mindweave 0.0.1 → 2.1.2

package/dist/drivers/deepseek/client.js

@@ -0,0 +1,168 @@
+import { compatStreamTurn, compatToolTurn, requireApiKey } from "../openaiCompat/wire.js";
+import { clientId } from "../clientId.js";
+import { extractSearch, SEARCH_MAX_USES, SEARCH_SYSTEM } from "../searchBlocks.js";
+import { parseInlineToolCalls } from "./inlineTools.js";
+import { DEFAULT_MODEL } from "./manifest.js";
+const BASE_URL = process.env.MINDWEAVE_BASE_URL ?? "https://api.deepseek.com";
+const MODEL = process.env.MINDWEAVE_MODEL ?? DEFAULT_MODEL;
+/**
+ * Where DeepSeek serves its own web search.
+ *
+ * DeepSeek speaks two protocols on one account and one key. Chat and tools run over
+ * the OpenAI-compatible path above; the native web search is served ONLY over this
+ * Anthropic-Messages-compatible one, as a server-side tool. So this driver talks to
+ * both, and which one a request uses is decided by the OPERATION rather than by
+ * configuration — there is nothing for the user to choose and no second key.
+ *
+ * Only the search call moves. Putting the whole driver here would cost real things:
+ * this endpoint ignores `cache_control` (so the prompt cache the OpenAI path relies
+ * on would be gone) and does not carry MCP tools, images, or code execution.
+ */
+const ANTHROPIC_BASE_URL = process.env.MINDWEAVE_DEEPSEEK_ANTHROPIC_URL ?? "https://api.deepseek.com/anthropic";
+/** Output ceiling for the buffered search call. */
+const SEARCH_MAX_TOKENS = 8_000;
+/**
+ * DeepSeek's reasoning fields.
+ *
+ * The `thinking` field is ALWAYS sent, one way or the other — DeepSeek's own docs
+ * say omitting it entirely defaults to ENABLED at high effort, not disabled. So
+ * every internal call that never sets `req.model` (the session-memory summarizer,
+ * the web-page distiller) was once silently paying for full reasoning it never
+ * asked for and the UI was never going to show, since reasoning deltas are
+ * discarded unconditionally. Sending an explicit `{ type: "disabled" }` is what
+ * actually turns it off.
+ *
+ * This turned out to be the family pattern rather than a DeepSeek quirk: Qwen and
+ * GLM default to thinking-on in exactly the same way, and their drivers send the
+ * off switch explicitly for exactly this reason.
+ */
+export function reasoningFields(config) {
+    if (!config?.thinking)
+        return { thinking: { type: "disabled" } };
+    return { thinking: { type: "enabled" }, reasoning_effort: config.effort };
+}
+/**
+ * DeepSeek's documented sampling settings for agent use.
+ *
+ * Two facts from their docs, and the second is why this is a function rather than a
+ * constant spread into every body:
+ *
+ *  1. The recommended operating point is `temperature = 1.0` with `top_p = 0.95` for
+ *     agentic use (`1.0` otherwise). We were sending NEITHER, so every call ran on
+ *     whatever the endpoint's default happened to be — fine until it changes, and
+ *     invisible when it does.
+ *  2. "Thinking mode does not support the temperature, top_p, presence_penalty, or
+ *     frequency_penalty parameters. These parameters will not trigger an error but
+ *     will have no effect." So with thinking ON there is nothing to set.
+ *
+ * Which means this bites exactly where thinking is off: the Standard `/think` level,
+ * and every internal buffered call — the session summarizer, the web-page distiller —
+ * since `reasoningFields` sends `{ type: "disabled" }` for those. Those are the calls
+ * that were silently un-tuned, and they are also the ones whose output feeds straight
+ * back into the transcript.
+ *
+ * Temperature is NOT lowered to shorten output, ever, however tempting that looks for
+ * a summarizer: DeepSeek warns that lower temperatures collapse the reasoning trace
+ * and degrade the answer, and that length belongs to `max_tokens`. Written down here
+ * because it is exactly the "optimisation" a later reader would otherwise try.
+ */
+export function sampling(config) {
+    if (config?.thinking)
+        return {};
+    return { temperature: 1.0, top_p: 0.95 };
+}
+/**
+ * `insufficient_system_resource` is DeepSeek-specific, not part of the OpenAI
+ * spec: the request was cut off by DeepSeek's own infrastructure, not a token limit
+ * or a safety decision. Without a case for it, it falls through to `"end"` — a
+ * resource-starved, incomplete reply reported as a clean finish.
+ */
+export function extraStop(reason) {
+    return reason === "insufficient_system_resource" ? "overloaded" : undefined;
+}
+/**
+ * DeepSeek reports BOTH sides of the cache split directly, as siblings of
+ * `prompt_tokens`, rather than nesting a `cached_tokens` figure that the miss side
+ * has to be derived from. Read as given: deriving one from the other here would
+ * disagree with the provider whenever the two do not sum to the prompt exactly.
+ */
+export function cacheSplit(usage) {
+    const num = (key) => (typeof usage[key] === "number" ? usage[key] : 0);
+    const hit = num("prompt_cache_hit_tokens");
+    const miss = num("prompt_cache_miss_tokens");
+    if (hit === 0 && miss === 0)
+        return undefined;
+    return { hit, miss };
+}
+/**
+ * Recover tool calls DeepSeek leaked into the TEXT channel as DSML markup, and
+ * strip that markup from the content either way.
+ *
+ * The recovery only fires when the structured list is empty: a model that emitted
+ * a proper `tool_calls` array and also narrated one in prose has already had its
+ * call registered, and adding the narrated copy would run it twice.
+ */
+export function repairContent(content, toolCalls) {
+    const inline = parseInlineToolCalls(content);
+    const recovered = toolCalls.length === 0 ? inline.toolCalls : toolCalls;
+    return { content: inline.cleaned, toolCalls: recovered };
+}
+/** Everything the shared wire layer needs to talk to DeepSeek. */
+export const deepseekProvider = {
+    label: "DeepSeek",
+    baseUrl: BASE_URL,
+    apiKeyEnv: "DEEPSEEK_API_KEY",
+    defaultModel: MODEL,
+    reasoningFields,
+    sampling,
+    extraStop,
+    cacheSplit,
+    repairContent,
+};
+/** Ask the model for one turn. An empty `tools` array forces a plain-text answer. */
+export async function toolTurn(req, options = {}) {
+    return compatToolTurn(deepseekProvider, req, options.signal);
+}
+/** Ask the model for one turn, streaming deltas to `options.onEvent`. */
+export async function streamTurn(req, options = {}) {
+    return compatStreamTurn(deepseekProvider, req, options.onEvent, options.signal);
+}
+/**
+ * Search the web, over DeepSeek's own Anthropic-protocol endpoint.
+ *
+ * DeepSeek runs the search on its own servers with the key the user already has.
+ * Nothing third-party is involved and there is no second key — the same account,
+ * reached over the protocol that happens to carry its search.
+ *
+ * The SDK is imported HERE rather than at the top of the file so that a DeepSeek
+ * session that never searches does not load it. That lazy split is measured and
+ * deliberate (it is why drivers load on demand at all), and a top-level import would
+ * quietly undo it for every DeepSeek user.
+ *
+ * The model id is passed through as DeepSeek's own (`deepseek-v4-pro`), which is what
+ * their documented example does. Their Claude-name mapping is a compatibility shim
+ * for clients that only know Anthropic model names, and routing through it would mean
+ * naming a Claude model to get a DeepSeek one — indirection with nothing to gain.
+ */
+export async function webSearch(query, options = {}) {
+    const { default: Anthropic } = await import("@anthropic-ai/sdk");
+    const client = new Anthropic({
+        apiKey: requireApiKey(deepseekProvider),
+        baseURL: ANTHROPIC_BASE_URL,
+        // DeepSeek authenticates this endpoint with `x-api-key`, which is what the SDK
+        // sends for `apiKey`. It ignores `anthropic-version` rather than requiring it.
+        defaultHeaders: { "User-Agent": clientId() },
+    });
+    const message = await client.messages.create({
+        model: MODEL,
+        max_tokens: SEARCH_MAX_TOKENS,
+        system: SEARCH_SYSTEM,
+        messages: [{ role: "user", content: query }],
+        // The original tool version, not Anthropic's newer dated one: this is
+        // DeepSeek's implementation of the protocol, and the widely-implemented
+        // version is the one to send to a compatibility endpoint.
+        tools: [{ type: "web_search_20250305", name: "web_search", max_uses: SEARCH_MAX_USES }],
+    }, { signal: options.signal });
+    return extractSearch(message);
+}
+//# sourceMappingURL=client.js.map

package/dist/drivers/deepseek/client.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"client.js","sourceRoot":"","sources":["../../../src/drivers/deepseek/client.ts"],"names":[],"mappings":"AA0BA,OAAO,EAAE,gBAAgB,EAAE,cAAc,EAAE,aAAa,EAAuB,MAAM,yBAAyB,CAAC;AAC/G,OAAO,EAAE,QAAQ,EAAE,MAAM,gBAAgB,CAAC;AAC1C,OAAO,EAAE,aAAa,EAAE,eAAe,EAAE,aAAa,EAAE,MAAM,oBAAoB,CAAC;AACnF,OAAO,EAAE,oBAAoB,EAAE,MAAM,kBAAkB,CAAC;AACxD,OAAO,EAAE,aAAa,EAAE,MAAM,eAAe,CAAC;AAE9C,MAAM,QAAQ,GAAG,OAAO,CAAC,GAAG,CAAC,kBAAkB,IAAI,0BAA0B,CAAC;AAC9E,MAAM,KAAK,GAAG,OAAO,CAAC,GAAG,CAAC,eAAe,IAAI,aAAa,CAAC;AAE3D;;;;;;;;;;;;GAYG;AACH,MAAM,kBAAkB,GAAG,OAAO,CAAC,GAAG,CAAC,gCAAgC,IAAI,oCAAoC,CAAC;AAEhH,mDAAmD;AACnD,MAAM,iBAAiB,GAAG,KAAK,CAAC;AAEhC;;;;;;;;;;;;;;GAcG;AACH,MAAM,UAAU,eAAe,CAAC,MAA+B;IAC7D,IAAI,CAAC,MAAM,EAAE,QAAQ;QAAE,OAAO,EAAE,QAAQ,EAAE,EAAE,IAAI,EAAE,UAAU,EAAE,EAAE,CAAC;IACjE,OAAO,EAAE,QAAQ,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE,EAAE,gBAAgB,EAAE,MAAM,CAAC,MAAM,EAAE,CAAC;AAC5E,CAAC;AAED;;;;;;;;;;;;;;;;;;;;;;;;GAwBG;AACH,MAAM,UAAU,QAAQ,CAAC,MAA+B;IACtD,IAAI,MAAM,EAAE,QAAQ;QAAE,OAAO,EAAE,CAAC;IAChC,OAAO,EAAE,WAAW,EAAE,GAAG,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC;AAC3C,CAAC;AAED;;;;;GAKG;AACH,MAAM,UAAU,SAAS,CAAC,MAAc;IACtC,OAAO,MAAM,KAAK,8BAA8B,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,SAAS,CAAC;AAC9E,CAAC;AAED;;;;;GAKG;AACH,MAAM,UAAU,UAAU,CAAC,KAA8B;IACvD,MAAM,GAAG,GAAG,CAAC,GAAW,EAAU,EAAE,CAAC,CAAC,OAAO,KAAK,CAAC,GAAG,CAAC,KAAK,QAAQ,CAAC,CAAC,CAAE,KAAK,CAAC,GAAG,CAAY,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IACnG,MAAM,GAAG,GAAG,GAAG,CAAC,yBAAyB,CAAC,CAAC;IAC3C,MAAM,IAAI,GAAG,GAAG,CAAC,0BAA0B,CAAC,CAAC;IAC7C,IAAI,GAAG,KAAK,CAAC,IAAI,IAAI,KAAK,CAAC;QAAE,OAAO,SAAS,CAAC;IAC9C,OAAO,EAAE,GAAG,EAAE,IAAI,EAAE,CAAC;AACvB,CAAC;AAED;;;;;;;GAOG;AACH,MAAM,UAAU,aAAa,CAAC,OAAe,EAAE,SAAqB;IAClE,MAAM,MAAM,GAAG,oBAAoB,CAAC,OAAO,CAAC,CAAC;IAC7C,MAAM,SAAS,GAAG,SAAS,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC;IACxE,OAAO,EAAE,OAAO,EAAE,MAAM,CAAC,OAAO,EAAE,SAAS,EAAE,SAAS,EAAE,CAAC;AAC3D,CAAC;AAED,kEAAkE;AAClE,MAAM,CAAC,MAAM,gBAAgB,GAAmB;IAC9C,KAAK,EAAE,UAAU;IACjB,OAAO,EAAE,QAAQ;IACjB,SAAS,EAAE,kBAAkB;IAC7B,YAAY,EAAE,KAAK;IACnB,eAAe;IACf,QAAQ;IACR,SAAS;IACT,UAAU;IACV,aAAa;CACd,CAAC;AAEF,qFAAqF;AACrF,MAAM,CAAC,KAAK,UAAU,QAAQ,CAAC,GAAiB,EAAE,UAAuB,EAAE;IACzE,OAAO,cAAc,CAAC,gBAAgB,EAAE,GAAG,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC;AAC/D,CAAC;AAED,yEAAyE;AACzE,MAAM,CAAC,KAAK,UAAU,UAAU,CAAC,GAAiB,EAAE,UAAyB,EAAE;IAC7E,OAAO,gBAAgB,CAAC,gBAAgB,EAAE,GAAG,EAAE,OAAO,CAAC,OAAO,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC;AAClF,CAAC;AAED;;;;;;;;;;;;;;;;GAgBG;AACH,MAAM,CAAC,KAAK,UAAU,SAAS,CAAC,KAAa,EAAE,UAAyB,EAAE;IACxE,MAAM,EAAE,OAAO,EAAE,SAAS,EAAE,GAAG,MAAM,MAAM,CAAC,mBAAmB,CAAC,CAAC;IACjE,MAAM,MAAM,GAAG,IAAI,SAAS,CAAC;QAC3B,MAAM,EAAE,aAAa,CAAC,gBAAgB,CAAC;QACvC,OAAO,EAAE,kBAAkB;QAC3B,+EAA+E;QAC/E,+EAA+E;QAC/E,cAAc,EAAE,EAAE,YAAY,EAAE,QAAQ,EAAE,EAAE;KAC7C,CAAC,CAAC;IACH,MAAM,OAAO,GAAG,MAAM,MAAM,CAAC,QAAQ,CAAC,MAAM,CAC1C;QACE,KAAK,EAAE,KAAK;QACZ,UAAU,EAAE,iBAAiB;QAC7B,MAAM,EAAE,aAAa;QACrB,QAAQ,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,KAAK,EAAE,CAAC;QAC5C,sEAAsE;QACtE,wEAAwE;QACxE,0DAA0D;QAC1D,KAAK,EAAE,CAAC,EAAE,IAAI,EAAE,qBAAqB,EAAE,IAAI,EAAE,YAAY,EAAE,QAAQ,EAAE,eAAe,EAAE,CAAC;KACxF,EACD,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,CAC3B,CAAC;IACF,OAAO,aAAa,CAAC,OAAO,CAAC,CAAC;AAChC,CAAC"}

package/dist/drivers/deepseek/index.js

@@ -0,0 +1,16 @@
+import { streamTurn, toolTurn, webSearch } from "./client.js";
+import { stripInlineToolCalls } from "./inlineTools.js";
+import { deepseekManifest } from "./manifest.js";
+export const deepseekDriver = {
+    ...deepseekManifest,
+    toolTurn,
+    streamTurn,
+    // Served over DeepSeek's own Anthropic-protocol endpoint, with the same key. Chat
+    // stays on the OpenAI path; only this call changes protocol.
+    webSearch,
+    // The live UI renders raw text deltas, which is the one place DeepSeek's leaked
+    // DSML markup can still reach the screen — the assembled turn is already clean.
+    sanitizeText: stripInlineToolCalls,
+};
+export default deepseekDriver;
+//# sourceMappingURL=index.js.map

package/dist/drivers/deepseek/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../../src/drivers/deepseek/index.ts"],"names":[],"mappings":"AAYA,OAAO,EAAE,UAAU,EAAE,QAAQ,EAAE,SAAS,EAAE,MAAM,aAAa,CAAC;AAC9D,OAAO,EAAE,oBAAoB,EAAE,MAAM,kBAAkB,CAAC;AACxD,OAAO,EAAE,gBAAgB,EAAE,MAAM,eAAe,CAAC;AAEjD,MAAM,CAAC,MAAM,cAAc,GAAW;IACpC,GAAG,gBAAgB;IACnB,QAAQ;IACR,UAAU;IACV,kFAAkF;IAClF,6DAA6D;IAC7D,SAAS;IACT,gFAAgF;IAChF,gFAAgF;IAChF,YAAY,EAAE,oBAAoB;CACnC,CAAC;AAEF,eAAe,cAAc,CAAC"}

package/dist/drivers/deepseek/inlineTools.js

@@ -0,0 +1,46 @@
+const BLOCK_RE = /<[^<>]*DSML[^<>]*tool_calls\s*>[\s\S]*?<\/[^<>]*DSML[^<>]*tool_calls\s*>/g;
+const INVOKE_RE = /<[^<>]*DSML[^<>]*invoke[^<>]*name="([^"]+)"[^<>]*>([\s\S]*?)<\/[^<>]*DSML[^<>]*invoke[^<>]*>/g;
+const PARAM_RE = /<[^<>]*DSML[^<>]*parameter[^<>]*name="([^"]+)"([^<>]*)>([\s\S]*?)<\/[^<>]*DSML[^<>]*parameter[^<>]*>/g;
+/** True if the text contains a leaked DSML tool-call block (cheap pre-check). */
+export function hasInlineToolCalls(content) {
+    BLOCK_RE.lastIndex = 0;
+    return BLOCK_RE.test(content);
+}
+/** Strip leaked tool-call markup from text for DISPLAY (no parsing). */
+export function stripInlineToolCalls(content) {
+    return content.replace(BLOCK_RE, "").replace(/\n{3,}/g, "\n\n").trim();
+}
+/** Parse leaked tool-call blocks into real tool calls and return the cleaned text. */
+export function parseInlineToolCalls(content) {
+    const toolCalls = [];
+    let n = 0;
+    for (const block of content.match(BLOCK_RE) ?? []) {
+        INVOKE_RE.lastIndex = 0;
+        let inv;
+        while ((inv = INVOKE_RE.exec(block)) !== null) {
+            const name = inv[1];
+            const args = {};
+            PARAM_RE.lastIndex = 0;
+            let p;
+            while ((p = PARAM_RE.exec(inv[2])) !== null) {
+                const key = p[1];
+                const attrs = p[2] ?? "";
+                const raw = p[3].trim();
+                // The convention DeepSeek follows: a parameter is a string unless it's
+                // flagged `string="false"`, in which case its value is JSON.
+                args[key] = /string="false"/i.test(attrs) ? tryJson(raw) : raw;
+            }
+            toolCalls.push({ id: `inline_${Date.now()}_${n++}`, name, arguments: JSON.stringify(args) });
+        }
+    }
+    return { cleaned: stripInlineToolCalls(content), toolCalls };
+}
+function tryJson(raw) {
+    try {
+        return JSON.parse(raw);
+    }
+    catch {
+        return raw; // not valid JSON → keep the literal text
+    }
+}
+//# sourceMappingURL=inlineTools.js.map

package/dist/drivers/deepseek/inlineTools.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"inlineTools.js","sourceRoot":"","sources":["../../../src/drivers/deepseek/inlineTools.ts"],"names":[],"mappings":"AAwBA,MAAM,QAAQ,GAAG,2EAA2E,CAAC;AAC7F,MAAM,SAAS,GAAG,+FAA+F,CAAC;AAClH,MAAM,QAAQ,GAAG,uGAAuG,CAAC;AASzH,iFAAiF;AACjF,MAAM,UAAU,kBAAkB,CAAC,OAAe;IAChD,QAAQ,CAAC,SAAS,GAAG,CAAC,CAAC;IACvB,OAAO,QAAQ,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;AAChC,CAAC;AAED,wEAAwE;AACxE,MAAM,UAAU,oBAAoB,CAAC,OAAe;IAClD,OAAO,OAAO,CAAC,OAAO,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC,OAAO,CAAC,SAAS,EAAE,MAAM,CAAC,CAAC,IAAI,EAAE,CAAC;AACzE,CAAC;AAED,sFAAsF;AACtF,MAAM,UAAU,oBAAoB,CAAC,OAAe;IAClD,MAAM,SAAS,GAAe,EAAE,CAAC;IACjC,IAAI,CAAC,GAAG,CAAC,CAAC;IACV,KAAK,MAAM,KAAK,IAAI,OAAO,CAAC,KAAK,CAAC,QAAQ,CAAC,IAAI,EAAE,EAAE,CAAC;QAClD,SAAS,CAAC,SAAS,GAAG,CAAC,CAAC;QACxB,IAAI,GAA2B,CAAC;QAChC,OAAO,CAAC,GAAG,GAAG,SAAS,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,KAAK,IAAI,EAAE,CAAC;YAC9C,MAAM,IAAI,GAAG,GAAG,CAAC,CAAC,CAAE,CAAC;YACrB,MAAM,IAAI,GAA4B,EAAE,CAAC;YACzC,QAAQ,CAAC,SAAS,GAAG,CAAC,CAAC;YACvB,IAAI,CAAyB,CAAC;YAC9B,OAAO,CAAC,CAAC,GAAG,QAAQ,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAE,CAAC,CAAC,KAAK,IAAI,EAAE,CAAC;gBAC7C,MAAM,GAAG,GAAG,CAAC,CAAC,CAAC,CAAE,CAAC;gBAClB,MAAM,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;gBACzB,MAAM,GAAG,GAAG,CAAC,CAAC,CAAC,CAAE,CAAC,IAAI,EAAE,CAAC;gBACzB,uEAAuE;gBACvE,6DAA6D;gBAC7D,IAAI,CAAC,GAAG,CAAC,GAAG,iBAAiB,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC;YACjE,CAAC;YACD,SAAS,CAAC,IAAI,CAAC,EAAE,EAAE,EAAE,UAAU,IAAI,CAAC,GAAG,EAAE,IAAI,CAAC,EAAE,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QAC/F,CAAC;IACH,CAAC;IACD,OAAO,EAAE,OAAO,EAAE,oBAAoB,CAAC,OAAO,CAAC,EAAE,SAAS,EAAE,CAAC;AAC/D,CAAC;AAED,SAAS,OAAO,CAAC,GAAW;IAC1B,IAAI,CAAC;QACH,OAAO,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;IACzB,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,GAAG,CAAC,CAAC,yCAAyC;IACvD,CAAC;AACH,CAAC"}

package/dist/drivers/deepseek/manifest.js

@@ -0,0 +1,166 @@
+export const FLASH = "deepseek-v4-flash";
+export const PRO = "deepseek-v4-pro";
+/** The multimodal model, added 2026-08-21. `-exp` is DeepSeek's own suffix: they
+ *  ship it as experimental, and the id is theirs, not a label we chose. */
+export const VISION = "deepseek-v4-flash-vision-exp";
+/** The model used when nothing is saved and no env override is set. */
+export const DEFAULT_MODEL = FLASH;
+/** The models offered by `/model`. First entry is the default. */
+export const MODELS = [
+    { id: FLASH, label: "DeepSeek V4 Flash", description: "fast & cheap — the default" },
+    { id: PRO, label: "DeepSeek V4 Pro", description: "stronger, for harder work" },
+    { id: VISION, label: "DeepSeek V4 Flash Vision", description: "reads images — experimental" },
+];
+/**
+ * The effort values DeepSeek's API actually accepts for `reasoning_effort`.
+ *
+ * The shared `Effort` type is the union of every provider's ladder and includes
+ * rungs DeepSeek has never had (`medium`, `xhigh` — those are Anthropic's). Sending
+ * one is not a soft failure: it's a value the API does not recognize. So this set
+ * is the authority, `normalize` clamps to it, and a test asserts every level we
+ * advertise survives that clamp.
+ */
+const ACCEPTED_EFFORTS = new Set(["low", "high", "max"]);
+/**
+ * The reasoning levels offered by `/think`. DeepSeek V4 exposes thinking as a toggle
+ * on the same model id plus a `reasoning_effort` budget, so the whole space is:
+ *
+ *   Standard (no thinking) · High (thinking, high) · Maximum (thinking, max)
+ *
+ * Pro's Maximum sends `max`. It previously sent `xhigh`, which DeepSeek does not
+ * accept, so that level had never done anything — the rung had leaked in from the
+ * shared type when a second provider was added.
+ *
+ * FLASH HAS A MAXIMUM TIER TOO, and used to be denied one here. The fix that removed
+ * `xhigh` also scoped Maximum to Pro, on the assumption that the cheaper model had a
+ * shorter ladder. It does not: DeepSeek documents `reasoning_effort` as low/high/max
+ * for V4 Flash as well, unscoped by model. Withholding it meant the DEFAULT model —
+ * the one most sessions run — silently could not reach its top reasoning setting.
+ *
+ * The ladder is identical for both, so it is built once. What differs between Flash
+ * and Pro is the size of the model underneath, not the settings it accepts.
+ */
+export function thinkLevels(model) {
+    const standard = { label: "Standard", description: "answer directly — fastest", thinking: false, effort: "high" };
+    // The vision model is offered WITHOUT a reasoning ladder, and the omission is
+    // deliberate. DeepSeek's vision guide documents the request shape, the formats and
+    // the image budget, and says nothing at all about `reasoning_effort` on this id.
+    // This driver has already been bitten by assuming a rung exists: `xhigh` was
+    // advertised for a year, is not a value DeepSeek accepts, and so the setting had
+    // never once done anything. Advertising a level that may be rejected is worse than
+    // withholding one that turns out to work, because only the first breaks a request.
+    // Add the other two the moment the docs name them.
+    if (model === VISION)
+        return [standard];
+    return [
+        standard,
+        { label: "High", description: "think first, then answer", thinking: true, effort: "high" },
+        { label: "Maximum", description: "maximum reasoning budget", thinking: true, effort: "max" },
+    ];
+}
+// DeepSeek list prices (USD / 1M). Cache hits are ~1/10 of misses — the whole
+// reason re-sent context stays cheap. `-pro` is estimated higher; correct it if
+// needed. These are best-effort defaults a user can override without a rebuild.
+const PRICES = {
+    [FLASH]: { cacheHit: 0.014, cacheMiss: 0.14, output: 0.28 },
+    [PRO]: { cacheHit: 0.028, cacheMiss: 0.28, output: 0.56 },
+    // DeepSeek's price list gives vision exactly Flash's rates, so it is written as the
+    // same numbers rather than derived from them: if Flash's estimate is corrected one
+    // day, that is a judgment about Flash and should not silently move a second model.
+    [VISION]: { cacheHit: 0.014, cacheMiss: 0.14, output: 0.28 },
+};
+const DEFAULT_PRICE = PRICES[FLASH];
+/** Cache-aware list price for a model, falling back to Flash's for unknown ids. */
+export function price(model) {
+    return PRICES[model] ?? DEFAULT_PRICE;
+}
+/**
+ * The model's USABLE context window — where retrieval and attention stay reliable,
+ * not the raw storage cap. Both models store 1M tokens (native pretraining, not a
+ * RoPE-extended stretch), but V4 runs hybrid sparse attention (CSA at 4x KV
+ * compression alternating with HCA at 128x), and compression is where accuracy
+ * leaks at range.
+ *
+ * The number that matters for an agent is MULTI-needle retrieval, not single: a
+ * coding session recalls many scattered facts (files read, decisions made, which
+ * command failed), which is the multi-needle shape. That is also V4's weakest
+ * axis — its single-to-multi drop at 1M is the largest in the field.
+ *
+ * Published V4-Pro figures (NIAH-2 / MRCR):
+ *   200K  single 96%   multi-8 84%
+ *   256K              multi-8 ~0.82   ← still flat
+ *   1M    single 78%   multi-8 41%    ← cliff
+ *
+ * So Pro anchors at 256K: the top of the demonstrated flat region. Past it the
+ * evidence thins to a single bad endpoint, and on BYOK the user pays for every
+ * token we let the transcript grow into.
+ */
+const PRO_WINDOW = 256_000;
+/**
+ * Flash gets its own, lower value rather than inheriting Pro's curve.
+ *
+ * Flash is 284B with 13B active against Pro's 1.6T/49B, and there is NO published
+ * multi-needle data for it at any length. The one datapoint (100% single-needle
+ * NIAH at 435K) is the easy axis and says nothing about the axis we care about.
+ * 192K is a deliberate judgment call under absent data: clearly above the old
+ * shared 128K, clearly inside Pro's proven-flat region, and revisable the moment
+ * someone publishes a Flash multi-needle curve.
+ */
+const FLASH_WINDOW = 192_000;
+export function contextWindow(model) {
+    // Vision anchors to Flash. DeepSeek documents the same 1M store for all three and
+    // states its pure-text ability is on par with Flash, so the sharp-window judgment
+    // made for Flash is the one that applies; inventing a separate number for a model
+    // with no published multi-needle curve of its own would be a guess wearing a
+    // decimal point.
+    return model === PRO ? PRO_WINDOW : FLASH_WINDOW;
+}
+/**
+ * Only the vision model takes image input, and only since 2026-08-21.
+ *
+ * This used to be absent entirely, with a comment recording that as a fact rather
+ * than an oversight. It is now true of exactly one id, so the check is by id and not
+ * by provider: pointing an image at Flash or Pro still degrades before anything is
+ * sent, which is what core does with a false answer here.
+ */
+export function acceptsImages(model) {
+    return model === VISION;
+}
+/**
+ * Coerce a stored or unknown config onto a model this driver actually serves, and
+ * keep the reasoning intent valid. DeepSeek accepts three of the five shared effort
+ * rungs (`low`, `high`, `max`), so anything else clamps to `high`.
+ *
+ * No model-scoped step-down any more: both models take the same three rungs, so
+ * switching between them preserves the user's reasoning choice instead of quietly
+ * demoting it. See thinkLevels for why the old Pro-only Maximum was wrong.
+ */
+export function normalize(config) {
+    // Three ids now, so this can no longer be "PRO or else FLASH": that shape would
+    // have quietly rewritten a vision selection back to Flash, and the user would have
+    // watched their chosen model change itself with no message.
+    const model = config.model === PRO ? PRO : config.model === VISION ? VISION : FLASH;
+    // Thinking is forced off on vision for the reason thinkLevels gives none: the flag
+    // is undocumented there, and this is the gate that stops a config saved on another
+    // model carrying one in.
+    const thinking = model === VISION ? false : config.thinking === true;
+    // Anything outside DeepSeek's accepted set becomes `high`. That covers a config
+    // saved by an older build (which stored `xhigh`) and a rung belonging to another
+    // provider. `max` is accepted on BOTH models and is no longer stepped down.
+    const effort = ACCEPTED_EFFORTS.has(config.effort) ? config.effort : "high";
+    return { model, thinking, effort };
+}
+/** The cheap metadata half of this driver — see `index.ts` for the wire half. */
+export const deepseekManifest = {
+    id: "deepseek",
+    label: "DeepSeek",
+    apiKeyEnv: "DEEPSEEK_API_KEY",
+    keysUrl: "https://platform.deepseek.com/api_keys",
+    models: MODELS,
+    thinkLevels,
+    price,
+    contextWindow,
+    acceptsImages,
+    normalize,
+};
+//# sourceMappingURL=manifest.js.map

package/dist/drivers/deepseek/manifest.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"manifest.js","sourceRoot":"","sources":["../../../src/drivers/deepseek/manifest.ts"],"names":[],"mappings":"AAkBA,MAAM,CAAC,MAAM,KAAK,GAAG,mBAAmB,CAAC;AACzC,MAAM,CAAC,MAAM,GAAG,GAAG,iBAAiB,CAAC;AACrC;2EAC2E;AAC3E,MAAM,CAAC,MAAM,MAAM,GAAG,8BAA8B,CAAC;AAErD,uEAAuE;AACvE,MAAM,CAAC,MAAM,aAAa,GAAG,KAAK,CAAC;AAEnC,kEAAkE;AAClE,MAAM,CAAC,MAAM,MAAM,GAAkB;IACnC,EAAE,EAAE,EAAE,KAAK,EAAE,KAAK,EAAE,mBAAmB,EAAE,WAAW,EAAE,4BAA4B,EAAE;IACpF,EAAE,EAAE,EAAE,GAAG,EAAE,KAAK,EAAE,iBAAiB,EAAE,WAAW,EAAE,2BAA2B,EAAE;IAC/E,EAAE,EAAE,EAAE,MAAM,EAAE,KAAK,EAAE,0BAA0B,EAAE,WAAW,EAAE,6BAA6B,EAAE;CAC9F,CAAC;AAEF;;;;;;;;GAQG;AACH,MAAM,gBAAgB,GAAG,IAAI,GAAG,CAAS,CAAC,KAAK,EAAE,MAAM,EAAE,KAAK,CAAC,CAAC,CAAC;AAEjE;;;;;;;;;;;;;;;;;;GAkBG;AACH,MAAM,UAAU,WAAW,CAAC,KAAc;IACxC,MAAM,QAAQ,GAAe,EAAE,KAAK,EAAE,UAAU,EAAE,WAAW,EAAE,2BAA2B,EAAE,QAAQ,EAAE,KAAK,EAAE,MAAM,EAAE,MAAM,EAAE,CAAC;IAC9H,8EAA8E;IAC9E,mFAAmF;IACnF,iFAAiF;IACjF,6EAA6E;IAC7E,iFAAiF;IACjF,mFAAmF;IACnF,mFAAmF;IACnF,mDAAmD;IACnD,IAAI,KAAK,KAAK,MAAM;QAAE,OAAO,CAAC,QAAQ,CAAC,CAAC;IACxC,OAAO;QACL,QAAQ;QACR,EAAE,KAAK,EAAE,MAAM,EAAE,WAAW,EAAE,0BAA0B,EAAE,QAAQ,EAAE,IAAI,EAAE,MAAM,EAAE,MAAM,EAAE;QAC1F,EAAE,KAAK,EAAE,SAAS,EAAE,WAAW,EAAE,0BAA0B,EAAE,QAAQ,EAAE,IAAI,EAAE,MAAM,EAAE,KAAK,EAAE;KAC7F,CAAC;AACJ,CAAC;AAED,8EAA8E;AAC9E,gFAAgF;AAChF,gFAAgF;AAChF,MAAM,MAAM,GAA+B;IACzC,CAAC,KAAK,CAAC,EAAE,EAAE,QAAQ,EAAE,KAAK,EAAE,SAAS,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE;IAC3D,CAAC,GAAG,CAAC,EAAE,EAAE,QAAQ,EAAE,KAAK,EAAE,SAAS,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE;IACzD,oFAAoF;IACpF,mFAAmF;IACnF,mFAAmF;IACnF,CAAC,MAAM,CAAC,EAAE,EAAE,QAAQ,EAAE,KAAK,EAAE,SAAS,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE;CAC7D,CAAC;AACF,MAAM,aAAa,GAAe,MAAM,CAAC,KAAK,CAAE,CAAC;AAEjD,mFAAmF;AACnF,MAAM,UAAU,KAAK,CAAC,KAAc;IAClC,OAAO,MAAM,CAAC,KAAK,CAAC,IAAI,aAAa,CAAC;AACxC,CAAC;AAED;;;;;;;;;;;;;;;;;;;;GAoBG;AACH,MAAM,UAAU,GAAG,OAAO,CAAC;AAE3B;;;;;;;;;GASG;AACH,MAAM,YAAY,GAAG,OAAO,CAAC;AAE7B,MAAM,UAAU,aAAa,CAAC,KAAc;IAC1C,kFAAkF;IAClF,kFAAkF;IAClF,kFAAkF;IAClF,6EAA6E;IAC7E,iBAAiB;IACjB,OAAO,KAAK,KAAK,GAAG,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,YAAY,CAAC;AACnD,CAAC;AAED;;;;;;;GAOG;AACH,MAAM,UAAU,aAAa,CAAC,KAAc;IAC1C,OAAO,KAAK,KAAK,MAAM,CAAC;AAC1B,CAAC;AAED;;;;;;;;GAQG;AACH,MAAM,UAAU,SAAS,CAAC,MAAmB;IAC3C,gFAAgF;IAChF,mFAAmF;IACnF,4DAA4D;IAC5D,MAAM,KAAK,GAAY,MAAM,CAAC,KAAK,KAAK,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,KAAK,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,KAAK,CAAC;IAC7F,mFAAmF;IACnF,mFAAmF;IACnF,yBAAyB;IACzB,MAAM,QAAQ,GAAG,KAAK,KAAK,MAAM,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,QAAQ,KAAK,IAAI,CAAC;IACrE,gFAAgF;IAChF,iFAAiF;IACjF,4EAA4E;IAC5E,MAAM,MAAM,GAAW,gBAAgB,CAAC,GAAG,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC;IACpF,OAAO,EAAE,KAAK,EAAE,QAAQ,EAAE,MAAM,EAAE,CAAC;AACrC,CAAC;AAED,iFAAiF;AACjF,MAAM,CAAC,MAAM,gBAAgB,GAAmB;IAC9C,EAAE,EAAE,UAAU;IACd,KAAK,EAAE,UAAU;IACjB,SAAS,EAAE,kBAAkB;IAC7B,OAAO,EAAE,wCAAwC;IACjD,MAAM,EAAE,MAAM;IACd,WAAW;IACX,KAAK;IACL,aAAa;IACb,aAAa;IACb,SAAS;CACV,CAAC"}

package/dist/drivers/gemini/client.js

@@ -0,0 +1,38 @@
+import { compatStreamTurn, compatToolTurn, standardCacheSplit } from "../openaiCompat/wire.js";
+import { BUFFERED_OUTPUT_TOKENS, DEFAULT_MODEL } from "./manifest.js";
+const BASE_URL = process.env.MINDWEAVE_GEMINI_URL ?? "https://generativelanguage.googleapis.com/v1beta/openai";
+/**
+ * Gemini's reasoning field.
+ *
+ * Google documents its OpenAI-compatible endpoint translating the standard
+ * `reasoning_effort` parameter onto its own `thinking_level` internally, so this
+ * is a bare pass-through rather than a renamed field like Qwen's or xAI's. Every
+ * model in this lineup always reasons (see the manifest header), so unlike xAI or
+ * DeepSeek there is no "send nothing, thinking is off" branch to guard here — an
+ * effort is always sent.
+ */
+export function reasoningFields(config) {
+    return { reasoning_effort: config?.effort ?? "low" };
+}
+/** Gemini reports its cache hit the standard OpenAI-compatible way, the same
+ *  shape Qwen, GLM, xAI, Mistral, Groq and Cerebras all return. */
+export const cacheSplit = standardCacheSplit;
+/** Everything the shared wire layer needs to talk to Gemini. */
+export const geminiProvider = {
+    label: "Gemini",
+    baseUrl: BASE_URL,
+    apiKeyEnv: "GEMINI_API_KEY",
+    defaultModel: process.env.MINDWEAVE_MODEL ?? DEFAULT_MODEL,
+    reasoningFields,
+    cacheSplit,
+    bufferedMaxTokens: BUFFERED_OUTPUT_TOKENS,
+};
+/** Ask the model for one turn. */
+export async function toolTurn(req, options = {}) {
+    return compatToolTurn(geminiProvider, req, options.signal);
+}
+/** Ask the model for one turn, streaming deltas to `options.onEvent`. */
+export async function streamTurn(req, options = {}) {
+    return compatStreamTurn(geminiProvider, req, options.onEvent, options.signal);
+}
+//# sourceMappingURL=client.js.map

package/dist/drivers/gemini/client.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"client.js","sourceRoot":"","sources":["../../../src/drivers/gemini/client.ts"],"names":[],"mappings":"AAUA,OAAO,EAAE,gBAAgB,EAAE,cAAc,EAAE,kBAAkB,EAAuB,MAAM,yBAAyB,CAAC;AACpH,OAAO,EAAE,sBAAsB,EAAE,aAAa,EAAE,MAAM,eAAe,CAAC;AAEtE,MAAM,QAAQ,GAAG,OAAO,CAAC,GAAG,CAAC,oBAAoB,IAAI,yDAAyD,CAAC;AAE/G;;;;;;;;;GASG;AACH,MAAM,UAAU,eAAe,CAAC,MAA+B;IAC7D,OAAO,EAAE,gBAAgB,EAAE,MAAM,EAAE,MAAM,IAAI,KAAK,EAAE,CAAC;AACvD,CAAC;AAED;mEACmE;AACnE,MAAM,CAAC,MAAM,UAAU,GAAG,kBAAkB,CAAC;AAE7C,gEAAgE;AAChE,MAAM,CAAC,MAAM,cAAc,GAAmB;IAC5C,KAAK,EAAE,QAAQ;IACf,OAAO,EAAE,QAAQ;IACjB,SAAS,EAAE,gBAAgB;IAC3B,YAAY,EAAE,OAAO,CAAC,GAAG,CAAC,eAAe,IAAI,aAAa;IAC1D,eAAe;IACf,UAAU;IACV,iBAAiB,EAAE,sBAAsB;CAC1C,CAAC;AAEF,kCAAkC;AAClC,MAAM,CAAC,KAAK,UAAU,QAAQ,CAAC,GAAiB,EAAE,UAAuB,EAAE;IACzE,OAAO,cAAc,CAAC,cAAc,EAAE,GAAG,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC;AAC7D,CAAC;AAED,yEAAyE;AACzE,MAAM,CAAC,KAAK,UAAU,UAAU,CAAC,GAAiB,EAAE,UAAyB,EAAE;IAC7E,OAAO,gBAAgB,CAAC,cAAc,EAAE,GAAG,EAAE,OAAO,CAAC,OAAO,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC;AAChF,CAAC"}

package/dist/drivers/gemini/index.js

@@ -0,0 +1,5 @@
+import { streamTurn, toolTurn } from "./client.js";
+import { geminiManifest } from "./manifest.js";
+export const geminiDriver = { ...geminiManifest, toolTurn, streamTurn };
+export default geminiDriver;
+//# sourceMappingURL=index.js.map

package/dist/drivers/gemini/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../../src/drivers/gemini/index.ts"],"names":[],"mappings":"AAWA,OAAO,EAAE,UAAU,EAAE,QAAQ,EAAE,MAAM,aAAa,CAAC;AACnD,OAAO,EAAE,cAAc,EAAE,MAAM,eAAe,CAAC;AAE/C,MAAM,CAAC,MAAM,YAAY,GAAW,EAAE,GAAG,cAAc,EAAE,QAAQ,EAAE,UAAU,EAAE,CAAC;AAEhF,eAAe,YAAY,CAAC"}

package/dist/drivers/gemini/manifest.js

@@ -0,0 +1,150 @@
+export const FLASH_37 = "gemini-3.7-flash";
+export const PRO_31 = "gemini-3.1-pro-preview";
+export const FLASH_LITE_35 = "gemini-3.5-flash-lite";
+export const FLASH_36 = "gemini-3.6-flash";
+export const FLASH_35 = "gemini-3.5-flash";
+export const FLASH_LITE_31 = "gemini-3.1-flash-lite";
+/** The model used when nothing is saved and no env override is set. */
+export const DEFAULT_MODEL = FLASH_37;
+/** The models offered by `/model`. First entry is this provider's default. */
+export const MODELS = [
+    { id: FLASH_37, label: "Gemini 3.7 Flash", description: "the current flash model, built for coding — the default" },
+    { id: PRO_31, label: "Gemini 3.1 Pro", description: "the deepest reasoning tier, for the hardest problems" },
+    { id: FLASH_LITE_35, label: "Gemini 3.5 Flash-Lite", description: "cheap and quick, for high-volume work" },
+    { id: FLASH_36, label: "Gemini 3.6 Flash", description: "the previous flash generation" },
+    { id: FLASH_35, label: "Gemini 3.5 Flash", description: "older flash, stronger than its Lite sibling" },
+    { id: FLASH_LITE_31, label: "Gemini 3.1 Flash-Lite", description: "the cheapest model here" },
+];
+/**
+ * The reasoning levels offered by `/think`.
+ *
+ * No off switch — see the file header. All three rungs are real settings a Gemini
+ * 3 model will run at, not a stub standing in for one that does not exist.
+ */
+export function thinkLevels(_model) {
+    return [
+        { label: "Standard", description: "lighter reasoning budget — this model always reasons", thinking: true, effort: "low" },
+        { label: "Thinking", description: "more reasoning before answering", thinking: true, effort: "medium" },
+        { label: "Maximum", description: "maximum reasoning budget", thinking: true, effort: "high" },
+    ];
+}
+/**
+ * List prices (USD / 1M tokens), the ≤200K-token tier where one applies.
+ *
+ * Two of these carry a promotional rate that Google has stated ends 2026-12-31,
+ * after which Flash's input/output both double. Not modelled here — a rate that
+ * changes on a calendar date belongs in a review before that date, not a branch in
+ * this function guessing today's date against it.
+ */
+const PRICES = {
+    [FLASH_37]: { cacheHit: 0.075, cacheMiss: 0.75, output: 3.75 },
+    [PRO_31]: { cacheHit: 0.2, cacheMiss: 2, output: 12 },
+    [FLASH_LITE_35]: { cacheHit: 0.03, cacheMiss: 0.3, output: 2.5 },
+    // 3.6 Flash carries the same promotional rate as 3.7 (both double on 2027-01-01);
+    // see the note above on why that date is not branched on here.
+    [FLASH_36]: { cacheHit: 0.075, cacheMiss: 0.75, output: 3.75 },
+    // 3.5 Flash is the odd one: it costs MORE than either newer Flash above, because
+    // those two are promotionally discounted and it is not. Listed as published rather
+    // than reordered to look tidy — the cost line has to match the bill, not intuition.
+    [FLASH_35]: { cacheHit: 0.15, cacheMiss: 1.5, output: 9 },
+    [FLASH_LITE_31]: { cacheHit: 0.025, cacheMiss: 0.25, output: 1.5 },
+};
+/**
+ * Whether this model has its OWN entry in the table above.
+ *
+ * Exists because `price()` cannot answer it: a model with no entry silently returns the
+ * default's rate, and when two models genuinely share a rate (3.7 and 3.6 Flash carry
+ * the same promotional price) comparing values cannot tell a real entry from a
+ * fallthrough. A model added to MODELS but forgotten in PRICES would then be costed as
+ * something else, wrongly and invisibly, in a figure shown to the user as money.
+ */
+export function hasListedPrice(model) {
+    return Object.hasOwn(PRICES, model);
+}
+/** Cache-aware list price for a model, falling back to the default model's. */
+export function price(model) {
+    return PRICES[model] ?? PRICES[DEFAULT_MODEL];
+}
+/**
+ * The model's USABLE context window.
+ *
+ * Gemini 3.1 Pro is billed in two tiers — the price table above is the ≤200K-token
+ * rate, and crossing that line DOUBLES the rate on the whole request, the same
+ * shape xAI's Grok 4.3 has. 128K keeps ordinary sessions clear of it, for the same
+ * reason and the same number as that driver.
+ *
+ * The two Flash models carry no such cliff; 200K is the ordinary ceiling this
+ * codebase uses elsewhere once a window is chosen for cost rather than storage —
+ * both models store far more, but the point of a usable window is what the user's
+ * money should still be paying to carry, not what the provider will hold.
+ */
+export function contextWindow(model) {
+    return model === PRO_31 ? 128_000 : 200_000;
+}
+/**
+ * The ceiling this driver puts on a single BUFFERED (non-streaming) call. Core
+ * reserves exactly this much room below the window, and `client.ts` sends the same
+ * constant, so the request and the reservation cannot drift apart.
+ */
+export const BUFFERED_OUTPUT_TOKENS = 8_000;
+export function bufferedOutputTokens(_model) {
+    return BUFFERED_OUTPUT_TOKENS;
+}
+/**
+ * Vision is deliberately NOT declared, even though Gemini itself reads images
+ * fine. The shared OpenAI-compatible wire layer (`../openaiCompat/wire.ts`) never
+ * renders `ChatMessage.images` into the request body at all — only the two
+ * SDK-backed drivers (Anthropic, OpenAI) do that today. Declaring it here would
+ * mean core attaching bytes this path silently never sends. Add it once the shared
+ * layer carries image parts.
+ */
+/** The effort rungs this provider accepts, weakest first — note the absent top two. */
+const EFFORTS = ["low", "medium", "high"];
+/**
+ * Coerce a stored or unknown config onto a model this provider actually serves.
+ *
+ * Thinking is forced true unconditionally — there is no model in this lineup for
+ * which it could legally be false — and the effort then snaps onto a rung
+ * `/think` actually lists, which is what keeps `xhigh`/`max` from another
+ * provider's config from ever reaching the wire.
+ */
+export function normalize(config) {
+    const model = PRICES[config.model] ? config.model : DEFAULT_MODEL;
+    const effort = EFFORTS.includes(config.effort) ? config.effort : "low";
+    return { model, thinking: true, effort: snapToOfferedRung(model, effort) };
+}
+/**
+ * Move an effort onto the nearest rung this model's `/think` ladder offers. Ties
+ * break DOWNWARD: an unlisted setting resolves to the cheaper neighbour, because
+ * silently spending more of the user's money is the worse way to be wrong.
+ */
+function snapToOfferedRung(model, effort) {
+    const ladder = ["low", "medium", "high", "xhigh", "max"];
+    const offered = thinkLevels(model);
+    if (offered.some((l) => l.effort === effort))
+        return effort;
+    const want = ladder.indexOf(effort);
+    let best = offered[0];
+    for (const level of offered) {
+        const d = Math.abs(ladder.indexOf(level.effort) - want);
+        const bestD = Math.abs(ladder.indexOf(best.effort) - want);
+        if (d < bestD || (d === bestD && ladder.indexOf(level.effort) < ladder.indexOf(best.effort))) {
+            best = level;
+        }
+    }
+    return best.effort;
+}
+/** The cheap metadata half of this driver — see `index.ts` for the wire half. */
+export const geminiManifest = {
+    id: "gemini",
+    label: "Gemini",
+    apiKeyEnv: "GEMINI_API_KEY",
+    keysUrl: "https://aistudio.google.com/apikey",
+    models: MODELS,
+    thinkLevels,
+    price,
+    contextWindow,
+    bufferedOutputTokens,
+    normalize,
+};
+//# sourceMappingURL=manifest.js.map

package/dist/drivers/gemini/manifest.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"manifest.js","sourceRoot":"","sources":["../../../src/drivers/gemini/manifest.ts"],"names":[],"mappings":"AA4BA,MAAM,CAAC,MAAM,QAAQ,GAAG,kBAAkB,CAAC;AAC3C,MAAM,CAAC,MAAM,MAAM,GAAG,wBAAwB,CAAC;AAC/C,MAAM,CAAC,MAAM,aAAa,GAAG,uBAAuB,CAAC;AACrD,MAAM,CAAC,MAAM,QAAQ,GAAG,kBAAkB,CAAC;AAC3C,MAAM,CAAC,MAAM,QAAQ,GAAG,kBAAkB,CAAC;AAC3C,MAAM,CAAC,MAAM,aAAa,GAAG,uBAAuB,CAAC;AAErD,uEAAuE;AACvE,MAAM,CAAC,MAAM,aAAa,GAAG,QAAQ,CAAC;AAEtC,8EAA8E;AAC9E,MAAM,CAAC,MAAM,MAAM,GAAkB;IACnC,EAAE,EAAE,EAAE,QAAQ,EAAE,KAAK,EAAE,kBAAkB,EAAE,WAAW,EAAE,yDAAyD,EAAE;IACnH,EAAE,EAAE,EAAE,MAAM,EAAE,KAAK,EAAE,gBAAgB,EAAE,WAAW,EAAE,sDAAsD,EAAE;IAC5G,EAAE,EAAE,EAAE,aAAa,EAAE,KAAK,EAAE,uBAAuB,EAAE,WAAW,EAAE,uCAAuC,EAAE;IAC3G,EAAE,EAAE,EAAE,QAAQ,EAAE,KAAK,EAAE,kBAAkB,EAAE,WAAW,EAAE,+BAA+B,EAAE;IACzF,EAAE,EAAE,EAAE,QAAQ,EAAE,KAAK,EAAE,kBAAkB,EAAE,WAAW,EAAE,6CAA6C,EAAE;IACvG,EAAE,EAAE,EAAE,aAAa,EAAE,KAAK,EAAE,uBAAuB,EAAE,WAAW,EAAE,yBAAyB,EAAE;CAC9F,CAAC;AAEF;;;;;GAKG;AACH,MAAM,UAAU,WAAW,CAAC,MAAe;IACzC,OAAO;QACL,EAAE,KAAK,EAAE,UAAU,EAAE,WAAW,EAAE,sDAAsD,EAAE,QAAQ,EAAE,IAAI,EAAE,MAAM,EAAE,KAAK,EAAE;QACzH,EAAE,KAAK,EAAE,UAAU,EAAE,WAAW,EAAE,iCAAiC,EAAE,QAAQ,EAAE,IAAI,EAAE,MAAM,EAAE,QAAQ,EAAE;QACvG,EAAE,KAAK,EAAE,SAAS,EAAE,WAAW,EAAE,0BAA0B,EAAE,QAAQ,EAAE,IAAI,EAAE,MAAM,EAAE,MAAM,EAAE;KAC9F,CAAC;AACJ,CAAC;AAED;;;;;;;GAOG;AACH,MAAM,MAAM,GAA+B;IACzC,CAAC,QAAQ,CAAC,EAAE,EAAE,QAAQ,EAAE,KAAK,EAAE,SAAS,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE;IAC9D,CAAC,MAAM,CAAC,EAAE,EAAE,QAAQ,EAAE,GAAG,EAAE,SAAS,EAAE,CAAC,EAAE,MAAM,EAAE,EAAE,EAAE;IACrD,CAAC,aAAa,CAAC,EAAE,EAAE,QAAQ,EAAE,IAAI,EAAE,SAAS,EAAE,GAAG,EAAE,MAAM,EAAE,GAAG,EAAE;IAChE,kFAAkF;IAClF,+DAA+D;IAC/D,CAAC,QAAQ,CAAC,EAAE,EAAE,QAAQ,EAAE,KAAK,EAAE,SAAS,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE;IAC9D,iFAAiF;IACjF,mFAAmF;IACnF,oFAAoF;IACpF,CAAC,QAAQ,CAAC,EAAE,EAAE,QAAQ,EAAE,IAAI,EAAE,SAAS,EAAE,GAAG,EAAE,MAAM,EAAE,CAAC,EAAE;IACzD,CAAC,aAAa,CAAC,EAAE,EAAE,QAAQ,EAAE,KAAK,EAAE,SAAS,EAAE,IAAI,EAAE,MAAM,EAAE,GAAG,EAAE;CACnE,CAAC;AAEF;;;;;;;;GAQG;AACH,MAAM,UAAU,cAAc,CAAC,KAAc;IAC3C,OAAO,MAAM,CAAC,MAAM,CAAC,MAAM,EAAE,KAAK,CAAC,CAAC;AACtC,CAAC;AAED,+EAA+E;AAC/E,MAAM,UAAU,KAAK,CAAC,KAAc;IAClC,OAAO,MAAM,CAAC,KAAK,CAAC,IAAI,MAAM,CAAC,aAAa,CAAE,CAAC;AACjD,CAAC;AAED;;;;;;;;;;;;GAYG;AACH,MAAM,UAAU,aAAa,CAAC,KAAc;IAC1C,OAAO,KAAK,KAAK,MAAM,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,OAAO,CAAC;AAC9C,CAAC;AAED;;;;GAIG;AACH,MAAM,CAAC,MAAM,sBAAsB,GAAG,KAAK,CAAC;AAE5C,MAAM,UAAU,oBAAoB,CAAC,MAAe;IAClD,OAAO,sBAAsB,CAAC;AAChC,CAAC;AAED;;;;;;;GAOG;AAEH,uFAAuF;AACvF,MAAM,OAAO,GAAa,CAAC,KAAK,EAAE,QAAQ,EAAE,MAAM,CAAC,CAAC;AAEpD;;;;;;;GAOG;AACH,MAAM,UAAU,SAAS,CAAC,MAAmB;IAC3C,MAAM,KAAK,GAAY,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,aAAa,CAAC;IAC3E,MAAM,MAAM,GAAW,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,KAAK,CAAC;IAC/E,OAAO,EAAE,KAAK,EAAE,QAAQ,EAAE,IAAI,EAAE,MAAM,EAAE,iBAAiB,CAAC,KAAK,EAAE,MAAM,CAAC,EAAE,CAAC;AAC7E,CAAC;AAED;;;;GAIG;AACH,SAAS,iBAAiB,CAAC,KAAc,EAAE,MAAc;IACvD,MAAM,MAAM,GAAa,CAAC,KAAK,EAAE,QAAQ,EAAE,MAAM,EAAE,OAAO,EAAE,KAAK,CAAC,CAAC;IACnE,MAAM,OAAO,GAAG,WAAW,CAAC,KAAK,CAAC,CAAC;IACnC,IAAI,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,MAAM,CAAC;QAAE,OAAO,MAAM,CAAC;IAE5D,MAAM,IAAI,GAAG,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;IACpC,IAAI,IAAI,GAAG,OAAO,CAAC,CAAC,CAAE,CAAC;IACvB,KAAK,MAAM,KAAK,IAAI,OAAO,EAAE,CAAC;QAC5B,MAAM,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,MAAM,CAAC,OAAO,CAAC,KAAK,CAAC,MAAM,CAAC,GAAG,IAAI,CAAC,CAAC;QACxD,MAAM,KAAK,GAAG,IAAI,CAAC,GAAG,CAAC,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,GAAG,IAAI,CAAC,CAAC;QAC3D,IAAI,CAAC,GAAG,KAAK,IAAI,CAAC,CAAC,KAAK,KAAK,IAAI,MAAM,CAAC,OAAO,CAAC,KAAK,CAAC,MAAM,CAAC,GAAG,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,EAAE,CAAC;YAC7F,IAAI,GAAG,KAAK,CAAC;QACf,CAAC;IACH,CAAC;IACD,OAAO,IAAI,CAAC,MAAM,CAAC;AACrB,CAAC;AAED,iFAAiF;AACjF,MAAM,CAAC,MAAM,cAAc,GAAmB;IAC5C,EAAE,EAAE,QAAQ;IACZ,KAAK,EAAE,QAAQ;IACf,SAAS,EAAE,gBAAgB;IAC3B,OAAO,EAAE,oCAAoC;IAC7C,MAAM,EAAE,MAAM;IACd,WAAW;IACX,KAAK;IACL,aAAa;IACb,oBAAoB;IACpB,SAAS;CACV,CAAC"}