAC,UAAU,CAAC;IAC7E,KAAK,CAAC,IAAI,CAAC,KAAK,GAAG,KAAK,CAAC,IAAI,CAAC,KAAK,GAAG,KAAK,CAAC,IAAI,CAAC,MAAM,GAAG,KAAK,CAAC,IAAI,CAAC,SAAS,GAAG,KAAK,CAAC,IAAI,CAAC,UAAU,CAAC;IACvG,OAAO,KAAK,CAAC,IAAI,CAAC;AAAA,CAClB;AAQD,sFAAsF;AACtF,SAAS,mBAAmB,CAAC,OAAe,EAAmC;IAC9E,MAAM,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC,uDAAuD,CAAC,CAAC;IACrF,IAAI,CAAC,KAAK;QAAE,OAAO,SAAS,CAAC;IAC7B,OAAO;QACN,MAAM,EAAE,KAAK,CAAC,CAAC,CAAkC;QACjD,KAAK,EAAE,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC;QACpC,KAAK,EAAE,KAAK,CAAC,CAAC,CAAC,KAAK,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC;KACzE,CAAC;AAAA,CACF;AAED,mEAAmE;AACnE,SAAS,aAAa,CAAC,MAA2B,EAAE,KAAa,EAAE,KAAK,GAAG,CAAC,EAAW;IACtF,OAAO,MAAM,CAAC,KAAK,GAAG,KAAK,IAAI,CAAC,MAAM,CAAC,KAAK,KAAK,KAAK,IAAI,CAAC,MAAM,CAAC,KAAK,IAAI,CAAC,CAAC,IAAI,KAAK,CAAC,CAAC;AAAA,CACxF;AAED;;;;;;;GAOG;AACH,SAAS,0BAA0B,CAAmB,KAAkB,EAAW;IAClF,OAAO,KAAK,CAAC,GAAG,KAAK,oBAAoB,IAAI,KAAK,CAAC,GAAG,KAAK,yBAAyB,CAAC;AAAA,CACrF;AAED,mEAAmE;AACnE,SAAS,YAAY,CAAmB,KAAkB,EAAmC;IAC5F,OAAO,0BAA0B,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,mBAAmB,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;AAAA,CACrF;AAED;;;;;;;GAOG;AACH,SAAS,iBAAiB,CAAC,OAAe,EAAgD;IACzF,MAAM,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC,+CAA+C,CAAC,CAAC;IAC7E,IAAI,CAAC,KAAK;QAAE,OAAO,SAAS,CAAC;IAC7B,OAAO;QACN,KAAK,EAAE,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC;QACpC,KAAK,EAAE,KAAK,CAAC,CAAC,CAAC,KAAK,SAAS,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC;KACjE,CAAC;AAAA,CACF;AAED;;;GAGG;AACH,MAAM,UAAU,eAAe,CAAC,OAAe,EAAW;IACzD,MAAM,IAAI,GAAG,iBAAiB,CAAC,OAAO,CAAC,CAAC;IACxC,IAAI,CAAC,IAAI;QAAE,OAAO,KAAK,CAAC;IACxB,OAAO,IAAI,CAAC,KAAK,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,KAAK,CAAC,IAAI,IAAI,CAAC,KAAK,IAAI,CAAC,CAAC,CAAC;AAAA,CAC/D;AAED;;;;;;;;GAQG;AACH,MAAM,UAAU,WAAW,CAAmB,KAAkB,EAAW;IAC1E,MAAM,EAAE,GAAG,KAAK,CAAC,EAAE,CAAC,WAAW,EAAE,CAAC;IAClC,oEAAoE;IACpE,gCAAgC;IAChC,IAAI,4BAA4B,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,yBAAyB,CAAC,IAAI,CAAC,EAAE,CAAC;QAAE,OAAO,IAAI,CAAC;IAE7F,MAAM,MAAM,GAAG,YAAY,CAAC,KAAK,CAAC,CAAC;IACnC,IAAI,CAAC,MAAM;QAAE,OAAO,KAAK,CAAC;IAC1B,IAAI,MAAM,CAAC,MAAM,KAAK,OAAO;QAAE,OAAO,aAAa,CAAC,MAAM,EAAE,CAAC,EAAE,CAAC,CAAC,CAAC;IAClE,OAAO,aAAa,CAAC,MAAM,EAAE,CAAC,EAAE,CAAC,CAAC,CAAC;AAAA,CACnC;AAED;;;;;;;;;;;;GAYG;AACH,MAAM,UAAU,aAAa,CAAmB,KAAkB,EAAW;IAC5E,IAAI,KAAK,CAAC,QAAQ,KAAK,mBAAmB,IAAI,KAAK,CAAC,EAAE,KAAK,IAAI;QAAE,OAAO,IAAI,CAAC;IAC7E,IAAI,eAAe,CAAC,KAAK,CAAC,EAAE,CAAC;QAAE,OAAO,IAAI,CAAC;IAC3C,IAAI,oBAAoB,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC;QAAE,OAAO,IAAI,CAAC;IAErD,MAAM,MAAM,GAAG,mBAAmB,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;IAC7C,IAAI,CAAC,MAAM;QAAE,OAAO,KAAK,CAAC;IAC1B,IAAI,0BAA0B,CAAC,KAAK,CAAC,EAAE,CAAC;QACvC,IAAI,MAAM,CAAC,MAAM,KAAK,MAAM;YAAE,OAAO,aAAa,CAAC,MAAM,EAAE,CAAC,EAAE,CAAC,CAAC,CAAC;QACjE,IAAI,MAAM,CAAC,MAAM,KAAK,QAAQ;YAAE,OAAO,aAAa,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;QAChE,OAAO,aAAa,CAAC,MAAM,EAAE,CAAC,EAAE,CAAC,CAAC,CAAC;IACpC,CAAC;IACD,gFAAgF;IAChF,IAAI,MAAM,CAAC,MAAM,KAAK,MAAM;QAAE,OAAO,aAAa,CAAC,MAAM,EAAE,CAAC,EAAE,CAAC,CAAC,CAAC;IACjE,IAAI,MAAM,CAAC,MAAM,KAAK,QAAQ;QAAE,OAAO,aAAa,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;IAChE,OAAO,KAAK,CAAC;AAAA,CACb;AAED;;;;GAIG;AACH,MAAM,UAAU,wBAAwB,CAAmB,KAAkB,EAAW;IACvF,MAAM,MAAM,GAAG,mBAAmB,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;IAC7C,IAAI,CAAC,MAAM;QAAE,OAAO,KAAK,CAAC;IAC1B,IAAI,MAAM,CAAC,MAAM,KAAK,OAAO;QAAE,OAAO,aAAa,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;IAC/D,OAAO,aAAa,CAAC,MAAM,EAAE,CAAC,EAAE,CAAC,CAAC,CAAC;AAAA,CACnC;AAED;;;;;;;GAOG;AACH,MAAM,UAAU,mBAAmB,CAAmB,KAAkB,EAAW;IAClF,OAAO,wBAAwB,CAAC,KAAK,CAAC,KAAK,MAAM,CAAC;AAAA,CAClD;AAKD;;;;;;;;GAQG;AACH,MAAM,UAAU,wBAAwB,CAAmB,KAAkB,EAAyB;IACrG,MAAM,MAAM,GAAG,YAAY,CAAC,KAAK,CAAC,CAAC;IACnC,IAAI,CAAC,MAAM;QAAE,OAAO,UAAU,CAAC;IAC/B,IAAI,MAAM,CAAC,MAAM,KAAK,MAAM,IAAI,aAAa,CAAC,MAAM,EAAE,CAAC,EAAE,CAAC,CAAC;QAAE,OAAO,MAAM,CAAC;IAC3E,IAAI,MAAM,CAAC,MAAM,KAAK,QAAQ,IAAI,aAAa,CAAC,MAAM,EAAE,CAAC,EAAE,CAAC,CAAC;QAAE,OAAO,eAAe,CAAC;IACtF,OAAO,UAAU,CAAC;AAAA,CAClB;AAED;;;;GAIG;AACH,MAAM,UAAU,qBAAqB,CAAmB,KAAkB,EAAW;IACpF,MAAM,MAAM,GAAG,mBAAmB,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;IAC7C,OAAO,MAAM,KAAK,SAAS,IAAI,MAAM,CAAC,KAAK,IAAI,CAAC,CAAC;AAAA,CACjD;AAED,qDAAqD;AACrD,MAAM,oBAAoB,GAAuC;IAChE,KAAK;IACL,SAAS;IACT,KAAK;IACL,QAAQ;IACR,MAAM;IACN,OAAO;IACP,KAAK;CACL,CAAC;AAEF;;;;;GAKG;AACH,MAAM,UAAU,0BAA0B,CACzC,KAA8B,EACF;IAC5B,IAAI,CAAC,KAAK,EAAE,SAAS;QAAE,OAAO,CAAC,KAAK,CAAC,CAAC;IACtC,OAAO,oBAAoB,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC;QAC7C,IAAI,KAAK,KAAK,KAAK;YAAE,OAAO,mBAAmB,CAAC,KAAK,CAAC,CAAC;QACvD,IAAI,KAAK,KAAK,OAAO;YAAE,OAAO,aAAa,CAAC,KAAK,CAAC,CAAC;QACnD,IAAI,KAAK,KAAK,KAAK;YAAE,OAAO,WAAW,CAAC,KAAK,CAAC,CAAC;QAC/C,OAAO,IAAI,CAAC;IAAA,CACZ,CAAC,CAAC;AAAA,CACH;AAED;;;;;;;;;GASG;AACH,MAAM,UAAU,kBAAkB,CACjC,KAA8B,EAC9B,KAA4B,EACJ;IACxB,MAAM,SAAS,GAAG,0BAA0B,CAAC,KAAK,CAAC,CAAC;IACpD,MAAM,SAAS,GAAG,oBAAoB,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC;IACtD,OAAO,CACN,SAAS,CAAC,IAAI,CAAC,CAAC,SAAS,EAAE,EAAE,CAAC,oBAAoB,CAAC,OAAO,CAAC,SAAS,CAAC,IAAI,SAAS,CAAC;QACnF,SAAS,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,CAAC,CAC/B,CAAC;AAAA,CACF;AAED;;;;GAIG;AACH,MAAM,UAAU,eAAe,CAC9B,KAAkB,EAClB,KAAgC,EACI;IACpC,MAAM,OAAO,GAAG,KAAK,KAAK,SAAS,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,kBAAkB,CAAC,KAAK,EAAE,KAAK,CAAC,CAAC;IAChF,OAAO,OAAO,KAAK,SAAS,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,OAAO,KAAK,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,OAAO,CAAC;AAAA,CAC5E;AAED;;;GAGG;AACH,MAAM,UAAU,cAAc,CAC7B,CAAiC,EACjC,CAAiC,EACvB;IACV,IAAI,CAAC,CAAC,IAAI,CAAC,CAAC;QAAE,OAAO,KAAK,CAAC;IAC3B,OAAO,CAAC,CAAC,EAAE,KAAK,CAAC,CAAC,EAAE,IAAI,CAAC,CAAC,QAAQ,KAAK,CAAC,CAAC,QAAQ,CAAC;AAAA,CAClD","sourcesContent":["import { MODELS } from \"./models.generated.js\";\nimport type { Api, KnownProvider, Model, ThinkingLevel, Usage } from \"./types.js\";\n\nconst modelRegistry: Map<string, Map<string, Model<Api>>> = new Map();\n\n// Initialize registry from MODELS on module load\nfor (const [provider, models] of Object.entries(MODELS)) {\n\tconst providerModels = new Map<string, Model<Api>>();\n\tfor (const [id, model] of Object.entries(models)) {\n\t\tproviderModels.set(id, model as Model<Api>);\n\t}\n\tmodelRegistry.set(provider, providerModels);\n}\n\ntype ModelApi<\n\tTProvider extends KnownProvider,\n\tTModelId extends keyof (typeof MODELS)[TProvider],\n> = (typeof MODELS)[TProvider][TModelId] extends { api: infer TApi } ? (TApi extends Api ? TApi : never) : never;\n\nexport function getModel<TProvider extends KnownProvider, TModelId extends keyof (typeof MODELS)[TProvider]>(\n\tprovider: TProvider,\n\tmodelId: TModelId,\n): Model<ModelApi<TProvider, TModelId>> {\n\tconst providerModels = modelRegistry.get(provider);\n\treturn providerModels?.get(modelId as string) as Model<ModelApi<TProvider, TModelId>>;\n}\n\n/**\n * Check if a model ID looks like an alias (no date suffix).\n * Aliases are preferred over dated versions when fuzzy matching.\n *\n * IDs ending with `-latest` are treated as aliases.\n * IDs ending with a date pattern (`-YYYYMMDD`) are treated as dated versions.\n */\nexport function isModelAlias(id: string): boolean {\n\tif (id.endsWith(\"-latest\")) return true;\n\treturn !/-\\d{8}$/.test(id);\n}\n\n/**\n * Find a model by fuzzy matching against the provider's registered models.\n *\n * Resolution order:\n * 1. Exact match by provider + model ID (via registry Map.get)\n * 2. Case-insensitive substring match against model ID and display name\n * 3. Among matches, prefer aliases (non-dated IDs) over dated versions\n * 4. Among ties, pick the lexicographically highest (latest) ID\n *\n * This is the same matching logic used by the CLI, subagent model resolution,\n * and interactive mode — centralised here so tests can exercise the real path.\n *\n * @example\n * findModel(\"anthropic\", \"sonnet\")  // → latest claude-sonnet alias\n * findModel(\"anthropic\", \"haiku\")   // → latest claude-haiku alias\n * findModel(\"openai\", \"gpt-5\")     // → latest gpt-5 alias\n */\nexport function findModel(provider: string, pattern: string): Model<Api> | undefined {\n\tconst providerModels = modelRegistry.get(provider);\n\tif (!providerModels) return undefined;\n\n\t// Try exact match first\n\tconst exact = providerModels.get(pattern);\n\tif (exact) return exact;\n\n\t// Substring match (case-insensitive)\n\tconst normalizedPattern = pattern.toLowerCase();\n\tconst matches = Array.from(providerModels.values()).filter(\n\t\t(m) => m.id.toLowerCase().includes(normalizedPattern) || m.name?.toLowerCase().includes(normalizedPattern),\n\t);\n\n\tif (matches.length === 0) return undefined;\n\tif (matches.length === 1) return matches[0];\n\n\t// Multiple matches — separate into aliases and dated versions\n\tconst aliases = matches.filter((m) => isModelAlias(m.id));\n\tconst datedVersions = matches.filter((m) => !isModelAlias(m.id));\n\n\tif (aliases.length > 0) {\n\t\t// Prefer alias — if multiple, pick the lexicographically highest\n\t\taliases.sort((a, b) => b.id.localeCompare(a.id));\n\t\treturn aliases[0];\n\t}\n\n\t// All dated — prefer the latest\n\tdatedVersions.sort((a, b) => b.id.localeCompare(a.id));\n\treturn datedVersions[0];\n}\n\n/**\n * Find a model by fuzzy matching against a flat array of models.\n * Same algorithm as findModel() but operates on an arbitrary model list\n * instead of the built-in registry.\n *\n * Used by model-resolver.ts and other code that manages its own model lists.\n */\nexport function findModelInList(pattern: string, models: Model<Api>[]): Model<Api> | undefined {\n\tif (models.length === 0) return undefined;\n\n\tconst normalizedPattern = pattern.toLowerCase();\n\n\t// Exact ID match (case-insensitive)\n\tconst exactById = models.find((m) => m.id.toLowerCase() === normalizedPattern);\n\tif (exactById) return exactById;\n\n\t// Substring match (case-insensitive)\n\tconst matches = models.filter(\n\t\t(m) => m.id.toLowerCase().includes(normalizedPattern) || m.name?.toLowerCase().includes(normalizedPattern),\n\t);\n\n\tif (matches.length === 0) return undefined;\n\tif (matches.length === 1) return matches[0];\n\n\t// Multiple matches — separate into aliases and dated versions\n\tconst aliases = matches.filter((m) => isModelAlias(m.id));\n\tconst datedVersions = matches.filter((m) => !isModelAlias(m.id));\n\n\tif (aliases.length > 0) {\n\t\taliases.sort((a, b) => b.id.localeCompare(a.id));\n\t\treturn aliases[0];\n\t}\n\n\tdatedVersions.sort((a, b) => b.id.localeCompare(a.id));\n\treturn datedVersions[0];\n}\n\nexport function getProviders(): KnownProvider[] {\n\treturn Array.from(modelRegistry.keys()) as KnownProvider[];\n}\n\nexport function getModels<TProvider extends KnownProvider>(\n\tprovider: TProvider,\n): Model<ModelApi<TProvider, keyof (typeof MODELS)[TProvider]>>[] {\n\tconst models = modelRegistry.get(provider);\n\treturn models ? (Array.from(models.values()) as Model<ModelApi<TProvider, keyof (typeof MODELS)[TProvider]>>[]) : [];\n}\n\nexport function calculateCost<TApi extends Api>(model: Model<TApi>, usage: Usage): Usage[\"cost\"] {\n\tusage.cost.input = (model.cost.input / 1000000) * usage.input;\n\tusage.cost.output = (model.cost.output / 1000000) * usage.output;\n\tusage.cost.cacheRead = (model.cost.cacheRead / 1000000) * usage.cacheRead;\n\tusage.cost.cacheWrite = (model.cost.cacheWrite / 1000000) * usage.cacheWrite;\n\tusage.cost.total = usage.cost.input + usage.cost.output + usage.cost.cacheRead + usage.cost.cacheWrite;\n\treturn usage.cost;\n}\n\ninterface ClaudeFamilyVersion {\n\tfamily: \"opus\" | \"sonnet\" | \"haiku\";\n\tmajor: number;\n\tminor?: number;\n}\n\n/** Parse Claude family versions without mistaking long date suffixes for versions. */\nfunction claudeFamilyVersion(modelId: string): ClaudeFamilyVersion | undefined {\n\tconst match = modelId.match(/(opus|sonnet|haiku)-(\\d{1,2})(?:[.-](\\d{1,2}))?(?!\\d)/);\n\tif (!match) return undefined;\n\treturn {\n\t\tfamily: match[1] as ClaudeFamilyVersion[\"family\"],\n\t\tmajor: Number.parseInt(match[2], 10),\n\t\tminor: match[3] === undefined ? undefined : Number.parseInt(match[3], 10),\n\t};\n}\n\n/** True when a parsed Claude version is at least `major.minor`. */\nfunction claudeAtLeast(claude: ClaudeFamilyVersion, major: number, minor = 0): boolean {\n\treturn claude.major > major || (claude.major === major && (claude.minor ?? 0) >= minor);\n}\n\n/**\n * Anthropic-native transports, where dreb controls the Anthropic `thinking` /\n * `output_config.effort` request fields directly: the Anthropic Messages API\n * (Anthropic, GitHub Copilot's `/v1/messages`, opencode, custom Anthropic-compatible\n * endpoints) and Amazon Bedrock. Claude served through OpenAI-compatible APIs\n * (OpenRouter, Copilot Fable, ...) keeps the legacy capability rules because those\n * transports map effort themselves.\n */\nfunction isAnthropicNativeTransport<TApi extends Api>(model: Model<TApi>): boolean {\n\treturn model.api === \"anthropic-messages\" || model.api === \"bedrock-converse-stream\";\n}\n\n/** Claude model on an Anthropic-native transport, or undefined. */\nfunction nativeClaude<TApi extends Api>(model: Model<TApi>): ClaudeFamilyVersion | undefined {\n\treturn isAnthropicNativeTransport(model) ? claudeFamilyVersion(model.id) : undefined;\n}\n\n/**\n * Parse Qwen family versions without mistaking parameter counts for minor versions.\n * Only a dot separates major from minor (qwen3.8-27b, Qwen-3.8); a hyphen after the\n * major version introduces the parameter size (qwen3-32b → 3.0, not 3.32). A capture\n * directly followed by a `b`-suffixed size token is a parameter count, not a version\n * (qwen-32b, Qwen-7B-Chat, deepseek-r1-distill-qwen-32b), and Qwen majors never reach\n * two digits, so only a single digit is accepted.\n */\nfunction qwenFamilyVersion(modelId: string): { major: number; minor: number } | undefined {\n\tconst match = modelId.match(/\\bqwen[\\s_-]*v?(\\d)(?:\\.(\\d+))?(?![\\d.]*b\\b)/i);\n\tif (!match) return undefined;\n\treturn {\n\t\tmajor: Number.parseInt(match[1], 10),\n\t\tminor: match[2] === undefined ? 0 : Number.parseInt(match[2], 10),\n\t};\n}\n\n/**\n * Check whether a model id belongs to Qwen 3.8 or a later Qwen generation — the first\n * Qwen family with graded `reasoning_effort` tiers (low/medium/xhigh, default xhigh).\n */\nexport function isQwen38OrLater(modelId: string): boolean {\n\tconst qwen = qwenFamilyVersion(modelId);\n\tif (!qwen) return false;\n\treturn qwen.major > 3 || (qwen.major === 3 && qwen.minor >= 8);\n}\n\n/**\n * Check if a model supports the native `max` reasoning tier.\n *\n * Supported today:\n * - GPT-5.6 model families (Sol, Terra, Luna, and the alias)\n * - GPT-6 model families (Astra and future variants)\n * - On Anthropic-native transports (Anthropic Messages, including GitHub Copilot, and\n *   Bedrock): Claude Opus/Sonnet 4.6+ (including the Claude 5 families) and Haiku 5.5+\n */\nexport function supportsMax<TApi extends Api>(model: Model<TApi>): boolean {\n\tconst id = model.id.toLowerCase();\n\t// Anchored to the id start or a gateway \"/\" prefix so unrelated ids\n\t// (e.g. \"gpt-60\") do not match.\n\tif (/(?:^|\\/)gpt-5\\.6(?:$|[-.])/.test(id) || /(?:^|\\/)gpt-6(?:$|[-.])/.test(id)) return true;\n\n\tconst claude = nativeClaude(model);\n\tif (!claude) return false;\n\tif (claude.family === \"haiku\") return claudeAtLeast(claude, 5, 5);\n\treturn claudeAtLeast(claude, 4, 6);\n}\n\n/**\n * Check if a model supports xhigh thinking level.\n *\n * Supported today:\n * - GPT-5.2 through GPT-5.6 model families\n * - GPT-6 model families (Astra and future variants)\n * - Claude on Anthropic-native transports: Opus 4.7+, Sonnet 5+, Haiku 5.5+\n *   (Opus 4.6 and Sonnet 4.6 have `max` but no native `xhigh`)\n * - Claude on OpenAI-compatible transports (OpenRouter, ...): Opus 4.6+ and Sonnet 5+\n *   (the legacy rule; those transports map effort themselves)\n * - Kimi Code K3 (xhigh maps to its advertised \"max\" effort)\n * - Qwen 3.8+ model families (xhigh is their top native effort tier)\n */\nexport function supportsXhigh<TApi extends Api>(model: Model<TApi>): boolean {\n\tif (model.provider === \"kimi-coding-oauth\" && model.id === \"k3\") return true;\n\tif (isQwen38OrLater(model.id)) return true;\n\tif (/gpt-(?:5\\.[2-6]|6)/.test(model.id)) return true;\n\n\tconst claude = claudeFamilyVersion(model.id);\n\tif (!claude) return false;\n\tif (isAnthropicNativeTransport(model)) {\n\t\tif (claude.family === \"opus\") return claudeAtLeast(claude, 4, 7);\n\t\tif (claude.family === \"sonnet\") return claudeAtLeast(claude, 5);\n\t\treturn claudeAtLeast(claude, 5, 5);\n\t}\n\t// Legacy rule for proxied Claude transports (unchanged; Haiku never qualified).\n\tif (claude.family === \"opus\") return claudeAtLeast(claude, 4, 6);\n\tif (claude.family === \"sonnet\") return claudeAtLeast(claude, 5);\n\treturn false;\n}\n\n/**\n * Check if a model uses adaptive thinking (Opus/Sonnet 4.6+, Haiku 5+, including the\n * Claude 5 families), where `output_config.effort` and the `thinkingDisplay` option\n * are honored. Mirrors the per-provider internal checks.\n */\nexport function supportsAdaptiveThinking<TApi extends Api>(model: Model<TApi>): boolean {\n\tconst claude = claudeFamilyVersion(model.id);\n\tif (!claude) return false;\n\tif (claude.family === \"haiku\") return claudeAtLeast(claude, 5);\n\treturn claudeAtLeast(claude, 4, 6);\n}\n\n/**\n * Check whether thinking can be turned off for a model (dreb level `off`).\n *\n * False only for Claude Opus 5.5+ on Anthropic-native transports, where adaptive\n * thinking is always on and `thinking: {type: \"disabled\"}` is rejected. Callers that\n * offer selectable levels should hide `off` for such models; {@link clampThinkingLevel}\n * maps a requested `off` to `minimal`.\n */\nexport function supportsThinkingOff<TApi extends Api>(model: Model<TApi>): boolean {\n\treturn getClaudeThinkingOffMode(model) !== \"omit\";\n}\n\n/** How a request turns Claude thinking off on an Anthropic-native transport. */\nexport type ClaudeThinkingOffMode = \"disabled\" | \"between_tools\" | \"omit\";\n\n/**\n * Select the request shape that turns thinking off for a Claude model:\n * - `\"omit\"`: send neither a `thinking` field nor an effort (Opus 5.5+, where thinking\n *   cannot be disabled; the model default effort applies)\n * - `\"between_tools\"`: `thinking: {type: \"between_tools\"}` (Sonnet 5.5+, where\n *   `disabled` is rejected)\n * - `\"disabled\"`: `thinking: {type: \"disabled\"}` (every other model, and every model on\n *   non-native transports)\n */\nexport function getClaudeThinkingOffMode<TApi extends Api>(model: Model<TApi>): ClaudeThinkingOffMode {\n\tconst claude = nativeClaude(model);\n\tif (!claude) return \"disabled\";\n\tif (claude.family === \"opus\" && claudeAtLeast(claude, 5, 5)) return \"omit\";\n\tif (claude.family === \"sonnet\" && claudeAtLeast(claude, 5, 5)) return \"between_tools\";\n\treturn \"disabled\";\n}\n\n/**\n * Check whether a Claude model thinks by default when a request omits the `thinking`\n * field (the Claude 5 generation). Transports that otherwise omit the field to turn\n * thinking off (Bedrock) must send an explicit off config for these models.\n */\nexport function claudeThinksByDefault<TApi extends Api>(model: Model<TApi>): boolean {\n\tconst claude = claudeFamilyVersion(model.id);\n\treturn claude !== undefined && claude.major >= 5;\n}\n\n/** Every dreb thinking level, in ascending order. */\nconst THINKING_LEVEL_ORDER: readonly (ThinkingLevel | \"off\")[] = [\n\t\"off\",\n\t\"minimal\",\n\t\"low\",\n\t\"medium\",\n\t\"high\",\n\t\"xhigh\",\n\t\"max\",\n];\n\n/**\n * Thinking levels a model supports, in ascending order: `[\"off\"]` for non-reasoning\n * models; otherwise the full scale minus `off` when thinking cannot be turned off\n * (Claude Opus 5.5+), `xhigh` without {@link supportsXhigh}, and `max` without\n * {@link supportsMax}.\n */\nexport function getAvailableThinkingLevels<TApi extends Api>(\n\tmodel: Model<TApi> | undefined,\n): (ThinkingLevel | \"off\")[] {\n\tif (!model?.reasoning) return [\"off\"];\n\treturn THINKING_LEVEL_ORDER.filter((level) => {\n\t\tif (level === \"off\") return supportsThinkingOff(model);\n\t\tif (level === \"xhigh\") return supportsXhigh(model);\n\t\tif (level === \"max\") return supportsMax(model);\n\t\treturn true;\n\t});\n}\n\n/**\n * Clamp a requested thinking level to what a model supports. This is the single\n * source of truth for thinking-level fallback across dreb: the nearest available\n * level at or above the request, else the highest available level. So:\n * - non-reasoning models → `off`\n * - `off` on models where thinking cannot be turned off → `minimal`\n * - unsupported `xhigh` → `max` when the model has `max` (Opus/Sonnet 4.6), else `high`\n * - unsupported `max` → `xhigh` when supported, else `high`\n * - an unknown level → the lowest available level\n */\nexport function clampThinkingLevel<TApi extends Api>(\n\tmodel: Model<TApi> | undefined,\n\tlevel: ThinkingLevel | \"off\",\n): ThinkingLevel | \"off\" {\n\tconst available = getAvailableThinkingLevels(model);\n\tconst requested = THINKING_LEVEL_ORDER.indexOf(level);\n\treturn (\n\t\tavailable.find((candidate) => THINKING_LEVEL_ORDER.indexOf(candidate) >= requested) ??\n\t\tavailable[available.length - 1]\n\t);\n}\n\n/**\n * Anthropic adaptive-thinking effort (`output_config.effort`) for a dreb thinking level,\n * after clamping it to the model. `minimal` → `low`; `xhigh` and `max` are sent as\n * themselves; an unset level (or `off` on a non-reasoning model) → `high`.\n */\nexport function getClaudeEffort<TApi extends Api>(\n\tmodel: Model<TApi>,\n\tlevel: ThinkingLevel | undefined,\n): Exclude<ThinkingLevel, \"minimal\"> {\n\tconst clamped = level === undefined ? \"high\" : clampThinkingLevel(model, level);\n\treturn clamped === \"minimal\" ? \"low\" : clamped === \"off\" ? \"high\" : clamped;\n}\n\n/**\n * Check if two models are equal by comparing both their id and provider.\n * Returns false if either model is null or undefined.\n */\nexport function modelsAreEqual<TApi extends Api>(\n\ta: Model<TApi> | null | undefined,\n\tb: Model<TApi> | null | undefined,\n): boolean {\n\tif (!a || !b) return false;\n\treturn a.id === b.id && a.provider === b.provider;\n}\n"]}

package/dist/providers/amazon-bedrock.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"amazon-bedrock.d.ts","sourceRoot":"","sources":["../../src/providers/amazon-bedrock.ts"],"names":[],"mappings":"AAuBA,OAAO,KAAK,EAKX,KAAK,EACL,mBAAmB,EAEnB,cAAc,EACd,aAAa,EAEb,eAAe,EAEf,aAAa,EAIb,MAAM,aAAa,CAAC;AAOrB,MAAM,WAAW,cAAe,SAAQ,aAAa;IACpD,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,UAAU,CAAC,EAAE,MAAM,GAAG,KAAK,GAAG,MAAM,GAAG;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAA;KAAE,CAAC;IAEtE,SAAS,CAAC,EAAE,aAAa,CAAC;IAE1B,eAAe,CAAC,EAAE,eAAe,CAAC;IAKlC,eAAe,CAAC,EAAE,YAAY,GAAG,SAAS,CAAC;IAE3C,mBAAmB,CAAC,EAAE,OAAO,CAAC;IAC9B;;;sGAGkG;IAClG,eAAe,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;CACzC;AAID,eAAO,MAAM,aAAa,EAAE,cAAc,CAAC,yBAAyB,EAAE,cAAc,CAqKnF,CAAC;AAEF,eAAO,MAAM,mBAAmB,EAAE,cAAc,CAAC,yBAAyB,EAAE,mBAAmB,CA2C9F,CAAC;AA4aF,wBAAgB,iCAAiC,CAChD,KAAK,EAAE,KAAK,CAAC,yBAAyB,CAAC,EACvC,OAAO,EAAE,cAAc,GACrB,MAAM,CAAC,MAAM,EAAE,GAAG,CAAC,GAAG,SAAS,CAiDjC","sourcesContent":["import {\n\tBedrockRuntimeClient,\n\ttype BedrockRuntimeClientConfig,\n\tStopReason as BedrockStopReason,\n\ttype Tool as BedrockTool,\n\tCachePointType,\n\tCacheTTL,\n\ttype ContentBlock,\n\ttype ContentBlockDeltaEvent,\n\ttype ContentBlockStartEvent,\n\ttype ContentBlockStopEvent,\n\tConversationRole,\n\tConverseStreamCommand,\n\ttype ConverseStreamMetadataEvent,\n\tImageFormat,\n\ttype Message,\n\ttype SystemContentBlock,\n\ttype ToolChoice,\n\ttype ToolConfiguration,\n\tToolResultStatus,\n} from \"@aws-sdk/client-bedrock-runtime\";\n\nimport { calculateCost, supportsAdaptiveThinking, supportsXhigh } from \"../models.js\";\nimport type {\n\tApi,\n\tAssistantMessage,\n\tCacheRetention,\n\tContext,\n\tModel,\n\tSimpleStreamOptions,\n\tStopReason,\n\tStreamFunction,\n\tStreamOptions,\n\tTextContent,\n\tThinkingBudgets,\n\tThinkingContent,\n\tThinkingLevel,\n\tTool,\n\tToolCall,\n\tToolResultMessage,\n} from \"../types.js\";\nimport { AssistantMessageEventStream } from \"../utils/event-stream.js\";\nimport { parseStreamingJson } from \"../utils/json-parse.js\";\nimport { sanitizeSurrogates } from \"../utils/sanitize-unicode.js\";\nimport { adjustMaxTokensForThinking, buildBaseOptions, clampReasoning } from \"./simple-options.js\";\nimport { transformMessages } from \"./transform-messages.js\";\n\nexport interface BedrockOptions extends StreamOptions {\n\tregion?: string;\n\tprofile?: string;\n\ttoolChoice?: \"auto\" | \"any\" | \"none\" | { type: \"tool\"; name: string };\n\t/* See https://docs.aws.amazon.com/bedrock/latest/userguide/inference-reasoning.html for supported models. */\n\treasoning?: ThinkingLevel;\n\t/* Custom token budgets per thinking level. Overrides default budgets. */\n\tthinkingBudgets?: ThinkingBudgets;\n\t/* Controls whether thinking text is returned for adaptive-thinking Claude models.\n\t * \"summarized\" returns visible thinking; \"omitted\" returns only an encrypted signature.\n\t * When unset, the API default applies (Opus 4.7+ default to \"omitted\"). Ignored by\n\t * budget-based models. */\n\tthinkingDisplay?: \"summarized\" | \"omitted\";\n\t/* Only supported by Claude 4.x models, see https://docs.aws.amazon.com/bedrock/latest/userguide/claude-messages-extended-thinking.html#claude-messages-extended-thinking-tool-use-interleaved */\n\tinterleavedThinking?: boolean;\n\t/** Key-value pairs attached to the inference request for cost allocation tagging.\n\t * Keys: max 64 chars, no `aws:` prefix. Values: max 256 chars. Max 50 pairs.\n\t * Tags appear in AWS Cost Explorer split cost allocation data.\n\t * @see https://docs.aws.amazon.com/bedrock/latest/APIReference/API_runtime_ConverseStream.html */\n\trequestMetadata?: Record<string, string>;\n}\n\ntype Block = (TextContent | ThinkingContent | ToolCall) & { index?: number; partialJson?: string };\n\nexport const streamBedrock: StreamFunction<\"bedrock-converse-stream\", BedrockOptions> = (\n\tmodel: Model<\"bedrock-converse-stream\">,\n\tcontext: Context,\n\toptions: BedrockOptions = {},\n): AssistantMessageEventStream => {\n\tconst stream = new AssistantMessageEventStream();\n\n\t(async () => {\n\t\tconst output: AssistantMessage = {\n\t\t\trole: \"assistant\",\n\t\t\tcontent: [],\n\t\t\tapi: \"bedrock-converse-stream\" as Api,\n\t\t\tprovider: model.provider,\n\t\t\tmodel: model.id,\n\t\t\tusage: {\n\t\t\t\tinput: 0,\n\t\t\t\toutput: 0,\n\t\t\t\tcacheRead: 0,\n\t\t\t\tcacheWrite: 0,\n\t\t\t\ttotalTokens: 0,\n\t\t\t\tcost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },\n\t\t\t},\n\t\t\tstopReason: \"stop\",\n\t\t\ttimestamp: Date.now(),\n\t\t};\n\n\t\tconst blocks = output.content as Block[];\n\n\t\tconst config: BedrockRuntimeClientConfig = {\n\t\t\tprofile: options.profile,\n\t\t};\n\n\t\t// in Node.js/Bun environment only\n\t\tif (typeof process !== \"undefined\" && (process.versions?.node || process.versions?.bun)) {\n\t\t\t// Region resolution: explicit option > env vars > SDK default chain.\n\t\t\t// When AWS_PROFILE is set, we leave region undefined so the SDK can\n\t\t\t// resovle it from aws profile configs. Otherwise fall back to us-east-1.\n\t\t\tconst explicitRegion = options.region || process.env.AWS_REGION || process.env.AWS_DEFAULT_REGION;\n\t\t\tif (explicitRegion) {\n\t\t\t\tconfig.region = explicitRegion;\n\t\t\t} else if (!process.env.AWS_PROFILE) {\n\t\t\t\tconfig.region = \"us-east-1\";\n\t\t\t}\n\n\t\t\t// Support proxies that don't need authentication\n\t\t\tif (process.env.AWS_BEDROCK_SKIP_AUTH === \"1\") {\n\t\t\t\tconfig.credentials = {\n\t\t\t\t\taccessKeyId: \"dummy-access-key\",\n\t\t\t\t\tsecretAccessKey: \"dummy-secret-key\",\n\t\t\t\t};\n\t\t\t}\n\n\t\t\tif (\n\t\t\t\tprocess.env.HTTP_PROXY ||\n\t\t\t\tprocess.env.HTTPS_PROXY ||\n\t\t\t\tprocess.env.NO_PROXY ||\n\t\t\t\tprocess.env.http_proxy ||\n\t\t\t\tprocess.env.https_proxy ||\n\t\t\t\tprocess.env.no_proxy\n\t\t\t) {\n\t\t\t\tconst nodeHttpHandler = await import(\"@smithy/node-http-handler\");\n\t\t\t\tconst proxyAgent = await import(\"proxy-agent\");\n\n\t\t\t\tconst agent = new proxyAgent.ProxyAgent();\n\n\t\t\t\t// Bedrock runtime uses NodeHttp2Handler by default since v3.798.0, which is based\n\t\t\t\t// on `http2` module and has no support for http agent.\n\t\t\t\t// Use NodeHttpHandler to support http agent.\n\t\t\t\tconfig.requestHandler = new nodeHttpHandler.NodeHttpHandler({\n\t\t\t\t\thttpAgent: agent,\n\t\t\t\t\thttpsAgent: agent,\n\t\t\t\t});\n\t\t\t} else if (process.env.AWS_BEDROCK_FORCE_HTTP1 === \"1\") {\n\t\t\t\t// Some custom endpoints require HTTP/1.1 instead of HTTP/2\n\t\t\t\tconst nodeHttpHandler = await import(\"@smithy/node-http-handler\");\n\t\t\t\tconfig.requestHandler = new nodeHttpHandler.NodeHttpHandler();\n\t\t\t}\n\t\t} else {\n\t\t\t// Non-Node environment (browser): fall back to us-east-1 since\n\t\t\t// there's no config file resolution available.\n\t\t\tconfig.region = options.region || \"us-east-1\";\n\t\t}\n\n\t\ttry {\n\t\t\tconst client = new BedrockRuntimeClient(config);\n\n\t\t\tconst cacheRetention = resolveCacheRetention(options.cacheRetention);\n\t\t\tlet commandInput = {\n\t\t\t\tmodelId: model.id,\n\t\t\t\tmessages: convertMessages(context, model, cacheRetention),\n\t\t\t\tsystem: buildSystemPrompt(context.systemPrompt, model, cacheRetention),\n\t\t\t\tinferenceConfig: { maxTokens: options.maxTokens, temperature: options.temperature },\n\t\t\t\ttoolConfig: convertToolConfig(context.tools, options.toolChoice),\n\t\t\t\tadditionalModelRequestFields: buildAdditionalModelRequestFields(model, options),\n\t\t\t\t...(options.requestMetadata !== undefined && { requestMetadata: options.requestMetadata }),\n\t\t\t};\n\t\t\tconst nextCommandInput = await options?.onPayload?.(commandInput, model);\n\t\t\tif (nextCommandInput !== undefined) {\n\t\t\t\tcommandInput = nextCommandInput as typeof commandInput;\n\t\t\t}\n\t\t\tconst command = new ConverseStreamCommand(commandInput);\n\n\t\t\tconst response = await client.send(command, { abortSignal: options.signal });\n\t\t\tlet receivedMessageStop = false;\n\n\t\t\tfor await (const item of response.stream!) {\n\t\t\t\tif (item.messageStart) {\n\t\t\t\t\tif (item.messageStart.role !== ConversationRole.ASSISTANT) {\n\t\t\t\t\t\tthrow new Error(\"Unexpected assistant message start but got user message start instead\");\n\t\t\t\t\t}\n\t\t\t\t\tstream.push({ type: \"start\", partial: output });\n\t\t\t\t} else if (item.contentBlockStart) {\n\t\t\t\t\thandleContentBlockStart(item.contentBlockStart, blocks, output, stream);\n\t\t\t\t} else if (item.contentBlockDelta) {\n\t\t\t\t\thandleContentBlockDelta(item.contentBlockDelta, blocks, output, stream, options?.onWarning);\n\t\t\t\t} else if (item.contentBlockStop) {\n\t\t\t\t\thandleContentBlockStop(item.contentBlockStop, blocks, output, stream, options?.onWarning);\n\t\t\t\t} else if (item.messageStop) {\n\t\t\t\t\treceivedMessageStop = true;\n\t\t\t\t\toutput.stopReason = mapStopReason(item.messageStop.stopReason);\n\t\t\t\t\tif (item.messageStop.stopReason === BedrockStopReason.MODEL_CONTEXT_WINDOW_EXCEEDED) {\n\t\t\t\t\t\toutput.errorMessage = \"model_context_window_exceeded\";\n\t\t\t\t\t}\n\t\t\t\t} else if (item.metadata) {\n\t\t\t\t\thandleMetadata(item.metadata, model, output);\n\t\t\t\t} else if (item.internalServerException) {\n\t\t\t\t\tthrow new Error(`Internal server error: ${item.internalServerException.message}`);\n\t\t\t\t} else if (item.modelStreamErrorException) {\n\t\t\t\t\tthrow new Error(`Model stream error: ${item.modelStreamErrorException.message}`);\n\t\t\t\t} else if (item.validationException) {\n\t\t\t\t\tthrow new Error(`Validation error: ${item.validationException.message}`);\n\t\t\t\t} else if (item.throttlingException) {\n\t\t\t\t\tthrow new Error(`Throttling error: ${item.throttlingException.message}`);\n\t\t\t\t} else if (item.serviceUnavailableException) {\n\t\t\t\t\tthrow new Error(`Service unavailable: ${item.serviceUnavailableException.message}`);\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tif (options.signal?.aborted) {\n\t\t\t\tthrow new Error(\"Request was aborted\");\n\t\t\t}\n\n\t\t\tif (!receivedMessageStop) {\n\t\t\t\tthrow new Error(\"Stream ended without messageStop — connection likely dropped\");\n\t\t\t}\n\n\t\t\tif (output.stopReason === \"error\" || output.stopReason === \"aborted\") {\n\t\t\t\tthrow new Error(\"An unknown error occurred\");\n\t\t\t}\n\n\t\t\tstream.push({ type: \"done\", reason: output.stopReason, message: output });\n\t\t\tstream.end();\n\t\t} catch (error) {\n\t\t\tfor (const block of output.content) {\n\t\t\t\tdelete (block as Block).index;\n\t\t\t\tdelete (block as Block).partialJson;\n\t\t\t}\n\t\t\toutput.stopReason = options.signal?.aborted ? \"aborted\" : \"error\";\n\t\t\toutput.errorMessage = error instanceof Error ? error.message : JSON.stringify(error);\n\t\t\tstream.push({ type: \"error\", reason: output.stopReason, error: output });\n\t\t\tstream.end();\n\t\t}\n\t})();\n\n\treturn stream;\n};\n\nexport const streamSimpleBedrock: StreamFunction<\"bedrock-converse-stream\", SimpleStreamOptions> = (\n\tmodel: Model<\"bedrock-converse-stream\">,\n\tcontext: Context,\n\toptions?: SimpleStreamOptions,\n): AssistantMessageEventStream => {\n\tconst base = buildBaseOptions(model, options, undefined);\n\tif (!options?.reasoning) {\n\t\treturn streamBedrock(model, context, { ...base, reasoning: undefined } satisfies BedrockOptions);\n\t}\n\n\tif (model.id.includes(\"anthropic.claude\") || model.id.includes(\"anthropic/claude\")) {\n\t\tif (supportsAdaptiveThinking(model)) {\n\t\t\treturn streamBedrock(model, context, {\n\t\t\t\t...base,\n\t\t\t\treasoning: options.reasoning,\n\t\t\t\tthinkingBudgets: options.thinkingBudgets,\n\t\t\t\tthinkingDisplay: options.thinkingDisplay,\n\t\t\t} satisfies BedrockOptions);\n\t\t}\n\n\t\tconst adjusted = adjustMaxTokensForThinking(\n\t\t\tbase.maxTokens || 0,\n\t\t\tmodel.maxTokens,\n\t\t\toptions.reasoning,\n\t\t\toptions.thinkingBudgets,\n\t\t);\n\n\t\treturn streamBedrock(model, context, {\n\t\t\t...base,\n\t\t\tmaxTokens: adjusted.maxTokens,\n\t\t\treasoning: options.reasoning,\n\t\t\tthinkingBudgets: {\n\t\t\t\t...(options.thinkingBudgets || {}),\n\t\t\t\t[clampReasoning(options.reasoning)!]: adjusted.thinkingBudget,\n\t\t\t},\n\t\t} satisfies BedrockOptions);\n\t}\n\n\treturn streamBedrock(model, context, {\n\t\t...base,\n\t\treasoning: options.reasoning,\n\t\tthinkingBudgets: options.thinkingBudgets,\n\t} satisfies BedrockOptions);\n};\n\nfunction handleContentBlockStart(\n\tevent: ContentBlockStartEvent,\n\tblocks: Block[],\n\toutput: AssistantMessage,\n\tstream: AssistantMessageEventStream,\n): void {\n\tconst index = event.contentBlockIndex!;\n\tconst start = event.start;\n\n\tif (start?.toolUse) {\n\t\tconst block: Block = {\n\t\t\ttype: \"toolCall\",\n\t\t\tid: start.toolUse.toolUseId || \"\",\n\t\t\tname: start.toolUse.name || \"\",\n\t\t\targuments: {},\n\t\t\tpartialJson: \"\",\n\t\t\tindex,\n\t\t};\n\t\toutput.content.push(block);\n\t\tstream.push({ type: \"toolcall_start\", contentIndex: blocks.length - 1, partial: output });\n\t}\n}\n\nfunction handleContentBlockDelta(\n\tevent: ContentBlockDeltaEvent,\n\tblocks: Block[],\n\toutput: AssistantMessage,\n\tstream: AssistantMessageEventStream,\n\tonWarning?: (code: string, message: string) => void,\n): void {\n\tconst contentBlockIndex = event.contentBlockIndex!;\n\tconst delta = event.delta;\n\tlet index = blocks.findIndex((b) => b.index === contentBlockIndex);\n\tlet block = blocks[index];\n\n\tif (delta?.text !== undefined) {\n\t\t// If no text block exists yet, create one, as `handleContentBlockStart` is not sent for text blocks\n\t\tif (!block) {\n\t\t\tconst newBlock: Block = { type: \"text\", text: \"\", index: contentBlockIndex };\n\t\t\toutput.content.push(newBlock);\n\t\t\tindex = blocks.length - 1;\n\t\t\tblock = blocks[index];\n\t\t\tstream.push({ type: \"text_start\", contentIndex: index, partial: output });\n\t\t}\n\t\tif (block.type === \"text\") {\n\t\t\tblock.text += delta.text;\n\t\t\tstream.push({ type: \"text_delta\", contentIndex: index, delta: delta.text, partial: output });\n\t\t}\n\t} else if (delta?.toolUse && block?.type === \"toolCall\") {\n\t\tblock.partialJson = (block.partialJson || \"\") + (delta.toolUse.input || \"\");\n\t\tblock.arguments = parseStreamingJson(block.partialJson, onWarning);\n\t\tstream.push({ type: \"toolcall_delta\", contentIndex: index, delta: delta.toolUse.input || \"\", partial: output });\n\t} else if (delta?.reasoningContent) {\n\t\tlet thinkingBlock = block;\n\t\tlet thinkingIndex = index;\n\n\t\tif (!thinkingBlock) {\n\t\t\tconst newBlock: Block = { type: \"thinking\", thinking: \"\", thinkingSignature: \"\", index: contentBlockIndex };\n\t\t\toutput.content.push(newBlock);\n\t\t\tthinkingIndex = blocks.length - 1;\n\t\t\tthinkingBlock = blocks[thinkingIndex];\n\t\t\tstream.push({ type: \"thinking_start\", contentIndex: thinkingIndex, partial: output });\n\t\t}\n\n\t\tif (thinkingBlock?.type === \"thinking\") {\n\t\t\tif (delta.reasoningContent.text) {\n\t\t\t\tthinkingBlock.thinking += delta.reasoningContent.text;\n\t\t\t\tstream.push({\n\t\t\t\t\ttype: \"thinking_delta\",\n\t\t\t\t\tcontentIndex: thinkingIndex,\n\t\t\t\t\tdelta: delta.reasoningContent.text,\n\t\t\t\t\tpartial: output,\n\t\t\t\t});\n\t\t\t}\n\t\t\tif (delta.reasoningContent.signature) {\n\t\t\t\tthinkingBlock.thinkingSignature =\n\t\t\t\t\t(thinkingBlock.thinkingSignature || \"\") + delta.reasoningContent.signature;\n\t\t\t}\n\t\t}\n\t}\n}\n\nfunction handleMetadata(\n\tevent: ConverseStreamMetadataEvent,\n\tmodel: Model<\"bedrock-converse-stream\">,\n\toutput: AssistantMessage,\n): void {\n\tif (event.usage) {\n\t\toutput.usage.input = event.usage.inputTokens || 0;\n\t\toutput.usage.output = event.usage.outputTokens || 0;\n\t\toutput.usage.cacheRead = event.usage.cacheReadInputTokens || 0;\n\t\toutput.usage.cacheWrite = event.usage.cacheWriteInputTokens || 0;\n\t\toutput.usage.totalTokens = event.usage.totalTokens || output.usage.input + output.usage.output;\n\t\tcalculateCost(model, output.usage);\n\t}\n}\n\nfunction handleContentBlockStop(\n\tevent: ContentBlockStopEvent,\n\tblocks: Block[],\n\toutput: AssistantMessage,\n\tstream: AssistantMessageEventStream,\n\tonWarning?: (code: string, message: string) => void,\n): void {\n\tconst index = blocks.findIndex((b) => b.index === event.contentBlockIndex);\n\tconst block = blocks[index];\n\tif (!block) return;\n\tdelete (block as Block).index;\n\n\tswitch (block.type) {\n\t\tcase \"text\":\n\t\t\tstream.push({ type: \"text_end\", contentIndex: index, content: block.text, partial: output });\n\t\t\tbreak;\n\t\tcase \"thinking\":\n\t\t\tstream.push({ type: \"thinking_end\", contentIndex: index, content: block.thinking, partial: output });\n\t\t\tbreak;\n\t\tcase \"toolCall\":\n\t\t\tblock.arguments = parseStreamingJson(block.partialJson, onWarning);\n\t\t\tdelete (block as Block).partialJson;\n\t\t\tstream.push({ type: \"toolcall_end\", contentIndex: index, toolCall: block, partial: output });\n\t\t\tbreak;\n\t}\n}\n\nfunction mapThinkingLevelToEffort(\n\tlevel: SimpleStreamOptions[\"reasoning\"],\n\tmodel: Model<\"bedrock-converse-stream\">,\n): \"low\" | \"medium\" | \"high\" | \"max\" {\n\tswitch (level) {\n\t\tcase \"minimal\":\n\t\tcase \"low\":\n\t\t\treturn \"low\";\n\t\tcase \"medium\":\n\t\t\treturn \"medium\";\n\t\tcase \"high\":\n\t\t\treturn \"high\";\n\t\tcase \"xhigh\":\n\t\tcase \"max\":\n\t\t\treturn supportsXhigh(model) ? \"max\" : \"high\";\n\t\tdefault:\n\t\t\treturn \"high\";\n\t}\n}\n\n/**\n * Resolve cache retention preference.\n * Defaults to \"short\" and uses DREB_CACHE_RETENTION for backward compatibility.\n */\nfunction resolveCacheRetention(cacheRetention?: CacheRetention): CacheRetention {\n\tif (cacheRetention) {\n\t\treturn cacheRetention;\n\t}\n\tif (typeof process !== \"undefined\" && process.env.DREB_CACHE_RETENTION === \"long\") {\n\t\treturn \"long\";\n\t}\n\treturn \"short\";\n}\n\n/**\n * Check if the model supports prompt caching.\n * Supported: Claude 3.5 Haiku, Claude 3.7 Sonnet, Claude 4.x models\n *\n * For base models and system-defined inference profiles the model ID / ARN\n * contains the model name, so we can decide locally.\n *\n * For application inference profiles (whose ARNs don't contain the model name),\n * set AWS_BEDROCK_FORCE_CACHE=1 to enable cache points.  Amazon Nova models\n * have automatic caching and don't need explicit cache points.\n */\nfunction supportsPromptCaching(model: Model<\"bedrock-converse-stream\">): boolean {\n\tconst id = model.id.toLowerCase();\n\tif (!id.includes(\"claude\")) {\n\t\t// Application inference profiles don't contain the model name in the ARN.\n\t\t// Allow users to force cache points via environment variable.\n\t\tif (typeof process !== \"undefined\" && process.env.AWS_BEDROCK_FORCE_CACHE === \"1\") return true;\n\t\treturn false;\n\t}\n\t// Claude 4.x models (opus-4, sonnet-4, haiku-4)\n\tif (id.includes(\"-4-\") || id.includes(\"-4.\")) return true;\n\t// Claude 3.7 Sonnet\n\tif (id.includes(\"claude-3-7-sonnet\")) return true;\n\t// Claude 3.5 Haiku\n\tif (id.includes(\"claude-3-5-haiku\")) return true;\n\treturn false;\n}\n\n/**\n * Check if the model supports thinking signatures in reasoningContent.\n * Only Anthropic Claude models support the signature field.\n * Other models (OpenAI, Qwen, Minimax, Moonshot, etc.) reject it with:\n * \"This model doesn't support the reasoningContent.reasoningText.signature field\"\n */\nfunction supportsThinkingSignature(model: Model<\"bedrock-converse-stream\">): boolean {\n\tconst id = model.id.toLowerCase();\n\treturn id.includes(\"anthropic.claude\") || id.includes(\"anthropic/claude\");\n}\n\nfunction buildSystemPrompt(\n\tsystemPrompt: string | undefined,\n\tmodel: Model<\"bedrock-converse-stream\">,\n\tcacheRetention: CacheRetention,\n): SystemContentBlock[] | undefined {\n\tif (!systemPrompt) return undefined;\n\n\tconst blocks: SystemContentBlock[] = [{ text: sanitizeSurrogates(systemPrompt) }];\n\n\t// Add cache point for supported Claude models when caching is enabled\n\tif (cacheRetention !== \"none\" && supportsPromptCaching(model)) {\n\t\tblocks.push({\n\t\t\tcachePoint: { type: CachePointType.DEFAULT, ...(cacheRetention === \"long\" ? { ttl: CacheTTL.ONE_HOUR } : {}) },\n\t\t});\n\t}\n\n\treturn blocks;\n}\n\nfunction normalizeToolCallId(id: string): string {\n\tconst sanitized = id.replace(/[^a-zA-Z0-9_-]/g, \"_\");\n\treturn sanitized.length > 64 ? sanitized.slice(0, 64) : sanitized;\n}\n\nfunction convertMessages(\n\tcontext: Context,\n\tmodel: Model<\"bedrock-converse-stream\">,\n\tcacheRetention: CacheRetention,\n): Message[] {\n\tconst result: Message[] = [];\n\tconst transformedMessages = transformMessages(context.messages, model, normalizeToolCallId);\n\n\tfor (let i = 0; i < transformedMessages.length; i++) {\n\t\tconst m = transformedMessages[i];\n\n\t\tswitch (m.role) {\n\t\t\tcase \"user\":\n\t\t\t\tresult.push({\n\t\t\t\t\trole: ConversationRole.USER,\n\t\t\t\t\tcontent:\n\t\t\t\t\t\ttypeof m.content === \"string\"\n\t\t\t\t\t\t\t? [{ text: sanitizeSurrogates(m.content) }]\n\t\t\t\t\t\t\t: m.content.map((c) => {\n\t\t\t\t\t\t\t\t\tswitch (c.type) {\n\t\t\t\t\t\t\t\t\t\tcase \"text\":\n\t\t\t\t\t\t\t\t\t\t\treturn { text: sanitizeSurrogates(c.text) };\n\t\t\t\t\t\t\t\t\t\tcase \"image\":\n\t\t\t\t\t\t\t\t\t\t\treturn { image: createImageBlock(c.mimeType, c.data) };\n\t\t\t\t\t\t\t\t\t\tdefault:\n\t\t\t\t\t\t\t\t\t\t\tthrow new Error(\"Unknown user content type\");\n\t\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t}),\n\t\t\t\t});\n\t\t\t\tbreak;\n\t\t\tcase \"assistant\": {\n\t\t\t\t// Skip assistant messages with empty content (e.g., from aborted requests)\n\t\t\t\t// Bedrock rejects messages with empty content arrays\n\t\t\t\tif (m.content.length === 0) {\n\t\t\t\t\tcontinue;\n\t\t\t\t}\n\t\t\t\tconst contentBlocks: ContentBlock[] = [];\n\t\t\t\tfor (const c of m.content) {\n\t\t\t\t\tswitch (c.type) {\n\t\t\t\t\t\tcase \"text\":\n\t\t\t\t\t\t\t// Skip empty text blocks\n\t\t\t\t\t\t\tif (c.text.trim().length === 0) continue;\n\t\t\t\t\t\t\tcontentBlocks.push({ text: sanitizeSurrogates(c.text) });\n\t\t\t\t\t\t\tbreak;\n\t\t\t\t\t\tcase \"toolCall\":\n\t\t\t\t\t\t\tcontentBlocks.push({\n\t\t\t\t\t\t\t\ttoolUse: { toolUseId: c.id, name: c.name, input: c.arguments },\n\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\tbreak;\n\t\t\t\t\t\tcase \"thinking\":\n\t\t\t\t\t\t\t// Skip empty thinking blocks\n\t\t\t\t\t\t\tif (c.thinking.trim().length === 0) continue;\n\t\t\t\t\t\t\t// Only Anthropic models support the signature field in reasoningText.\n\t\t\t\t\t\t\t// For other models, we omit the signature to avoid errors like:\n\t\t\t\t\t\t\t// \"This model doesn't support the reasoningContent.reasoningText.signature field\"\n\t\t\t\t\t\t\tif (supportsThinkingSignature(model)) {\n\t\t\t\t\t\t\t\t// Signatures arrive after thinking deltas. If a partial or externally\n\t\t\t\t\t\t\t\t// persisted message lacks a signature, Bedrock rejects the replayed\n\t\t\t\t\t\t\t\t// reasoning block. Fall back to plain text, matching Anthropic.\n\t\t\t\t\t\t\t\tif (!c.thinkingSignature || c.thinkingSignature.trim().length === 0) {\n\t\t\t\t\t\t\t\t\tcontentBlocks.push({ text: sanitizeSurrogates(c.thinking) });\n\t\t\t\t\t\t\t\t} else {\n\t\t\t\t\t\t\t\t\tcontentBlocks.push({\n\t\t\t\t\t\t\t\t\t\treasoningContent: {\n\t\t\t\t\t\t\t\t\t\t\treasoningText: {\n\t\t\t\t\t\t\t\t\t\t\t\ttext: sanitizeSurrogates(c.thinking),\n\t\t\t\t\t\t\t\t\t\t\t\tsignature: c.thinkingSignature,\n\t\t\t\t\t\t\t\t\t\t\t},\n\t\t\t\t\t\t\t\t\t\t},\n\t\t\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t} else {\n\t\t\t\t\t\t\t\tcontentBlocks.push({\n\t\t\t\t\t\t\t\t\treasoningContent: {\n\t\t\t\t\t\t\t\t\t\treasoningText: { text: sanitizeSurrogates(c.thinking) },\n\t\t\t\t\t\t\t\t\t},\n\t\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\tbreak;\n\t\t\t\t\t\tdefault:\n\t\t\t\t\t\t\tthrow new Error(\"Unknown assistant content type\");\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\t// Skip if all content blocks were filtered out\n\t\t\t\tif (contentBlocks.length === 0) {\n\t\t\t\t\tcontinue;\n\t\t\t\t}\n\t\t\t\tresult.push({\n\t\t\t\t\trole: ConversationRole.ASSISTANT,\n\t\t\t\t\tcontent: contentBlocks,\n\t\t\t\t});\n\t\t\t\tbreak;\n\t\t\t}\n\t\t\tcase \"toolResult\": {\n\t\t\t\t// Collect all consecutive toolResult messages into a single user message\n\t\t\t\t// Bedrock requires all tool results to be in one message\n\t\t\t\tconst toolResults: ContentBlock.ToolResultMember[] = [];\n\n\t\t\t\t// Add current tool result with all content blocks combined\n\t\t\t\ttoolResults.push({\n\t\t\t\t\ttoolResult: {\n\t\t\t\t\t\ttoolUseId: m.toolCallId,\n\t\t\t\t\t\tcontent: m.content.map((c) =>\n\t\t\t\t\t\t\tc.type === \"image\"\n\t\t\t\t\t\t\t\t? { image: createImageBlock(c.mimeType, c.data) }\n\t\t\t\t\t\t\t\t: { text: sanitizeSurrogates(c.text) },\n\t\t\t\t\t\t),\n\t\t\t\t\t\tstatus: m.isError ? ToolResultStatus.ERROR : ToolResultStatus.SUCCESS,\n\t\t\t\t\t},\n\t\t\t\t});\n\n\t\t\t\t// Look ahead for consecutive toolResult messages\n\t\t\t\tlet j = i + 1;\n\t\t\t\twhile (j < transformedMessages.length && transformedMessages[j].role === \"toolResult\") {\n\t\t\t\t\tconst nextMsg = transformedMessages[j] as ToolResultMessage;\n\t\t\t\t\ttoolResults.push({\n\t\t\t\t\t\ttoolResult: {\n\t\t\t\t\t\t\ttoolUseId: nextMsg.toolCallId,\n\t\t\t\t\t\t\tcontent: nextMsg.content.map((c) =>\n\t\t\t\t\t\t\t\tc.type === \"image\"\n\t\t\t\t\t\t\t\t\t? { image: createImageBlock(c.mimeType, c.data) }\n\t\t\t\t\t\t\t\t\t: { text: sanitizeSurrogates(c.text) },\n\t\t\t\t\t\t\t),\n\t\t\t\t\t\t\tstatus: nextMsg.isError ? ToolResultStatus.ERROR : ToolResultStatus.SUCCESS,\n\t\t\t\t\t\t},\n\t\t\t\t\t});\n\t\t\t\t\tj++;\n\t\t\t\t}\n\n\t\t\t\t// Skip the messages we've already processed\n\t\t\t\ti = j - 1;\n\n\t\t\t\tresult.push({\n\t\t\t\t\trole: ConversationRole.USER,\n\t\t\t\t\tcontent: toolResults,\n\t\t\t\t});\n\t\t\t\tbreak;\n\t\t\t}\n\t\t\tdefault:\n\t\t\t\tthrow new Error(\"Unknown message role\");\n\t\t}\n\t}\n\n\t// Add cache point to the last user message for supported Claude models when caching is enabled\n\tif (cacheRetention !== \"none\" && supportsPromptCaching(model) && result.length > 0) {\n\t\tconst lastMessage = result[result.length - 1];\n\t\tif (lastMessage.role === ConversationRole.USER && lastMessage.content) {\n\t\t\t(lastMessage.content as ContentBlock[]).push({\n\t\t\t\tcachePoint: {\n\t\t\t\t\ttype: CachePointType.DEFAULT,\n\t\t\t\t\t...(cacheRetention === \"long\" ? { ttl: CacheTTL.ONE_HOUR } : {}),\n\t\t\t\t},\n\t\t\t});\n\t\t}\n\t}\n\n\treturn result;\n}\n\nfunction convertToolConfig(\n\ttools: Tool[] | undefined,\n\ttoolChoice: BedrockOptions[\"toolChoice\"],\n): ToolConfiguration | undefined {\n\tif (!tools?.length || toolChoice === \"none\") return undefined;\n\n\tconst bedrockTools: BedrockTool[] = tools.map((tool) => ({\n\t\ttoolSpec: {\n\t\t\tname: tool.name,\n\t\t\tdescription: tool.description,\n\t\t\tinputSchema: { json: tool.parameters },\n\t\t},\n\t}));\n\n\tlet bedrockToolChoice: ToolChoice | undefined;\n\tswitch (toolChoice) {\n\t\tcase \"auto\":\n\t\t\tbedrockToolChoice = { auto: {} };\n\t\t\tbreak;\n\t\tcase \"any\":\n\t\t\tbedrockToolChoice = { any: {} };\n\t\t\tbreak;\n\t\tdefault:\n\t\t\tif (toolChoice?.type === \"tool\") {\n\t\t\t\tbedrockToolChoice = { tool: { name: toolChoice.name } };\n\t\t\t}\n\t}\n\n\treturn { tools: bedrockTools, toolChoice: bedrockToolChoice };\n}\n\nfunction mapStopReason(reason: string | undefined): StopReason {\n\tswitch (reason) {\n\t\tcase BedrockStopReason.END_TURN:\n\t\tcase BedrockStopReason.STOP_SEQUENCE:\n\t\t\treturn \"stop\";\n\t\tcase BedrockStopReason.MAX_TOKENS:\n\t\tcase BedrockStopReason.MODEL_CONTEXT_WINDOW_EXCEEDED:\n\t\t\treturn \"length\";\n\t\tcase BedrockStopReason.TOOL_USE:\n\t\t\treturn \"toolUse\";\n\t\tdefault:\n\t\t\treturn \"error\";\n\t}\n}\n\nexport function buildAdditionalModelRequestFields(\n\tmodel: Model<\"bedrock-converse-stream\">,\n\toptions: BedrockOptions,\n): Record<string, any> | undefined {\n\tif (!options.reasoning || !model.reasoning) {\n\t\treturn undefined;\n\t}\n\n\tif (model.id.includes(\"anthropic.claude\") || model.id.includes(\"anthropic/claude\")) {\n\t\tconst result: Record<string, any> = supportsAdaptiveThinking(model)\n\t\t\t? (() => {\n\t\t\t\t\t// Adaptive thinking: the `display` field controls whether thinking text is\n\t\t\t\t\t// returned. Opus 4.7+ default to \"omitted\"; send \"summarized\" to make it visible.\n\t\t\t\t\tconst thinking: Record<string, any> = { type: \"adaptive\" };\n\t\t\t\t\tif (options.thinkingDisplay) {\n\t\t\t\t\t\tthinking.display = options.thinkingDisplay;\n\t\t\t\t\t}\n\t\t\t\t\treturn {\n\t\t\t\t\t\tthinking,\n\t\t\t\t\t\toutput_config: { effort: mapThinkingLevelToEffort(options.reasoning, model) },\n\t\t\t\t\t};\n\t\t\t\t})()\n\t\t\t: (() => {\n\t\t\t\t\tconst defaultBudgets: Record<ThinkingLevel, number> = {\n\t\t\t\t\t\tminimal: 1024,\n\t\t\t\t\t\tlow: 2048,\n\t\t\t\t\t\tmedium: 8192,\n\t\t\t\t\t\thigh: 16384,\n\t\t\t\t\t\txhigh: 16384, // Claude doesn't support xhigh, clamp to high\n\t\t\t\t\t\tmax: 16384, // Normalized max falls back through xhigh to high here\n\t\t\t\t\t};\n\n\t\t\t\t\t// Custom budgets do not define xhigh/max, so both use high.\n\t\t\t\t\tconst level = options.reasoning === \"xhigh\" || options.reasoning === \"max\" ? \"high\" : options.reasoning;\n\t\t\t\t\tconst budget = options.thinkingBudgets?.[level] ?? defaultBudgets[options.reasoning];\n\n\t\t\t\t\treturn {\n\t\t\t\t\t\tthinking: {\n\t\t\t\t\t\t\ttype: \"enabled\",\n\t\t\t\t\t\t\tbudget_tokens: budget,\n\t\t\t\t\t\t},\n\t\t\t\t\t};\n\t\t\t\t})();\n\n\t\tif (!supportsAdaptiveThinking(model) && (options.interleavedThinking ?? true)) {\n\t\t\tresult.anthropic_beta = [\"interleaved-thinking-2025-05-14\"];\n\t\t}\n\n\t\treturn result;\n\t}\n\n\treturn undefined;\n}\n\nfunction createImageBlock(mimeType: string, data: string) {\n\tlet format: ImageFormat;\n\tswitch (mimeType) {\n\t\tcase \"image/jpeg\":\n\t\tcase \"image/jpg\":\n\t\t\tformat = ImageFormat.JPEG;\n\t\t\tbreak;\n\t\tcase \"image/png\":\n\t\t\tformat = ImageFormat.PNG;\n\t\t\tbreak;\n\t\tcase \"image/gif\":\n\t\t\tformat = ImageFormat.GIF;\n\t\t\tbreak;\n\t\tcase \"image/webp\":\n\t\t\tformat = ImageFormat.WEBP;\n\t\t\tbreak;\n\t\tdefault:\n\t\t\tthrow new Error(`Unknown image type: ${mimeType}`);\n\t}\n\n\tconst binaryString = atob(data);\n\tconst bytes = new Uint8Array(binaryString.length);\n\tfor (let i = 0; i < binaryString.length; i++) {\n\t\tbytes[i] = binaryString.charCodeAt(i);\n\t}\n\n\treturn { source: { bytes }, format };\n}\n"]}
+{"version":3,"file":"amazon-bedrock.d.ts","sourceRoot":"","sources":["../../src/providers/amazon-bedrock.ts"],"names":[],"mappings":"AA6BA,OAAO,KAAK,EAKX,KAAK,EACL,mBAAmB,EAEnB,cAAc,EACd,aAAa,EAEb,eAAe,EAEf,aAAa,EAIb,MAAM,aAAa,CAAC;AAOrB,MAAM,WAAW,cAAe,SAAQ,aAAa;IACpD,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,UAAU,CAAC,EAAE,MAAM,GAAG,KAAK,GAAG,MAAM,GAAG;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAA;KAAE,CAAC;IAEtE,SAAS,CAAC,EAAE,aAAa,CAAC;IAE1B,eAAe,CAAC,EAAE,eAAe,CAAC;IAKlC,eAAe,CAAC,EAAE,YAAY,GAAG,SAAS,CAAC;IAE3C,mBAAmB,CAAC,EAAE,OAAO,CAAC;IAC9B;;;sGAGkG;IAClG,eAAe,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;CACzC;AAID,eAAO,MAAM,aAAa,EAAE,cAAc,CAAC,yBAAyB,EAAE,cAAc,CAqKnF,CAAC;AAEF,eAAO,MAAM,mBAAmB,EAAE,cAAc,CAAC,yBAAyB,EAAE,mBAAmB,CA2C9F,CAAC;AAwZF,wBAAgB,iCAAiC,CAChD,KAAK,EAAE,KAAK,CAAC,yBAAyB,CAAC,EACvC,OAAO,EAAE,cAAc,GACrB,MAAM,CAAC,MAAM,EAAE,GAAG,CAAC,GAAG,SAAS,CA4DjC","sourcesContent":["import {\n\tBedrockRuntimeClient,\n\ttype BedrockRuntimeClientConfig,\n\tStopReason as BedrockStopReason,\n\ttype Tool as BedrockTool,\n\tCachePointType,\n\tCacheTTL,\n\ttype ContentBlock,\n\ttype ContentBlockDeltaEvent,\n\ttype ContentBlockStartEvent,\n\ttype ContentBlockStopEvent,\n\tConversationRole,\n\tConverseStreamCommand,\n\ttype ConverseStreamMetadataEvent,\n\tImageFormat,\n\ttype Message,\n\ttype SystemContentBlock,\n\ttype ToolChoice,\n\ttype ToolConfiguration,\n\tToolResultStatus,\n} from \"@aws-sdk/client-bedrock-runtime\";\n\nimport {\n\tcalculateCost,\n\tclaudeThinksByDefault,\n\tgetClaudeEffort,\n\tgetClaudeThinkingOffMode,\n\tsupportsAdaptiveThinking,\n} from \"../models.js\";\nimport type {\n\tApi,\n\tAssistantMessage,\n\tCacheRetention,\n\tContext,\n\tModel,\n\tSimpleStreamOptions,\n\tStopReason,\n\tStreamFunction,\n\tStreamOptions,\n\tTextContent,\n\tThinkingBudgets,\n\tThinkingContent,\n\tThinkingLevel,\n\tTool,\n\tToolCall,\n\tToolResultMessage,\n} from \"../types.js\";\nimport { AssistantMessageEventStream } from \"../utils/event-stream.js\";\nimport { parseStreamingJson } from \"../utils/json-parse.js\";\nimport { sanitizeSurrogates } from \"../utils/sanitize-unicode.js\";\nimport { adjustMaxTokensForThinking, buildBaseOptions, clampReasoning } from \"./simple-options.js\";\nimport { transformMessages } from \"./transform-messages.js\";\n\nexport interface BedrockOptions extends StreamOptions {\n\tregion?: string;\n\tprofile?: string;\n\ttoolChoice?: \"auto\" | \"any\" | \"none\" | { type: \"tool\"; name: string };\n\t/* See https://docs.aws.amazon.com/bedrock/latest/userguide/inference-reasoning.html for supported models. */\n\treasoning?: ThinkingLevel;\n\t/* Custom token budgets per thinking level. Overrides default budgets. */\n\tthinkingBudgets?: ThinkingBudgets;\n\t/* Controls whether thinking text is returned for adaptive-thinking Claude models.\n\t * \"summarized\" returns visible thinking; \"omitted\" returns only an encrypted signature.\n\t * When unset, the API default applies (Opus 4.7+ default to \"omitted\"). Ignored by\n\t * budget-based models. */\n\tthinkingDisplay?: \"summarized\" | \"omitted\";\n\t/* Only supported by Claude 4.x models, see https://docs.aws.amazon.com/bedrock/latest/userguide/claude-messages-extended-thinking.html#claude-messages-extended-thinking-tool-use-interleaved */\n\tinterleavedThinking?: boolean;\n\t/** Key-value pairs attached to the inference request for cost allocation tagging.\n\t * Keys: max 64 chars, no `aws:` prefix. Values: max 256 chars. Max 50 pairs.\n\t * Tags appear in AWS Cost Explorer split cost allocation data.\n\t * @see https://docs.aws.amazon.com/bedrock/latest/APIReference/API_runtime_ConverseStream.html */\n\trequestMetadata?: Record<string, string>;\n}\n\ntype Block = (TextContent | ThinkingContent | ToolCall) & { index?: number; partialJson?: string };\n\nexport const streamBedrock: StreamFunction<\"bedrock-converse-stream\", BedrockOptions> = (\n\tmodel: Model<\"bedrock-converse-stream\">,\n\tcontext: Context,\n\toptions: BedrockOptions = {},\n): AssistantMessageEventStream => {\n\tconst stream = new AssistantMessageEventStream();\n\n\t(async () => {\n\t\tconst output: AssistantMessage = {\n\t\t\trole: \"assistant\",\n\t\t\tcontent: [],\n\t\t\tapi: \"bedrock-converse-stream\" as Api,\n\t\t\tprovider: model.provider,\n\t\t\tmodel: model.id,\n\t\t\tusage: {\n\t\t\t\tinput: 0,\n\t\t\t\toutput: 0,\n\t\t\t\tcacheRead: 0,\n\t\t\t\tcacheWrite: 0,\n\t\t\t\ttotalTokens: 0,\n\t\t\t\tcost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },\n\t\t\t},\n\t\t\tstopReason: \"stop\",\n\t\t\ttimestamp: Date.now(),\n\t\t};\n\n\t\tconst blocks = output.content as Block[];\n\n\t\tconst config: BedrockRuntimeClientConfig = {\n\t\t\tprofile: options.profile,\n\t\t};\n\n\t\t// in Node.js/Bun environment only\n\t\tif (typeof process !== \"undefined\" && (process.versions?.node || process.versions?.bun)) {\n\t\t\t// Region resolution: explicit option > env vars > SDK default chain.\n\t\t\t// When AWS_PROFILE is set, we leave region undefined so the SDK can\n\t\t\t// resovle it from aws profile configs. Otherwise fall back to us-east-1.\n\t\t\tconst explicitRegion = options.region || process.env.AWS_REGION || process.env.AWS_DEFAULT_REGION;\n\t\t\tif (explicitRegion) {\n\t\t\t\tconfig.region = explicitRegion;\n\t\t\t} else if (!process.env.AWS_PROFILE) {\n\t\t\t\tconfig.region = \"us-east-1\";\n\t\t\t}\n\n\t\t\t// Support proxies that don't need authentication\n\t\t\tif (process.env.AWS_BEDROCK_SKIP_AUTH === \"1\") {\n\t\t\t\tconfig.credentials = {\n\t\t\t\t\taccessKeyId: \"dummy-access-key\",\n\t\t\t\t\tsecretAccessKey: \"dummy-secret-key\",\n\t\t\t\t};\n\t\t\t}\n\n\t\t\tif (\n\t\t\t\tprocess.env.HTTP_PROXY ||\n\t\t\t\tprocess.env.HTTPS_PROXY ||\n\t\t\t\tprocess.env.NO_PROXY ||\n\t\t\t\tprocess.env.http_proxy ||\n\t\t\t\tprocess.env.https_proxy ||\n\t\t\t\tprocess.env.no_proxy\n\t\t\t) {\n\t\t\t\tconst nodeHttpHandler = await import(\"@smithy/node-http-handler\");\n\t\t\t\tconst proxyAgent = await import(\"proxy-agent\");\n\n\t\t\t\tconst agent = new proxyAgent.ProxyAgent();\n\n\t\t\t\t// Bedrock runtime uses NodeHttp2Handler by default since v3.798.0, which is based\n\t\t\t\t// on `http2` module and has no support for http agent.\n\t\t\t\t// Use NodeHttpHandler to support http agent.\n\t\t\t\tconfig.requestHandler = new nodeHttpHandler.NodeHttpHandler({\n\t\t\t\t\thttpAgent: agent,\n\t\t\t\t\thttpsAgent: agent,\n\t\t\t\t});\n\t\t\t} else if (process.env.AWS_BEDROCK_FORCE_HTTP1 === \"1\") {\n\t\t\t\t// Some custom endpoints require HTTP/1.1 instead of HTTP/2\n\t\t\t\tconst nodeHttpHandler = await import(\"@smithy/node-http-handler\");\n\t\t\t\tconfig.requestHandler = new nodeHttpHandler.NodeHttpHandler();\n\t\t\t}\n\t\t} else {\n\t\t\t// Non-Node environment (browser): fall back to us-east-1 since\n\t\t\t// there's no config file resolution available.\n\t\t\tconfig.region = options.region || \"us-east-1\";\n\t\t}\n\n\t\ttry {\n\t\t\tconst client = new BedrockRuntimeClient(config);\n\n\t\t\tconst cacheRetention = resolveCacheRetention(options.cacheRetention);\n\t\t\tlet commandInput = {\n\t\t\t\tmodelId: model.id,\n\t\t\t\tmessages: convertMessages(context, model, cacheRetention),\n\t\t\t\tsystem: buildSystemPrompt(context.systemPrompt, model, cacheRetention),\n\t\t\t\tinferenceConfig: { maxTokens: options.maxTokens, temperature: options.temperature },\n\t\t\t\ttoolConfig: convertToolConfig(context.tools, options.toolChoice),\n\t\t\t\tadditionalModelRequestFields: buildAdditionalModelRequestFields(model, options),\n\t\t\t\t...(options.requestMetadata !== undefined && { requestMetadata: options.requestMetadata }),\n\t\t\t};\n\t\t\tconst nextCommandInput = await options?.onPayload?.(commandInput, model);\n\t\t\tif (nextCommandInput !== undefined) {\n\t\t\t\tcommandInput = nextCommandInput as typeof commandInput;\n\t\t\t}\n\t\t\tconst command = new ConverseStreamCommand(commandInput);\n\n\t\t\tconst response = await client.send(command, { abortSignal: options.signal });\n\t\t\tlet receivedMessageStop = false;\n\n\t\t\tfor await (const item of response.stream!) {\n\t\t\t\tif (item.messageStart) {\n\t\t\t\t\tif (item.messageStart.role !== ConversationRole.ASSISTANT) {\n\t\t\t\t\t\tthrow new Error(\"Unexpected assistant message start but got user message start instead\");\n\t\t\t\t\t}\n\t\t\t\t\tstream.push({ type: \"start\", partial: output });\n\t\t\t\t} else if (item.contentBlockStart) {\n\t\t\t\t\thandleContentBlockStart(item.contentBlockStart, blocks, output, stream);\n\t\t\t\t} else if (item.contentBlockDelta) {\n\t\t\t\t\thandleContentBlockDelta(item.contentBlockDelta, blocks, output, stream, options?.onWarning);\n\t\t\t\t} else if (item.contentBlockStop) {\n\t\t\t\t\thandleContentBlockStop(item.contentBlockStop, blocks, output, stream, options?.onWarning);\n\t\t\t\t} else if (item.messageStop) {\n\t\t\t\t\treceivedMessageStop = true;\n\t\t\t\t\toutput.stopReason = mapStopReason(item.messageStop.stopReason);\n\t\t\t\t\tif (item.messageStop.stopReason === BedrockStopReason.MODEL_CONTEXT_WINDOW_EXCEEDED) {\n\t\t\t\t\t\toutput.errorMessage = \"model_context_window_exceeded\";\n\t\t\t\t\t}\n\t\t\t\t} else if (item.metadata) {\n\t\t\t\t\thandleMetadata(item.metadata, model, output);\n\t\t\t\t} else if (item.internalServerException) {\n\t\t\t\t\tthrow new Error(`Internal server error: ${item.internalServerException.message}`);\n\t\t\t\t} else if (item.modelStreamErrorException) {\n\t\t\t\t\tthrow new Error(`Model stream error: ${item.modelStreamErrorException.message}`);\n\t\t\t\t} else if (item.validationException) {\n\t\t\t\t\tthrow new Error(`Validation error: ${item.validationException.message}`);\n\t\t\t\t} else if (item.throttlingException) {\n\t\t\t\t\tthrow new Error(`Throttling error: ${item.throttlingException.message}`);\n\t\t\t\t} else if (item.serviceUnavailableException) {\n\t\t\t\t\tthrow new Error(`Service unavailable: ${item.serviceUnavailableException.message}`);\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tif (options.signal?.aborted) {\n\t\t\t\tthrow new Error(\"Request was aborted\");\n\t\t\t}\n\n\t\t\tif (!receivedMessageStop) {\n\t\t\t\tthrow new Error(\"Stream ended without messageStop — connection likely dropped\");\n\t\t\t}\n\n\t\t\tif (output.stopReason === \"error\" || output.stopReason === \"aborted\") {\n\t\t\t\tthrow new Error(\"An unknown error occurred\");\n\t\t\t}\n\n\t\t\tstream.push({ type: \"done\", reason: output.stopReason, message: output });\n\t\t\tstream.end();\n\t\t} catch (error) {\n\t\t\tfor (const block of output.content) {\n\t\t\t\tdelete (block as Block).index;\n\t\t\t\tdelete (block as Block).partialJson;\n\t\t\t}\n\t\t\toutput.stopReason = options.signal?.aborted ? \"aborted\" : \"error\";\n\t\t\toutput.errorMessage = error instanceof Error ? error.message : JSON.stringify(error);\n\t\t\tstream.push({ type: \"error\", reason: output.stopReason, error: output });\n\t\t\tstream.end();\n\t\t}\n\t})();\n\n\treturn stream;\n};\n\nexport const streamSimpleBedrock: StreamFunction<\"bedrock-converse-stream\", SimpleStreamOptions> = (\n\tmodel: Model<\"bedrock-converse-stream\">,\n\tcontext: Context,\n\toptions?: SimpleStreamOptions,\n): AssistantMessageEventStream => {\n\tconst base = buildBaseOptions(model, options, undefined);\n\tif (!options?.reasoning) {\n\t\treturn streamBedrock(model, context, { ...base, reasoning: undefined } satisfies BedrockOptions);\n\t}\n\n\tif (model.id.includes(\"anthropic.claude\") || model.id.includes(\"anthropic/claude\")) {\n\t\tif (supportsAdaptiveThinking(model)) {\n\t\t\treturn streamBedrock(model, context, {\n\t\t\t\t...base,\n\t\t\t\treasoning: options.reasoning,\n\t\t\t\tthinkingBudgets: options.thinkingBudgets,\n\t\t\t\tthinkingDisplay: options.thinkingDisplay,\n\t\t\t} satisfies BedrockOptions);\n\t\t}\n\n\t\tconst adjusted = adjustMaxTokensForThinking(\n\t\t\tbase.maxTokens || 0,\n\t\t\tmodel.maxTokens,\n\t\t\toptions.reasoning,\n\t\t\toptions.thinkingBudgets,\n\t\t);\n\n\t\treturn streamBedrock(model, context, {\n\t\t\t...base,\n\t\t\tmaxTokens: adjusted.maxTokens,\n\t\t\treasoning: options.reasoning,\n\t\t\tthinkingBudgets: {\n\t\t\t\t...(options.thinkingBudgets || {}),\n\t\t\t\t[clampReasoning(options.reasoning)!]: adjusted.thinkingBudget,\n\t\t\t},\n\t\t} satisfies BedrockOptions);\n\t}\n\n\treturn streamBedrock(model, context, {\n\t\t...base,\n\t\treasoning: options.reasoning,\n\t\tthinkingBudgets: options.thinkingBudgets,\n\t} satisfies BedrockOptions);\n};\n\nfunction handleContentBlockStart(\n\tevent: ContentBlockStartEvent,\n\tblocks: Block[],\n\toutput: AssistantMessage,\n\tstream: AssistantMessageEventStream,\n): void {\n\tconst index = event.contentBlockIndex!;\n\tconst start = event.start;\n\n\tif (start?.toolUse) {\n\t\tconst block: Block = {\n\t\t\ttype: \"toolCall\",\n\t\t\tid: start.toolUse.toolUseId || \"\",\n\t\t\tname: start.toolUse.name || \"\",\n\t\t\targuments: {},\n\t\t\tpartialJson: \"\",\n\t\t\tindex,\n\t\t};\n\t\toutput.content.push(block);\n\t\tstream.push({ type: \"toolcall_start\", contentIndex: blocks.length - 1, partial: output });\n\t}\n}\n\nfunction handleContentBlockDelta(\n\tevent: ContentBlockDeltaEvent,\n\tblocks: Block[],\n\toutput: AssistantMessage,\n\tstream: AssistantMessageEventStream,\n\tonWarning?: (code: string, message: string) => void,\n): void {\n\tconst contentBlockIndex = event.contentBlockIndex!;\n\tconst delta = event.delta;\n\tlet index = blocks.findIndex((b) => b.index === contentBlockIndex);\n\tlet block = blocks[index];\n\n\tif (delta?.text !== undefined) {\n\t\t// If no text block exists yet, create one, as `handleContentBlockStart` is not sent for text blocks\n\t\tif (!block) {\n\t\t\tconst newBlock: Block = { type: \"text\", text: \"\", index: contentBlockIndex };\n\t\t\toutput.content.push(newBlock);\n\t\t\tindex = blocks.length - 1;\n\t\t\tblock = blocks[index];\n\t\t\tstream.push({ type: \"text_start\", contentIndex: index, partial: output });\n\t\t}\n\t\tif (block.type === \"text\") {\n\t\t\tblock.text += delta.text;\n\t\t\tstream.push({ type: \"text_delta\", contentIndex: index, delta: delta.text, partial: output });\n\t\t}\n\t} else if (delta?.toolUse && block?.type === \"toolCall\") {\n\t\tblock.partialJson = (block.partialJson || \"\") + (delta.toolUse.input || \"\");\n\t\tblock.arguments = parseStreamingJson(block.partialJson, onWarning);\n\t\tstream.push({ type: \"toolcall_delta\", contentIndex: index, delta: delta.toolUse.input || \"\", partial: output });\n\t} else if (delta?.reasoningContent) {\n\t\tlet thinkingBlock = block;\n\t\tlet thinkingIndex = index;\n\n\t\tif (!thinkingBlock) {\n\t\t\tconst newBlock: Block = { type: \"thinking\", thinking: \"\", thinkingSignature: \"\", index: contentBlockIndex };\n\t\t\toutput.content.push(newBlock);\n\t\t\tthinkingIndex = blocks.length - 1;\n\t\t\tthinkingBlock = blocks[thinkingIndex];\n\t\t\tstream.push({ type: \"thinking_start\", contentIndex: thinkingIndex, partial: output });\n\t\t}\n\n\t\tif (thinkingBlock?.type === \"thinking\") {\n\t\t\tif (delta.reasoningContent.text) {\n\t\t\t\tthinkingBlock.thinking += delta.reasoningContent.text;\n\t\t\t\tstream.push({\n\t\t\t\t\ttype: \"thinking_delta\",\n\t\t\t\t\tcontentIndex: thinkingIndex,\n\t\t\t\t\tdelta: delta.reasoningContent.text,\n\t\t\t\t\tpartial: output,\n\t\t\t\t});\n\t\t\t}\n\t\t\tif (delta.reasoningContent.signature) {\n\t\t\t\tthinkingBlock.thinkingSignature =\n\t\t\t\t\t(thinkingBlock.thinkingSignature || \"\") + delta.reasoningContent.signature;\n\t\t\t}\n\t\t}\n\t}\n}\n\nfunction handleMetadata(\n\tevent: ConverseStreamMetadataEvent,\n\tmodel: Model<\"bedrock-converse-stream\">,\n\toutput: AssistantMessage,\n): void {\n\tif (event.usage) {\n\t\toutput.usage.input = event.usage.inputTokens || 0;\n\t\toutput.usage.output = event.usage.outputTokens || 0;\n\t\toutput.usage.cacheRead = event.usage.cacheReadInputTokens || 0;\n\t\toutput.usage.cacheWrite = event.usage.cacheWriteInputTokens || 0;\n\t\toutput.usage.totalTokens = event.usage.totalTokens || output.usage.input + output.usage.output;\n\t\tcalculateCost(model, output.usage);\n\t}\n}\n\nfunction handleContentBlockStop(\n\tevent: ContentBlockStopEvent,\n\tblocks: Block[],\n\toutput: AssistantMessage,\n\tstream: AssistantMessageEventStream,\n\tonWarning?: (code: string, message: string) => void,\n): void {\n\tconst index = blocks.findIndex((b) => b.index === event.contentBlockIndex);\n\tconst block = blocks[index];\n\tif (!block) return;\n\tdelete (block as Block).index;\n\n\tswitch (block.type) {\n\t\tcase \"text\":\n\t\t\tstream.push({ type: \"text_end\", contentIndex: index, content: block.text, partial: output });\n\t\t\tbreak;\n\t\tcase \"thinking\":\n\t\t\tstream.push({ type: \"thinking_end\", contentIndex: index, content: block.thinking, partial: output });\n\t\t\tbreak;\n\t\tcase \"toolCall\":\n\t\t\tblock.arguments = parseStreamingJson(block.partialJson, onWarning);\n\t\t\tdelete (block as Block).partialJson;\n\t\t\tstream.push({ type: \"toolcall_end\", contentIndex: index, toolCall: block, partial: output });\n\t\t\tbreak;\n\t}\n}\n\n/**\n * Resolve cache retention preference.\n * Defaults to \"short\" and uses DREB_CACHE_RETENTION for backward compatibility.\n */\nfunction resolveCacheRetention(cacheRetention?: CacheRetention): CacheRetention {\n\tif (cacheRetention) {\n\t\treturn cacheRetention;\n\t}\n\tif (typeof process !== \"undefined\" && process.env.DREB_CACHE_RETENTION === \"long\") {\n\t\treturn \"long\";\n\t}\n\treturn \"short\";\n}\n\n/**\n * Check if the model supports prompt caching.\n * Supported: Claude 3.5 Haiku, Claude 3.7 Sonnet, Claude 4.x models\n *\n * For base models and system-defined inference profiles the model ID / ARN\n * contains the model name, so we can decide locally.\n *\n * For application inference profiles (whose ARNs don't contain the model name),\n * set AWS_BEDROCK_FORCE_CACHE=1 to enable cache points.  Amazon Nova models\n * have automatic caching and don't need explicit cache points.\n */\nfunction supportsPromptCaching(model: Model<\"bedrock-converse-stream\">): boolean {\n\tconst id = model.id.toLowerCase();\n\tif (!id.includes(\"claude\")) {\n\t\t// Application inference profiles don't contain the model name in the ARN.\n\t\t// Allow users to force cache points via environment variable.\n\t\tif (typeof process !== \"undefined\" && process.env.AWS_BEDROCK_FORCE_CACHE === \"1\") return true;\n\t\treturn false;\n\t}\n\t// Claude 4.x models (opus-4, sonnet-4, haiku-4)\n\tif (id.includes(\"-4-\") || id.includes(\"-4.\")) return true;\n\t// Claude 3.7 Sonnet\n\tif (id.includes(\"claude-3-7-sonnet\")) return true;\n\t// Claude 3.5 Haiku\n\tif (id.includes(\"claude-3-5-haiku\")) return true;\n\treturn false;\n}\n\n/**\n * Check if the model supports thinking signatures in reasoningContent.\n * Only Anthropic Claude models support the signature field.\n * Other models (OpenAI, Qwen, Minimax, Moonshot, etc.) reject it with:\n * \"This model doesn't support the reasoningContent.reasoningText.signature field\"\n */\nfunction supportsThinkingSignature(model: Model<\"bedrock-converse-stream\">): boolean {\n\tconst id = model.id.toLowerCase();\n\treturn id.includes(\"anthropic.claude\") || id.includes(\"anthropic/claude\");\n}\n\nfunction buildSystemPrompt(\n\tsystemPrompt: string | undefined,\n\tmodel: Model<\"bedrock-converse-stream\">,\n\tcacheRetention: CacheRetention,\n): SystemContentBlock[] | undefined {\n\tif (!systemPrompt) return undefined;\n\n\tconst blocks: SystemContentBlock[] = [{ text: sanitizeSurrogates(systemPrompt) }];\n\n\t// Add cache point for supported Claude models when caching is enabled\n\tif (cacheRetention !== \"none\" && supportsPromptCaching(model)) {\n\t\tblocks.push({\n\t\t\tcachePoint: { type: CachePointType.DEFAULT, ...(cacheRetention === \"long\" ? { ttl: CacheTTL.ONE_HOUR } : {}) },\n\t\t});\n\t}\n\n\treturn blocks;\n}\n\nfunction normalizeToolCallId(id: string): string {\n\tconst sanitized = id.replace(/[^a-zA-Z0-9_-]/g, \"_\");\n\treturn sanitized.length > 64 ? sanitized.slice(0, 64) : sanitized;\n}\n\nfunction convertMessages(\n\tcontext: Context,\n\tmodel: Model<\"bedrock-converse-stream\">,\n\tcacheRetention: CacheRetention,\n): Message[] {\n\tconst result: Message[] = [];\n\tconst transformedMessages = transformMessages(context.messages, model, normalizeToolCallId);\n\n\tfor (let i = 0; i < transformedMessages.length; i++) {\n\t\tconst m = transformedMessages[i];\n\n\t\tswitch (m.role) {\n\t\t\tcase \"user\":\n\t\t\t\tresult.push({\n\t\t\t\t\trole: ConversationRole.USER,\n\t\t\t\t\tcontent:\n\t\t\t\t\t\ttypeof m.content === \"string\"\n\t\t\t\t\t\t\t? [{ text: sanitizeSurrogates(m.content) }]\n\t\t\t\t\t\t\t: m.content.map((c) => {\n\t\t\t\t\t\t\t\t\tswitch (c.type) {\n\t\t\t\t\t\t\t\t\t\tcase \"text\":\n\t\t\t\t\t\t\t\t\t\t\treturn { text: sanitizeSurrogates(c.text) };\n\t\t\t\t\t\t\t\t\t\tcase \"image\":\n\t\t\t\t\t\t\t\t\t\t\treturn { image: createImageBlock(c.mimeType, c.data) };\n\t\t\t\t\t\t\t\t\t\tdefault:\n\t\t\t\t\t\t\t\t\t\t\tthrow new Error(\"Unknown user content type\");\n\t\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t}),\n\t\t\t\t});\n\t\t\t\tbreak;\n\t\t\tcase \"assistant\": {\n\t\t\t\t// Skip assistant messages with empty content (e.g., from aborted requests)\n\t\t\t\t// Bedrock rejects messages with empty content arrays\n\t\t\t\tif (m.content.length === 0) {\n\t\t\t\t\tcontinue;\n\t\t\t\t}\n\t\t\t\tconst contentBlocks: ContentBlock[] = [];\n\t\t\t\tfor (const c of m.content) {\n\t\t\t\t\tswitch (c.type) {\n\t\t\t\t\t\tcase \"text\":\n\t\t\t\t\t\t\t// Skip empty text blocks\n\t\t\t\t\t\t\tif (c.text.trim().length === 0) continue;\n\t\t\t\t\t\t\tcontentBlocks.push({ text: sanitizeSurrogates(c.text) });\n\t\t\t\t\t\t\tbreak;\n\t\t\t\t\t\tcase \"toolCall\":\n\t\t\t\t\t\t\tcontentBlocks.push({\n\t\t\t\t\t\t\t\ttoolUse: { toolUseId: c.id, name: c.name, input: c.arguments },\n\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\tbreak;\n\t\t\t\t\t\tcase \"thinking\":\n\t\t\t\t\t\t\t// Skip empty thinking blocks\n\t\t\t\t\t\t\tif (c.thinking.trim().length === 0) continue;\n\t\t\t\t\t\t\t// Only Anthropic models support the signature field in reasoningText.\n\t\t\t\t\t\t\t// For other models, we omit the signature to avoid errors like:\n\t\t\t\t\t\t\t// \"This model doesn't support the reasoningContent.reasoningText.signature field\"\n\t\t\t\t\t\t\tif (supportsThinkingSignature(model)) {\n\t\t\t\t\t\t\t\t// Signatures arrive after thinking deltas. If a partial or externally\n\t\t\t\t\t\t\t\t// persisted message lacks a signature, Bedrock rejects the replayed\n\t\t\t\t\t\t\t\t// reasoning block. Fall back to plain text, matching Anthropic.\n\t\t\t\t\t\t\t\tif (!c.thinkingSignature || c.thinkingSignature.trim().length === 0) {\n\t\t\t\t\t\t\t\t\tcontentBlocks.push({ text: sanitizeSurrogates(c.thinking) });\n\t\t\t\t\t\t\t\t} else {\n\t\t\t\t\t\t\t\t\tcontentBlocks.push({\n\t\t\t\t\t\t\t\t\t\treasoningContent: {\n\t\t\t\t\t\t\t\t\t\t\treasoningText: {\n\t\t\t\t\t\t\t\t\t\t\t\ttext: sanitizeSurrogates(c.thinking),\n\t\t\t\t\t\t\t\t\t\t\t\tsignature: c.thinkingSignature,\n\t\t\t\t\t\t\t\t\t\t\t},\n\t\t\t\t\t\t\t\t\t\t},\n\t\t\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t} else {\n\t\t\t\t\t\t\t\tcontentBlocks.push({\n\t\t\t\t\t\t\t\t\treasoningContent: {\n\t\t\t\t\t\t\t\t\t\treasoningText: { text: sanitizeSurrogates(c.thinking) },\n\t\t\t\t\t\t\t\t\t},\n\t\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\tbreak;\n\t\t\t\t\t\tdefault:\n\t\t\t\t\t\t\tthrow new Error(\"Unknown assistant content type\");\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\t// Skip if all content blocks were filtered out\n\t\t\t\tif (contentBlocks.length === 0) {\n\t\t\t\t\tcontinue;\n\t\t\t\t}\n\t\t\t\tresult.push({\n\t\t\t\t\trole: ConversationRole.ASSISTANT,\n\t\t\t\t\tcontent: contentBlocks,\n\t\t\t\t});\n\t\t\t\tbreak;\n\t\t\t}\n\t\t\tcase \"toolResult\": {\n\t\t\t\t// Collect all consecutive toolResult messages into a single user message\n\t\t\t\t// Bedrock requires all tool results to be in one message\n\t\t\t\tconst toolResults: ContentBlock.ToolResultMember[] = [];\n\n\t\t\t\t// Add current tool result with all content blocks combined\n\t\t\t\ttoolResults.push({\n\t\t\t\t\ttoolResult: {\n\t\t\t\t\t\ttoolUseId: m.toolCallId,\n\t\t\t\t\t\tcontent: m.content.map((c) =>\n\t\t\t\t\t\t\tc.type === \"image\"\n\t\t\t\t\t\t\t\t? { image: createImageBlock(c.mimeType, c.data) }\n\t\t\t\t\t\t\t\t: { text: sanitizeSurrogates(c.text) },\n\t\t\t\t\t\t),\n\t\t\t\t\t\tstatus: m.isError ? ToolResultStatus.ERROR : ToolResultStatus.SUCCESS,\n\t\t\t\t\t},\n\t\t\t\t});\n\n\t\t\t\t// Look ahead for consecutive toolResult messages\n\t\t\t\tlet j = i + 1;\n\t\t\t\twhile (j < transformedMessages.length && transformedMessages[j].role === \"toolResult\") {\n\t\t\t\t\tconst nextMsg = transformedMessages[j] as ToolResultMessage;\n\t\t\t\t\ttoolResults.push({\n\t\t\t\t\t\ttoolResult: {\n\t\t\t\t\t\t\ttoolUseId: nextMsg.toolCallId,\n\t\t\t\t\t\t\tcontent: nextMsg.content.map((c) =>\n\t\t\t\t\t\t\t\tc.type === \"image\"\n\t\t\t\t\t\t\t\t\t? { image: createImageBlock(c.mimeType, c.data) }\n\t\t\t\t\t\t\t\t\t: { text: sanitizeSurrogates(c.text) },\n\t\t\t\t\t\t\t),\n\t\t\t\t\t\t\tstatus: nextMsg.isError ? ToolResultStatus.ERROR : ToolResultStatus.SUCCESS,\n\t\t\t\t\t\t},\n\t\t\t\t\t});\n\t\t\t\t\tj++;\n\t\t\t\t}\n\n\t\t\t\t// Skip the messages we've already processed\n\t\t\t\ti = j - 1;\n\n\t\t\t\tresult.push({\n\t\t\t\t\trole: ConversationRole.USER,\n\t\t\t\t\tcontent: toolResults,\n\t\t\t\t});\n\t\t\t\tbreak;\n\t\t\t}\n\t\t\tdefault:\n\t\t\t\tthrow new Error(\"Unknown message role\");\n\t\t}\n\t}\n\n\t// Add cache point to the last user message for supported Claude models when caching is enabled\n\tif (cacheRetention !== \"none\" && supportsPromptCaching(model) && result.length > 0) {\n\t\tconst lastMessage = result[result.length - 1];\n\t\tif (lastMessage.role === ConversationRole.USER && lastMessage.content) {\n\t\t\t(lastMessage.content as ContentBlock[]).push({\n\t\t\t\tcachePoint: {\n\t\t\t\t\ttype: CachePointType.DEFAULT,\n\t\t\t\t\t...(cacheRetention === \"long\" ? { ttl: CacheTTL.ONE_HOUR } : {}),\n\t\t\t\t},\n\t\t\t});\n\t\t}\n\t}\n\n\treturn result;\n}\n\nfunction convertToolConfig(\n\ttools: Tool[] | undefined,\n\ttoolChoice: BedrockOptions[\"toolChoice\"],\n): ToolConfiguration | undefined {\n\tif (!tools?.length || toolChoice === \"none\") return undefined;\n\n\tconst bedrockTools: BedrockTool[] = tools.map((tool) => ({\n\t\ttoolSpec: {\n\t\t\tname: tool.name,\n\t\t\tdescription: tool.description,\n\t\t\tinputSchema: { json: tool.parameters },\n\t\t},\n\t}));\n\n\tlet bedrockToolChoice: ToolChoice | undefined;\n\tswitch (toolChoice) {\n\t\tcase \"auto\":\n\t\t\tbedrockToolChoice = { auto: {} };\n\t\t\tbreak;\n\t\tcase \"any\":\n\t\t\tbedrockToolChoice = { any: {} };\n\t\t\tbreak;\n\t\tdefault:\n\t\t\tif (toolChoice?.type === \"tool\") {\n\t\t\t\tbedrockToolChoice = { tool: { name: toolChoice.name } };\n\t\t\t}\n\t}\n\n\treturn { tools: bedrockTools, toolChoice: bedrockToolChoice };\n}\n\nfunction mapStopReason(reason: string | undefined): StopReason {\n\tswitch (reason) {\n\t\tcase BedrockStopReason.END_TURN:\n\t\tcase BedrockStopReason.STOP_SEQUENCE:\n\t\t\treturn \"stop\";\n\t\tcase BedrockStopReason.MAX_TOKENS:\n\t\tcase BedrockStopReason.MODEL_CONTEXT_WINDOW_EXCEEDED:\n\t\t\treturn \"length\";\n\t\tcase BedrockStopReason.TOOL_USE:\n\t\t\treturn \"toolUse\";\n\t\tdefault:\n\t\t\treturn \"error\";\n\t}\n}\n\nexport function buildAdditionalModelRequestFields(\n\tmodel: Model<\"bedrock-converse-stream\">,\n\toptions: BedrockOptions,\n): Record<string, any> | undefined {\n\tif (!model.reasoning) {\n\t\treturn undefined;\n\t}\n\n\tconst isClaude = model.id.includes(\"anthropic.claude\") || model.id.includes(\"anthropic/claude\");\n\n\tif (!options.reasoning) {\n\t\t// Omitting the thinking field turns thinking off for older Claude models, but the\n\t\t// Claude 5 generation thinks by default, so they need the explicit off config.\n\t\t// Opus 5.5+ cannot turn thinking off at all: send no thinking field and no effort.\n\t\tif (!isClaude || !claudeThinksByDefault(model)) return undefined;\n\t\tconst offMode = getClaudeThinkingOffMode(model);\n\t\treturn offMode === \"omit\" ? undefined : { thinking: { type: offMode } };\n\t}\n\n\tif (isClaude) {\n\t\tconst result: Record<string, any> = supportsAdaptiveThinking(model)\n\t\t\t? (() => {\n\t\t\t\t\t// Adaptive thinking: the `display` field controls whether thinking text is\n\t\t\t\t\t// returned. Opus 4.7+ default to \"omitted\"; send \"summarized\" to make it visible.\n\t\t\t\t\tconst thinking: Record<string, any> = { type: \"adaptive\" };\n\t\t\t\t\tif (options.thinkingDisplay) {\n\t\t\t\t\t\tthinking.display = options.thinkingDisplay;\n\t\t\t\t\t}\n\t\t\t\t\treturn {\n\t\t\t\t\t\tthinking,\n\t\t\t\t\t\toutput_config: { effort: getClaudeEffort(model, options.reasoning) },\n\t\t\t\t\t};\n\t\t\t\t})()\n\t\t\t: (() => {\n\t\t\t\t\tconst defaultBudgets: Record<ThinkingLevel, number> = {\n\t\t\t\t\t\tminimal: 1024,\n\t\t\t\t\t\tlow: 2048,\n\t\t\t\t\t\tmedium: 8192,\n\t\t\t\t\t\thigh: 16384,\n\t\t\t\t\t\txhigh: 16384, // Claude doesn't support xhigh, clamp to high\n\t\t\t\t\t\tmax: 16384, // Normalized max falls back through xhigh to high here\n\t\t\t\t\t};\n\n\t\t\t\t\t// Custom budgets do not define xhigh/max, so both use high.\n\t\t\t\t\tconst level = options.reasoning === \"xhigh\" || options.reasoning === \"max\" ? \"high\" : options.reasoning;\n\t\t\t\t\tconst budget = options.thinkingBudgets?.[level] ?? defaultBudgets[options.reasoning];\n\n\t\t\t\t\treturn {\n\t\t\t\t\t\tthinking: {\n\t\t\t\t\t\t\ttype: \"enabled\",\n\t\t\t\t\t\t\tbudget_tokens: budget,\n\t\t\t\t\t\t},\n\t\t\t\t\t};\n\t\t\t\t})();\n\n\t\tif (!supportsAdaptiveThinking(model) && (options.interleavedThinking ?? true)) {\n\t\t\tresult.anthropic_beta = [\"interleaved-thinking-2025-05-14\"];\n\t\t}\n\n\t\treturn result;\n\t}\n\n\treturn undefined;\n}\n\nfunction createImageBlock(mimeType: string, data: string) {\n\tlet format: ImageFormat;\n\tswitch (mimeType) {\n\t\tcase \"image/jpeg\":\n\t\tcase \"image/jpg\":\n\t\t\tformat = ImageFormat.JPEG;\n\t\t\tbreak;\n\t\tcase \"image/png\":\n\t\t\tformat = ImageFormat.PNG;\n\t\t\tbreak;\n\t\tcase \"image/gif\":\n\t\t\tformat = ImageFormat.GIF;\n\t\t\tbreak;\n\t\tcase \"image/webp\":\n\t\t\tformat = ImageFormat.WEBP;\n\t\t\tbreak;\n\t\tdefault:\n\t\t\tthrow new Error(`Unknown image type: ${mimeType}`);\n\t}\n\n\tconst binaryString = atob(data);\n\tconst bytes = new Uint8Array(binaryString.length);\n\tfor (let i = 0; i < binaryString.length; i++) {\n\t\tbytes[i] = binaryString.charCodeAt(i);\n\t}\n\n\treturn { source: { bytes }, format };\n}\n"]}

package/dist/providers/amazon-bedrock.js

@@ -1,5 +1,5 @@
 import { BedrockRuntimeClient, StopReason as BedrockStopReason, CachePointType, CacheTTL, ConversationRole, ConverseStreamCommand, ImageFormat, ToolResultStatus, } from "@aws-sdk/client-bedrock-runtime";
-import { calculateCost, supportsAdaptiveThinking, supportsXhigh } from "../models.js";
+import { calculateCost, claudeThinksByDefault, getClaudeEffort, getClaudeThinkingOffMode, supportsAdaptiveThinking, } from "../models.js";
 import { AssistantMessageEventStream } from "../utils/event-stream.js";
 import { parseStreamingJson } from "../utils/json-parse.js";
 import { sanitizeSurrogates } from "../utils/sanitize-unicode.js";
@@ -290,22 +290,6 @@ function handleContentBlockStop(event, blocks, output, stream, onWarning) {
             break;
     }
 }
-function mapThinkingLevelToEffort(level, model) {
-    switch (level) {
-        case "minimal":
-        case "low":
-            return "low";
-        case "medium":
-            return "medium";
-        case "high":
-            return "high";
-        case "xhigh":
-        case "max":
-            return supportsXhigh(model) ? "max" : "high";
-        default:
-            return "high";
-    }
-}
 /**
  * Resolve cache retention preference.
  * Defaults to "short" and uses DREB_CACHE_RETENTION for backward compatibility.
@@ -561,10 +545,20 @@ function mapStopReason(reason) {
     }
 }
 export function buildAdditionalModelRequestFields(model, options) {
-    if (!options.reasoning || !model.reasoning) {
+    if (!model.reasoning) {
         return undefined;
     }
-    if (model.id.includes("anthropic.claude") || model.id.includes("anthropic/claude")) {
+    const isClaude = model.id.includes("anthropic.claude") || model.id.includes("anthropic/claude");
+    if (!options.reasoning) {
+        // Omitting the thinking field turns thinking off for older Claude models, but the
+        // Claude 5 generation thinks by default, so they need the explicit off config.
+        // Opus 5.5+ cannot turn thinking off at all: send no thinking field and no effort.
+        if (!isClaude || !claudeThinksByDefault(model))
+            return undefined;
+        const offMode = getClaudeThinkingOffMode(model);
+        return offMode === "omit" ? undefined : { thinking: { type: offMode } };
+    }
+    if (isClaude) {
         const result = supportsAdaptiveThinking(model)
             ? (() => {
                 // Adaptive thinking: the `display` field controls whether thinking text is
@@ -575,7 +569,7 @@ export function buildAdditionalModelRequestFields(model, options) {
                 }
                 return {
                     thinking,
-                    output_config: { effort: mapThinkingLevelToEffort(options.reasoning, model) },
+                    output_config: { effort: getClaudeEffort(model, options.reasoning) },
                 };
             })()
             : (() => {