@gaunt-sloth/core 2.0.0-alpha.8 → 2.0.0-beta.0

package/dist/providers/ollama.js

@@ -2,59 +2,89 @@ import { displayWarning } from '#src/utils/consoleUtils.js';
 import { env } from '#src/utils/systemUtils.js';
 import { writeConfigFileWithMessages } from '#src/utils/fileUtils.js';
 import { buildInitConfigContent, getCuratedFallbackModel } from '#src/providers/modelDiscovery.js';
+import { NATIVE_CLIENT_REASON, warnUnusedConfiguration, } from '#src/providers/configurationPassthrough.js';
 /**
- * Default Ollama daemon host, matching the Ollama CLI/library default. The
- * OpenAI-compatible surface lives under `/v1` on this host. Kept in sync with
- * `DEFAULT_OLLAMA_HOST` in `modelDiscovery.ts`.
+ * Default Ollama daemon host, matching the Ollama CLI/library default. GS2-59 — `ChatOllama` talks
+ * to the daemon's NATIVE endpoint (`/api/chat`), NOT the OpenAI-compatible `/v1` shim this provider
+ * used before. The same root host is reused by `modelDiscovery.ts`'s model picker, which still
+ * queries the `/v1/models` surface (discovery only); both derive from this root — they stay in sync
+ * at the host level even though the two use different endpoints under it.
  */
 const DEFAULT_OLLAMA_HOST = 'http://127.0.0.1:11434';
 /**
- * Ollama serves an unauthenticated local daemon, but `ChatOpenAI` requires a
- * non-empty `apiKey` string. Send a harmless placeholder so the client builds;
- * the local daemon ignores it.
+ * GS2-59 — default context window (`num_ctx`) for Ollama models. Ollama's OWN default is 4096, but
+ * gaunt-sloth's agentic prompt (system + full lean toolset + a tool result) already lands ~4000
+ * tokens; at 4096 a thinking model (e.g. gemma4:31b) spends its entire remaining budget on the
+ * reasoning field and emits EMPTY `content` on the turn after a tool executes — the GS2-59
+ * blank-answer regression. The previous `ChatOpenAI`→`/v1` path could not fix this because the
+ * OpenAI-compat shim IGNORES `num_ctx`; the native `/api/chat` path honors it.
+ *
+ * 16384 is chosen as the largest window that is BOTH safely above the ~4000-token starvation point
+ * (4× headroom for reasoning + a few tool results) AND fits constrained consumer VRAM: Ollama
+ * preallocates the KV cache at `num_ctx`, so on a box where a large model already spills partly to
+ * CPU (e.g. a 19GB model on ~18GB of GPU), a 32768 cache tips the GPU allocation into an
+ * out-of-memory error. 16384 was verified live to run the agentic tool→synthesis turn on such a
+ * box; 32768 OOM'd it. Overridable per config via `llm.numCtx` — raise it if you have the VRAM and
+ * run long sessions, lower it on very tight hardware. NOTE: a per-request `num_ctx` overrides the
+ * daemon's `OLLAMA_CONTEXT_LENGTH`, so a user who tuned their server window higher should set
+ * `llm.numCtx` to match rather than rely on the server default.
  */
-const OLLAMA_PLACEHOLDER_API_KEY = 'ollama';
+const DEFAULT_OLLAMA_NUM_CTX = 16384;
 /**
- * Resolve the OpenAI-compatible base URL for the local Ollama daemon.
- *
- * Honors the `OLLAMA_HOST` env override (the same variable the Ollama CLI uses).
- * `OLLAMA_HOST` is typically a full URL (`http://127.0.0.1:11434`) or a bare
- * `host:port`; either form is normalized to a `http(s)://host[:port]/v1` base.
+ * Resolve the base URL for the local Ollama daemon — the NATIVE root (no `/v1` suffix; `ChatOllama`
+ * appends its own `/api/chat` path). Honors the `OLLAMA_HOST` env override (the same variable the
+ * Ollama CLI uses): a full URL (`http://127.0.0.1:11434`) or a bare `host:port` is normalized to
+ * `http(s)://host[:port]` with any trailing slash stripped.
  */
 function resolveBaseUrl() {
     const host = env.OLLAMA_HOST;
-    let base;
-    if (!host) {
-        base = DEFAULT_OLLAMA_HOST;
-    }
-    else if (/^https?:\/\//.test(host)) {
-        base = host.replace(/\/+$/, '');
-    }
-    else {
-        base = `http://${host}`.replace(/\/+$/, '');
-    }
-    return `${base}/v1`;
+    if (!host)
+        return DEFAULT_OLLAMA_HOST;
+    const base = /^https?:\/\//.test(host) ? host : `http://${host}`;
+    return base.replace(/\/+$/, '');
 }
-// Function to process JSON config and create an Ollama (OpenAI-compatible) LLM instance
+/**
+ * GS2-59 — build a native `ChatOllama` client (`@langchain/ollama`, talking to Ollama's
+ * `/api/chat`). This replaces the previous `ChatOpenAI`→`/v1` client, which had two fatal
+ * properties for local thinking models: it DROPPED the model's separate reasoning field (the
+ * completions converter discards it — `__includeRawResponse` is openrouter-only), and it IGNORED
+ * `num_ctx` so a large agentic prompt starved the answer to empty content. `ChatOllama` fixes both:
+ * it honors `numCtx`, and it surfaces the model's thinking in `additional_kwargs.reasoning_content`
+ * — the exact field the reasoning pipeline (`pickReasoningDelta`) already reads — so the `/reasoning`
+ * panel populates with no downstream change.
+ *
+ * Ollama is a local, unauthenticated daemon, so the old OpenAI-client knobs are gone (2.0 migration
+ * note): point elsewhere with `OLLAMA_HOST`; the previous `configuration.baseURL` / placeholder
+ * `apiKey` fields no longer apply. If a reverse proxy in front of Ollama needs auth, set `headers`
+ * in the config (passed straight through to `ChatOllama`).
+ */
 // noinspection JSUnusedGlobalSymbols
 export async function processJsonConfig(llmConfig) {
-    const { ChatOpenAI } = await import('@langchain/openai');
-    // Ollama is local and unauthenticated; ChatOpenAI still needs a non-empty key.
-    const apiKey = llmConfig.apiKey || OLLAMA_PLACEHOLDER_API_KEY;
+    const { ChatOllama } = await import('@langchain/ollama');
     const configFields = {
         ...llmConfig,
-        apiKey,
         model: llmConfig.model || getCuratedFallbackModel('ollama'),
-        configuration: {
-            baseURL: resolveBaseUrl(),
-            ...(llmConfig.configuration || {}),
-        },
+        // Precedence: an explicit config `baseUrl` (the native ChatOllama field) wins, else the
+        // `OLLAMA_HOST` env, else the local default. Config is the more specific/intentional signal.
+        baseUrl: llmConfig.baseUrl ?? resolveBaseUrl(),
+        numCtx: llmConfig.numCtx ?? DEFAULT_OLLAMA_NUM_CTX,
     };
-    // eslint-disable-next-line @typescript-eslint/no-explicit-any
-    delete configFields.type;
-    // eslint-disable-next-line @typescript-eslint/no-explicit-any
-    delete configFields.apiKeyEnvironmentVariable;
-    return new ChatOpenAI(configFields);
+    // `ChatOllama` is a native client, so NOTHING in a `configuration` block reaches it — say so
+    // before dropping it, rather than letting a stale config behave differently than it reads.
+    warnUnusedConfiguration({
+        provider: 'ollama',
+        configuration: configFields.configuration,
+        consumedPaths: [],
+        reason: NATIVE_CLIENT_REASON,
+        guidance: 'Ollama runs as a local, unauthenticated daemon: point elsewhere with the OLLAMA_HOST ' +
+            'environment variable or the "baseUrl" field of the "llm" block, and set "headers" there ' +
+            'if a reverse proxy in front of it needs auth.',
+    });
+    // Strip OpenAI-client / gaunt-sloth-internal keys ChatOllama neither needs nor understands.
+    for (const key of ['type', 'apiKeyEnvironmentVariable', 'apiKey', 'configuration']) {
+        delete configFields[key];
+    }
+    return new ChatOllama(configFields);
 }
 export function init(configFileName, force = false, model) {
     // Determine which content to use based on file extension

package/dist/providers/ollama.js.map

@@ -1 +1 @@
-{"version":3,"file":"ollama.js","sourceRoot":"","sources":["../../src/providers/ollama.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,cAAc,EAAE,MAAM,4BAA4B,CAAC;AAC5D,OAAO,EAAE,GAAG,EAAE,MAAM,2BAA2B,CAAC;AAQhD,OAAO,EAAE,2BAA2B,EAAE,MAAM,yBAAyB,CAAC;AACtE,OAAO,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,MAAM,kCAAkC,CAAC;AAEnG;;;;GAIG;AACH,MAAM,mBAAmB,GAAG,wBAAwB,CAAC;AAErD;;;;GAIG;AACH,MAAM,0BAA0B,GAAG,QAAQ,CAAC;AAE5C;;;;;;GAMG;AACH,SAAS,cAAc;IACrB,MAAM,IAAI,GAAG,GAAG,CAAC,WAAW,CAAC;IAC7B,IAAI,IAAY,CAAC;IACjB,IAAI,CAAC,IAAI,EAAE,CAAC;QACV,IAAI,GAAG,mBAAmB,CAAC;IAC7B,CAAC;SAAM,IAAI,cAAc,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;QACrC,IAAI,GAAG,IAAI,CAAC,OAAO,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;IAClC,CAAC;SAAM,CAAC;QACN,IAAI,GAAG,UAAU,IAAI,EAAE,CAAC,OAAO,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;IAC9C,CAAC;IACD,OAAO,GAAG,IAAI,KAAK,CAAC;AACtB,CAAC;AAED,wFAAwF;AACxF,qCAAqC;AACrC,MAAM,CAAC,KAAK,UAAU,iBAAiB,CACrC,SAAmE;IAEnE,MAAM,EAAE,UAAU,EAAE,GAAG,MAAM,MAAM,CAAC,mBAAmB,CAAC,CAAC;IACzD,+EAA+E;IAC/E,MAAM,MAAM,GAAG,SAAS,CAAC,MAAM,IAAI,0BAA0B,CAAC;IAC9D,MAAM,YAAY,GAAG;QACnB,GAAG,SAAS;QACZ,MAAM;QACN,KAAK,EAAE,SAAS,CAAC,KAAK,IAAI,uBAAuB,CAAC,QAAQ,CAAC;QAC3D,aAAa,EAAE;YACb,OAAO,EAAE,cAAc,EAAE;YACzB,GAAG,CAAC,SAAS,CAAC,aAAa,IAAI,EAAE,CAAC;SACnC;KACF,CAAC;IACF,8DAA8D;IAC9D,OAAQ,YAAoB,CAAC,IAAI,CAAC;IAClC,8DAA8D;IAC9D,OAAQ,YAAoB,CAAC,yBAAyB,CAAC;IAEvD,OAAO,IAAI,UAAU,CAAC,YAAY,CAAC,CAAC;AACtC,CAAC;AAED,MAAM,UAAU,IAAI,CAAC,cAAsB,EAAE,KAAK,GAAG,KAAK,EAAE,KAAc;IACxE,yDAAyD;IACzD,IAAI,CAAC,cAAc,CAAC,QAAQ,CAAC,OAAO,CAAC,EAAE,CAAC;QACtC,MAAM,IAAI,KAAK,CAAC,gCAAgC,CAAC,CAAC;IACpD,CAAC;IAED,2BAA2B,CAAC,cAAc,EAAE,sBAAsB,CAAC,QAAQ,EAAE,KAAK,CAAC,EAAE,KAAK,CAAC,CAAC;IAC5F,cAAc,CACZ,yBAAyB,cAAc,2BAA2B;QAChE,2EAA2E;QAC3E,+CAA+C,CAClD,CAAC;AACJ,CAAC"}
+{"version":3,"file":"ollama.js","sourceRoot":"","sources":["../../src/providers/ollama.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,cAAc,EAAE,MAAM,4BAA4B,CAAC;AAC5D,OAAO,EAAE,GAAG,EAAE,MAAM,2BAA2B,CAAC;AAOhD,OAAO,EAAE,2BAA2B,EAAE,MAAM,yBAAyB,CAAC;AACtE,OAAO,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,MAAM,kCAAkC,CAAC;AACnG,OAAO,EACL,oBAAoB,EACpB,uBAAuB,GACxB,MAAM,4CAA4C,CAAC;AAEpD;;;;;;GAMG;AACH,MAAM,mBAAmB,GAAG,wBAAwB,CAAC;AAErD;;;;;;;;;;;;;;;;;GAiBG;AACH,MAAM,sBAAsB,GAAG,KAAK,CAAC;AAErC;;;;;GAKG;AACH,SAAS,cAAc;IACrB,MAAM,IAAI,GAAG,GAAG,CAAC,WAAW,CAAC;IAC7B,IAAI,CAAC,IAAI;QAAE,OAAO,mBAAmB,CAAC;IACtC,MAAM,IAAI,GAAG,cAAc,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,UAAU,IAAI,EAAE,CAAC;IACjE,OAAO,IAAI,CAAC,OAAO,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;AAClC,CAAC;AAED;;;;;;;;;;;;;;GAcG;AACH,qCAAqC;AACrC,MAAM,CAAC,KAAK,UAAU,iBAAiB,CACrC,SAAgD;IAEhD,MAAM,EAAE,UAAU,EAAE,GAAG,MAAM,MAAM,CAAC,mBAAmB,CAAC,CAAC;IACzD,MAAM,YAAY,GAA4B;QAC5C,GAAG,SAAS;QACZ,KAAK,EAAE,SAAS,CAAC,KAAK,IAAI,uBAAuB,CAAC,QAAQ,CAAC;QAC3D,wFAAwF;QACxF,6FAA6F;QAC7F,OAAO,EAAE,SAAS,CAAC,OAAO,IAAI,cAAc,EAAE;QAC9C,MAAM,EAAE,SAAS,CAAC,MAAM,IAAI,sBAAsB;KACnD,CAAC;IACF,6FAA6F;IAC7F,2FAA2F;IAC3F,uBAAuB,CAAC;QACtB,QAAQ,EAAE,QAAQ;QAClB,aAAa,EAAE,YAAY,CAAC,aAAa;QACzC,aAAa,EAAE,EAAE;QACjB,MAAM,EAAE,oBAAoB;QAC5B,QAAQ,EACN,uFAAuF;YACvF,0FAA0F;YAC1F,+CAA+C;KAClD,CAAC,CAAC;IACH,4FAA4F;IAC5F,KAAK,MAAM,GAAG,IAAI,CAAC,MAAM,EAAE,2BAA2B,EAAE,QAAQ,EAAE,eAAe,CAAC,EAAE,CAAC;QACnF,OAAO,YAAY,CAAC,GAAG,CAAC,CAAC;IAC3B,CAAC;IACD,OAAO,IAAI,UAAU,CAAC,YAA+B,CAAC,CAAC;AACzD,CAAC;AAED,MAAM,UAAU,IAAI,CAAC,cAAsB,EAAE,KAAK,GAAG,KAAK,EAAE,KAAc;IACxE,yDAAyD;IACzD,IAAI,CAAC,cAAc,CAAC,QAAQ,CAAC,OAAO,CAAC,EAAE,CAAC;QACtC,MAAM,IAAI,KAAK,CAAC,gCAAgC,CAAC,CAAC;IACpD,CAAC;IAED,2BAA2B,CAAC,cAAc,EAAE,sBAAsB,CAAC,QAAQ,EAAE,KAAK,CAAC,EAAE,KAAK,CAAC,CAAC;IAC5F,cAAc,CACZ,yBAAyB,cAAc,2BAA2B;QAChE,2EAA2E;QAC3E,+CAA+C,CAClD,CAAC;AACJ,CAAC"}

package/dist/providers/openai.js

@@ -16,6 +16,35 @@ const SUPPORTED_DEFAULT_TEMPERATURE = 1;
 function isTemperatureRestrictedModel(model) {
     return !!model && TEMPERATURE_RESTRICTED_MODEL.test(model);
 }
+/**
+ * Decide whether to force the OpenAI Responses API (`/v1/responses`) for this request.
+ *
+ * OpenAI's `/v1/chat/completions` endpoint rejects reasoning-capable models (the gpt-5.x / o-series
+ * families) when `reasoning_effort` is combined with function tools:
+ *   "Function tools with reasoning_effort are not supported for gpt-5.6-luna in
+ *    /v1/chat/completions. To use function tools, use /v1/responses or set reasoning_effort to
+ *    'none'."
+ * LangChain's auto-router only flips to the Responses API for OpenAI built-in tools,
+ * `reasoning.summary` (not `.effort`), custom tools, or a hardcoded model allowlist — none of which
+ * covers ordinary function tools on a reasoning model that isn't yet on that allowlist (e.g.
+ * gpt-5.6-luna). Crucially, these models **default `reasoning_effort` to a non-`none` value**, so the
+ * collision fires even when the gth config sets NO reasoning at all (the failing case: a bare
+ * `{ type: openai, model: gpt-5.6-luna }` still 400s on `chat/completions` + tools). So for any
+ * reasoning-capable model, when the user has not made an explicit choice, default `useResponsesApi`
+ * to true — do NOT gate on reasoning being explicitly configured (GS2-74's original bug: it required
+ * `reasoningEffort`/`reasoning` to be set, missing the far more common bare-config path).
+ *
+ * Guards: (1) only reasoning-capable models — leave gpt-4o etc. on completions; (2) never override an
+ * explicit `useResponsesApi` the user set (true OR false). The model-name guard also scopes this away
+ * from OpenAI-compatible providers that reuse this file (Inception/DeepSeek/xAI ids don't match the
+ * regex). Routing a no-tools/plain call to Responses is still correct FOR GTH because gth never
+ * injects reasoning params itself — so even a non-reasoning `gpt-5`-prefixed id (e.g.
+ * `gpt-5-chat-latest`, which this regex also sweeps) is accepted there. (Responses isn't a universal
+ * superset — it drops some chat/completions-only params like `n>1` — but gth uses none of those.)
+ */
+function shouldUseResponsesApi(fields) {
+    return fields.useResponsesApi === undefined && isTemperatureRestrictedModel(fields.model);
+}
 // Function to process JSON config and create OpenAI LLM instance
 // noinspection JSUnusedGlobalSymbols
 export async function processJsonConfig(llmConfig) {
@@ -43,6 +72,11 @@ export async function processJsonConfig(llmConfig) {
         // eslint-disable-next-line @typescript-eslint/no-explicit-any
         delete configFields.temperature;
     }
+    // Reasoning-capable OpenAI models 400 on `/v1/chat/completions` when `reasoning_effort` meets
+    // function tools; route them to the Responses API unless the user pinned `useResponsesApi`.
+    if (shouldUseResponsesApi(configFields)) {
+        configFields.useResponsesApi = true;
+    }
     return new ChatOpenAI(configFields);
 }
 function getApiKey(llmConfig) {

package/dist/providers/openai.js.map

@@ -1 +1 @@
-{"version":3,"file":"openai.js","sourceRoot":"","sources":["../../src/providers/openai.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,cAAc,EAAE,MAAM,4BAA4B,CAAC;AAC5D,OAAO,EAAE,GAAG,EAAE,MAAM,2BAA2B,CAAC;AAQhD,OAAO,EAAE,2BAA2B,EAAE,MAAM,yBAAyB,CAAC;AACtE,OAAO,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,MAAM,kCAAkC,CAAC;AAEnG;;;;;;;GAOG;AACH,MAAM,4BAA4B,GAAG,eAAe,CAAC;AAErD,yFAAyF;AACzF,MAAM,6BAA6B,GAAG,CAAC,CAAC;AAExC,SAAS,4BAA4B,CAAC,KAAyB;IAC7D,OAAO,CAAC,CAAC,KAAK,IAAI,4BAA4B,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;AAC7D,CAAC;AAED,iEAAiE;AACjE,qCAAqC;AACrC,MAAM,CAAC,KAAK,UAAU,iBAAiB,CACrC,SAAmE;IAEnE,MAAM,EAAE,UAAU,EAAE,GAAG,MAAM,MAAM,CAAC,mBAAmB,CAAC,CAAC;IACzD,wEAAwE;IACxE,MAAM,YAAY,GAAG,SAAS,CAAC,SAAS,CAAC,CAAC;IAC1C,MAAM,YAAY,GAAG;QACnB,GAAG,SAAS;QACZ,MAAM,EAAE,YAAY;QACpB,KAAK,EAAE,SAAS,CAAC,KAAK,IAAI,uBAAuB,CAAC,QAAQ,CAAC;KAC5D,CAAC;IACF,8DAA8D;IAC9D,OAAQ,YAAoB,CAAC,IAAI,CAAC;IAClC,8DAA8D;IAC9D,OAAQ,YAAoB,CAAC,yBAAyB,CAAC;IAEvD,6FAA6F;IAC7F,8FAA8F;IAC9F,6FAA6F;IAC7F,IACE,4BAA4B,CAAC,YAAY,CAAC,KAAK,CAAC;QAChD,YAAY,CAAC,WAAW,KAAK,SAAS;QACtC,YAAY,CAAC,WAAW,KAAK,6BAA6B,EAC1D,CAAC;QACD,cAAc,CACZ,UAAU,YAAY,CAAC,KAAK,0CAA0C;YACpE,qBAAqB,6BAA6B,gBAAgB;YAClE,0CAA0C,YAAY,CAAC,WAAW,GAAG,CACxE,CAAC;QACF,8DAA8D;QAC9D,OAAQ,YAAoB,CAAC,WAAW,CAAC;IAC3C,CAAC;IAED,OAAO,IAAI,UAAU,CAAC,YAAY,CAAC,CAAC;AACtC,CAAC;AAED,SAAS,SAAS,CAAC,SAAmE;IACpF,8DAA8D;IAC9D,MAAM,IAAI,GAAG,SAA0C,CAAC;IACxD,IAAI,IAAI,CAAC,yBAAyB,IAAI,GAAG,CAAC,IAAI,CAAC,yBAAyB,CAAC,EAAE,CAAC;QAC1E,OAAO,GAAG,CAAC,IAAI,CAAC,yBAAyB,CAAC,CAAC;IAC7C,CAAC;SAAM,CAAC;QACN,OAAO,SAAS,CAAC,MAAM,IAAI,GAAG,CAAC,cAAc,CAAC;IAChD,CAAC;AACH,CAAC;AAED,MAAM,UAAU,IAAI,CAAC,cAAsB,EAAE,KAAK,GAAG,KAAK,EAAE,KAAc;IACxE,yDAAyD;IACzD,IAAI,CAAC,cAAc,CAAC,QAAQ,CAAC,OAAO,CAAC,EAAE,CAAC;QACtC,MAAM,IAAI,KAAK,CAAC,gCAAgC,CAAC,CAAC;IACpD,CAAC;IAED,2BAA2B,CAAC,cAAc,EAAE,sBAAsB,CAAC,QAAQ,EAAE,KAAK,CAAC,EAAE,KAAK,CAAC,CAAC;IAC5F,cAAc,CACZ,yBAAyB,cAAc,uBAAuB;QAC5D,gDAAgD,CACnD,CAAC;AACJ,CAAC"}
+{"version":3,"file":"openai.js","sourceRoot":"","sources":["../../src/providers/openai.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,cAAc,EAAE,MAAM,4BAA4B,CAAC;AAC5D,OAAO,EAAE,GAAG,EAAE,MAAM,2BAA2B,CAAC;AAQhD,OAAO,EAAE,2BAA2B,EAAE,MAAM,yBAAyB,CAAC;AACtE,OAAO,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,MAAM,kCAAkC,CAAC;AAEnG;;;;;;;GAOG;AACH,MAAM,4BAA4B,GAAG,eAAe,CAAC;AAErD,yFAAyF;AACzF,MAAM,6BAA6B,GAAG,CAAC,CAAC;AAExC,SAAS,4BAA4B,CAAC,KAAyB;IAC7D,OAAO,CAAC,CAAC,KAAK,IAAI,4BAA4B,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;AAC7D,CAAC;AAED;;;;;;;;;;;;;;;;;;;;;;;;;GAyBG;AACH,SAAS,qBAAqB,CAAC,MAAqD;IAClF,OAAO,MAAM,CAAC,eAAe,KAAK,SAAS,IAAI,4BAA4B,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;AAC5F,CAAC;AAED,iEAAiE;AACjE,qCAAqC;AACrC,MAAM,CAAC,KAAK,UAAU,iBAAiB,CACrC,SAAmE;IAEnE,MAAM,EAAE,UAAU,EAAE,GAAG,MAAM,MAAM,CAAC,mBAAmB,CAAC,CAAC;IACzD,wEAAwE;IACxE,MAAM,YAAY,GAAG,SAAS,CAAC,SAAS,CAAC,CAAC;IAC1C,MAAM,YAAY,GAAG;QACnB,GAAG,SAAS;QACZ,MAAM,EAAE,YAAY;QACpB,KAAK,EAAE,SAAS,CAAC,KAAK,IAAI,uBAAuB,CAAC,QAAQ,CAAC;KAC5D,CAAC;IACF,8DAA8D;IAC9D,OAAQ,YAAoB,CAAC,IAAI,CAAC;IAClC,8DAA8D;IAC9D,OAAQ,YAAoB,CAAC,yBAAyB,CAAC;IAEvD,6FAA6F;IAC7F,8FAA8F;IAC9F,6FAA6F;IAC7F,IACE,4BAA4B,CAAC,YAAY,CAAC,KAAK,CAAC;QAChD,YAAY,CAAC,WAAW,KAAK,SAAS;QACtC,YAAY,CAAC,WAAW,KAAK,6BAA6B,EAC1D,CAAC;QACD,cAAc,CACZ,UAAU,YAAY,CAAC,KAAK,0CAA0C;YACpE,qBAAqB,6BAA6B,gBAAgB;YAClE,0CAA0C,YAAY,CAAC,WAAW,GAAG,CACxE,CAAC;QACF,8DAA8D;QAC9D,OAAQ,YAAoB,CAAC,WAAW,CAAC;IAC3C,CAAC;IAED,8FAA8F;IAC9F,4FAA4F;IAC5F,IAAI,qBAAqB,CAAC,YAAY,CAAC,EAAE,CAAC;QACxC,YAAY,CAAC,eAAe,GAAG,IAAI,CAAC;IACtC,CAAC;IAED,OAAO,IAAI,UAAU,CAAC,YAAY,CAAC,CAAC;AACtC,CAAC;AAED,SAAS,SAAS,CAAC,SAAmE;IACpF,8DAA8D;IAC9D,MAAM,IAAI,GAAG,SAA0C,CAAC;IACxD,IAAI,IAAI,CAAC,yBAAyB,IAAI,GAAG,CAAC,IAAI,CAAC,yBAAyB,CAAC,EAAE,CAAC;QAC1E,OAAO,GAAG,CAAC,IAAI,CAAC,yBAAyB,CAAC,CAAC;IAC7C,CAAC;SAAM,CAAC;QACN,OAAO,SAAS,CAAC,MAAM,IAAI,GAAG,CAAC,cAAc,CAAC;IAChD,CAAC;AACH,CAAC;AAED,MAAM,UAAU,IAAI,CAAC,cAAsB,EAAE,KAAK,GAAG,KAAK,EAAE,KAAc;IACxE,yDAAyD;IACzD,IAAI,CAAC,cAAc,CAAC,QAAQ,CAAC,OAAO,CAAC,EAAE,CAAC;QACtC,MAAM,IAAI,KAAK,CAAC,gCAAgC,CAAC,CAAC;IACpD,CAAC;IAED,2BAA2B,CAAC,cAAc,EAAE,sBAAsB,CAAC,QAAQ,EAAE,KAAK,CAAC,EAAE,KAAK,CAAC,CAAC;IAC5F,cAAc,CACZ,yBAAyB,cAAc,uBAAuB;QAC5D,gDAAgD,CACnD,CAAC;AACJ,CAAC"}

package/dist/providers/openrouter.d.ts

@@ -1,5 +1,27 @@
-import { BaseChatModel, type BaseChatModelParams } from '@langchain/core/language_models/chat_models';
-import { OpenAIChatInput } from '@langchain/openai';
-import { ChatOpenAIFields } from '@langchain/openai';
-export declare function processJsonConfig(llmConfig: OpenAIChatInput & ChatOpenAIFields & BaseChatModelParams): Promise<BaseChatModel>;
+import type { BaseChatModel, BaseChatModelParams } from '@langchain/core/language_models/chat_models';
+import type { ChatOpenRouterInput } from '@langchain/openrouter';
+/**
+ * Build a native `ChatOpenRouter` (`@langchain/openrouter`). It is NOT a `ChatOpenAI` subclass: it
+ * talks to the OpenRouter REST API through the global `fetch` with no SDK client in between, and
+ * `_llmType()` reports `openrouter`.
+ *
+ * OpenRouter's own options are therefore native TOP-LEVEL fields of the `llm` block, forwarded by
+ * the `...restConfig` spread below: `provider` (routing preferences), `models` (the fallback list),
+ * `route`, `plugins`, `transforms`, `trace`, `minP`, `topA`, `repetitionPenalty`, `seed`,
+ * `logitBias`, `topLogprobs`, `sessionId`. Write them beside `model`, not under `configuration`.
+ *
+ * Migration note — the OpenAI-client knobs a `configuration` block used to carry are gone, because
+ * there is no client to give them to. `configuration.baseURL` still works (mapped to the native
+ * `baseURL`), and the two attribution headers are still honoured as a fallback for
+ * `siteUrl`/`siteName`. What has no injection point at all is transport: `fetch`, `fetchOptions`,
+ * `timeout`, `maxRetries`, `defaultQuery`, `dangerouslyAllowBrowser`, and any other
+ * `defaultHeaders` entry. That is an upstream surface gap, not something this module can restore —
+ * per-provider transport would need a `fetch` option added to `@langchain/openrouter`.
+ *
+ * The replacement is PROCESS-WIDE rather than per-provider, and it is verified working: run node
+ * with `--use-env-proxy` (or set `NODE_USE_ENV_PROXY=1`) together with `HTTP_PROXY`/`HTTPS_PROXY`,
+ * and the global `fetch` this model calls routes through the proxy — measured end to end against a
+ * local proxy, with a no-proxy control that stayed direct.
+ */
+export declare function processJsonConfig(llmConfig: ChatOpenRouterInput & BaseChatModelParams & Record<string, unknown>): Promise<BaseChatModel>;
 export declare function init(configFileName: string, force?: boolean, model?: string): void;

package/dist/providers/openrouter.js

@@ -2,43 +2,100 @@ import { displayWarning } from '#src/utils/consoleUtils.js';
 import { env } from '#src/utils/systemUtils.js';
 import { writeConfigFileWithMessages } from '#src/utils/fileUtils.js';
 import { buildInitConfigContent, getCuratedFallbackModel } from '#src/providers/modelDiscovery.js';
-// Function to process JSON config and create OpenRouter LLM instance
+import { NATIVE_CLIENT_REASON, warnConfigurationOverridesTopLevelField, warnUnappliedConfigurationPath, warnUnusedConfiguration, } from '#src/providers/configurationPassthrough.js';
+/**
+ * The OpenRouter attribution headers, and the native `ChatOpenRouter` field each one is set from.
+ * Declared once because they are read in two places that MUST agree: the `siteUrl`/`siteName`
+ * fallback below, and the list of `configuration` paths this factory consumes — if those drift, the
+ * orphaned-config warning starts firing on a header that is in fact honoured.
+ */
+const ATTRIBUTION_HEADERS = { siteUrl: 'HTTP-Referer', siteName: 'X-Title' };
+/**
+ * The only paths inside a user's `configuration` block that this factory reads. Everything else in
+ * it reaches nothing — see the migration note on {@link processJsonConfig}.
+ */
+const CONSUMED_CONFIGURATION_PATHS = [
+    'baseURL',
+    `defaultHeaders.${ATTRIBUTION_HEADERS.siteUrl}`,
+    `defaultHeaders.${ATTRIBUTION_HEADERS.siteName}`,
+];
+/**
+ * Build a native `ChatOpenRouter` (`@langchain/openrouter`). It is NOT a `ChatOpenAI` subclass: it
+ * talks to the OpenRouter REST API through the global `fetch` with no SDK client in between, and
+ * `_llmType()` reports `openrouter`.
+ *
+ * OpenRouter's own options are therefore native TOP-LEVEL fields of the `llm` block, forwarded by
+ * the `...restConfig` spread below: `provider` (routing preferences), `models` (the fallback list),
+ * `route`, `plugins`, `transforms`, `trace`, `minP`, `topA`, `repetitionPenalty`, `seed`,
+ * `logitBias`, `topLogprobs`, `sessionId`. Write them beside `model`, not under `configuration`.
+ *
+ * Migration note — the OpenAI-client knobs a `configuration` block used to carry are gone, because
+ * there is no client to give them to. `configuration.baseURL` still works (mapped to the native
+ * `baseURL`), and the two attribution headers are still honoured as a fallback for
+ * `siteUrl`/`siteName`. What has no injection point at all is transport: `fetch`, `fetchOptions`,
+ * `timeout`, `maxRetries`, `defaultQuery`, `dangerouslyAllowBrowser`, and any other
+ * `defaultHeaders` entry. That is an upstream surface gap, not something this module can restore —
+ * per-provider transport would need a `fetch` option added to `@langchain/openrouter`.
+ *
+ * The replacement is PROCESS-WIDE rather than per-provider, and it is verified working: run node
+ * with `--use-env-proxy` (or set `NODE_USE_ENV_PROXY=1`) together with `HTTP_PROXY`/`HTTPS_PROXY`,
+ * and the global `fetch` this model calls routes through the proxy — measured end to end against a
+ * local proxy, with a no-proxy control that stayed direct.
+ */
 // noinspection JSUnusedGlobalSymbols
 export async function processJsonConfig(llmConfig) {
-    const { ChatOpenAI } = await import('@langchain/openai');
-    // Use environment variable if available, otherwise use the config value
+    const { ChatOpenRouter } = await import('@langchain/openrouter');
     const openRouterApiKey = getApiKey(llmConfig);
     if (!openRouterApiKey) {
         throw new Error('You need to define OPEN_ROUTER_API_KEY environment variable, or set apiKey in your config file.');
     }
-    const configFields = {
-        ...llmConfig,
+    const { type: _type, apiKeyEnvironmentVariable: _envVar, configuration, siteUrl, siteName, ...restConfig } = llmConfig;
+    const configObj = configuration;
+    // The one place that decides whether the user's `configuration.baseURL` is applied; the warning
+    // below reads THIS result instead of re-testing the value, so the guard and what the user is told
+    // cannot drift apart.
+    const baseURLOverride = configObj?.baseURL ? { baseURL: configObj.baseURL } : {};
+    warnUnusedConfiguration({
+        provider: 'openrouter',
+        configuration,
+        consumedPaths: CONSUMED_CONFIGURATION_PATHS,
+        reason: NATIVE_CLIENT_REASON,
+        // Scope the advice to the options that HAVE a top-level home: told plainly to "put it at the
+        // top level", a user with a transport key would move it from a warned location to an unwarned
+        // one, which is worse than where they started.
+        guidance: 'OpenRouter\'s own options are native top-level fields of the "llm" block — set "provider" ' +
+            '(routing preferences), "models", "route", "plugins" or "transforms" beside "model" rather ' +
+            'than under "configuration". Transport settings (proxy, custom fetch, timeout, other ' +
+            'headers) have no top-level field and no per-provider hook at all: run node with ' +
+            '--use-env-proxy (or set NODE_USE_ENV_PROXY=1) plus HTTP_PROXY/HTTPS_PROXY to route this ' +
+            'provider through a proxy process-wide.',
+    });
+    warnUnappliedConfigurationPath('openrouter', configuration, 'baseURL', 'baseURL' in baseURLOverride, 'Give it the full base URL of the OpenRouter-compatible endpoint, or remove it to use the ' +
+        'default https://openrouter.ai/api/v1.');
+    // `baseURL` is ALSO a native top-level field of `ChatOpenRouter`, forwarded by `...restConfig`
+    // below — and the block's value is spread after it, so it wins. Both are honoured surfaces, so
+    // the defect is only the silence: say which one takes effect rather than dropping the other
+    // without a word.
+    warnConfigurationOverridesTopLevelField('openrouter', 'baseURL', restConfig.baseURL, baseURLOverride.baseURL);
+    const resolvedSiteUrl = siteUrl ??
+        configObj?.defaultHeaders?.[ATTRIBUTION_HEADERS.siteUrl] ??
+        'https://gauntsloth.app/';
+    const resolvedSiteName = siteName ?? configObj?.defaultHeaders?.[ATTRIBUTION_HEADERS.siteName] ?? 'Gaunt Sloth';
+    return new ChatOpenRouter({
+        ...restConfig,
         apiKey: openRouterApiKey,
         model: llmConfig.model || getCuratedFallbackModel('openrouter'),
-        configuration: {
-            baseURL: 'https://openrouter.ai/api/v1',
-            ...(llmConfig.configuration || {}),
-            defaultHeaders: {
-                'HTTP-Referer': 'https://gauntsloth.app/',
-                'X-Title': 'Gaunt Sloth',
-            },
-        },
-    };
-    // eslint-disable-next-line @typescript-eslint/no-explicit-any
-    delete configFields.type;
-    // eslint-disable-next-line @typescript-eslint/no-explicit-any
-    delete configFields.apiKeyEnvironmentVariable;
-    return new ChatOpenAI(configFields);
+        siteUrl: resolvedSiteUrl,
+        siteName: resolvedSiteName,
+        ...baseURLOverride,
+    });
 }
 function getApiKey(llmConfig) {
-    // eslint-disable-next-line @typescript-eslint/no-explicit-any
-    const conf = llmConfig;
-    if (conf.apiKeyEnvironmentVariable && env[conf.apiKeyEnvironmentVariable]) {
-        return env[conf.apiKeyEnvironmentVariable];
-    }
-    else {
-        return llmConfig.apiKey || env.OPEN_ROUTER_API_KEY || env.OPENROUTER_API_KEY;
+    const envVarName = llmConfig.apiKeyEnvironmentVariable;
+    if (envVarName && env[envVarName]) {
+        return env[envVarName];
     }
+    return (llmConfig.apiKey || env.OPEN_ROUTER_API_KEY || env.OPENROUTER_API_KEY);
 }
 export function init(configFileName, force = false, model) {
     // Determine which content to use based on file extension

package/dist/providers/openrouter.js.map

@@ -1 +1 @@
-{"version":3,"file":"openrouter.js","sourceRoot":"","sources":["../../src/providers/openrouter.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,cAAc,EAAE,MAAM,4BAA4B,CAAC;AAC5D,OAAO,EAAE,GAAG,EAAE,MAAM,2BAA2B,CAAC;AAQhD,OAAO,EAAE,2BAA2B,EAAE,MAAM,yBAAyB,CAAC;AACtE,OAAO,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,MAAM,kCAAkC,CAAC;AAEnG,qEAAqE;AACrE,qCAAqC;AACrC,MAAM,CAAC,KAAK,UAAU,iBAAiB,CACrC,SAAmE;IAEnE,MAAM,EAAE,UAAU,EAAE,GAAG,MAAM,MAAM,CAAC,mBAAmB,CAAC,CAAC;IACzD,wEAAwE;IACxE,MAAM,gBAAgB,GAAG,SAAS,CAAC,SAAS,CAAC,CAAC;IAC9C,IAAI,CAAC,gBAAgB,EAAE,CAAC;QACtB,MAAM,IAAI,KAAK,CACb,iGAAiG,CAClG,CAAC;IACJ,CAAC;IACD,MAAM,YAAY,GAAG;QACnB,GAAG,SAAS;QACZ,MAAM,EAAE,gBAAgB;QACxB,KAAK,EAAE,SAAS,CAAC,KAAK,IAAI,uBAAuB,CAAC,YAAY,CAAC;QAC/D,aAAa,EAAE;YACb,OAAO,EAAE,8BAA8B;YACvC,GAAG,CAAC,SAAS,CAAC,aAAa,IAAI,EAAE,CAAC;YAClC,cAAc,EAAE;gBACd,cAAc,EAAE,yBAAyB;gBACzC,SAAS,EAAE,aAAa;aACzB;SACF;KACF,CAAC;IACF,8DAA8D;IAC9D,OAAQ,YAAoB,CAAC,IAAI,CAAC;IAClC,8DAA8D;IAC9D,OAAQ,YAAoB,CAAC,yBAAyB,CAAC;IACvD,OAAO,IAAI,UAAU,CAAC,YAAY,CAAC,CAAC;AACtC,CAAC;AAED,SAAS,SAAS,CAAC,SAAmE;IACpF,8DAA8D;IAC9D,MAAM,IAAI,GAAG,SAA0C,CAAC;IACxD,IAAI,IAAI,CAAC,yBAAyB,IAAI,GAAG,CAAC,IAAI,CAAC,yBAAyB,CAAC,EAAE,CAAC;QAC1E,OAAO,GAAG,CAAC,IAAI,CAAC,yBAAyB,CAAC,CAAC;IAC7C,CAAC;SAAM,CAAC;QACN,OAAO,SAAS,CAAC,MAAM,IAAI,GAAG,CAAC,mBAAmB,IAAI,GAAG,CAAC,kBAAkB,CAAC;IAC/E,CAAC;AACH,CAAC;AAED,MAAM,UAAU,IAAI,CAAC,cAAsB,EAAE,KAAK,GAAG,KAAK,EAAE,KAAc;IACxE,yDAAyD;IACzD,IAAI,CAAC,cAAc,CAAC,QAAQ,CAAC,OAAO,CAAC,EAAE,CAAC;QACtC,MAAM,IAAI,KAAK,CAAC,gCAAgC,CAAC,CAAC;IACpD,CAAC;IAED,2BAA2B,CAAC,cAAc,EAAE,sBAAsB,CAAC,YAAY,EAAE,KAAK,CAAC,EAAE,KAAK,CAAC,CAAC;IAChG,cAAc,CACZ,yBAAyB,cAAc,uBAAuB;QAC5D,qDAAqD,CACxD,CAAC;AACJ,CAAC"}
+{"version":3,"file":"openrouter.js","sourceRoot":"","sources":["../../src/providers/openrouter.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,cAAc,EAAE,MAAM,4BAA4B,CAAC;AAC5D,OAAO,EAAE,GAAG,EAAE,MAAM,2BAA2B,CAAC;AAOhD,OAAO,EAAE,2BAA2B,EAAE,MAAM,yBAAyB,CAAC;AACtE,OAAO,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,MAAM,kCAAkC,CAAC;AACnG,OAAO,EACL,oBAAoB,EACpB,uCAAuC,EACvC,8BAA8B,EAC9B,uBAAuB,GACxB,MAAM,4CAA4C,CAAC;AAEpD;;;;;GAKG;AACH,MAAM,mBAAmB,GAAG,EAAE,OAAO,EAAE,cAAc,EAAE,QAAQ,EAAE,SAAS,EAAW,CAAC;AAEtF;;;GAGG;AACH,MAAM,4BAA4B,GAAG;IACnC,SAAS;IACT,kBAAkB,mBAAmB,CAAC,OAAO,EAAE;IAC/C,kBAAkB,mBAAmB,CAAC,QAAQ,EAAE;CACxC,CAAC;AAEX;;;;;;;;;;;;;;;;;;;;;;GAsBG;AACH,qCAAqC;AACrC,MAAM,CAAC,KAAK,UAAU,iBAAiB,CACrC,SAA8E;IAE9E,MAAM,EAAE,cAAc,EAAE,GAAG,MAAM,MAAM,CAAC,uBAAuB,CAAC,CAAC;IAEjE,MAAM,gBAAgB,GAAG,SAAS,CAAC,SAAS,CAAC,CAAC;IAC9C,IAAI,CAAC,gBAAgB,EAAE,CAAC;QACtB,MAAM,IAAI,KAAK,CACb,iGAAiG,CAClG,CAAC;IACJ,CAAC;IAED,MAAM,EACJ,IAAI,EAAE,KAAK,EACX,yBAAyB,EAAE,OAAO,EAClC,aAAa,EACb,OAAO,EACP,QAAQ,EACR,GAAG,UAAU,EACd,GAAG,SAAS,CAAC;IAEd,MAAM,SAAS,GAAG,aACyD,CAAC;IAE5E,gGAAgG;IAChG,kGAAkG;IAClG,sBAAsB;IACtB,MAAM,eAAe,GAAG,SAAS,EAAE,OAAO,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,SAAS,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;IAEjF,uBAAuB,CAAC;QACtB,QAAQ,EAAE,YAAY;QACtB,aAAa;QACb,aAAa,EAAE,4BAA4B;QAC3C,MAAM,EAAE,oBAAoB;QAC5B,6FAA6F;QAC7F,8FAA8F;QAC9F,+CAA+C;QAC/C,QAAQ,EACN,4FAA4F;YAC5F,4FAA4F;YAC5F,sFAAsF;YACtF,kFAAkF;YAClF,0FAA0F;YAC1F,wCAAwC;KAC3C,CAAC,CAAC;IAEH,8BAA8B,CAC5B,YAAY,EACZ,aAAa,EACb,SAAS,EACT,SAAS,IAAI,eAAe,EAC5B,2FAA2F;QACzF,uCAAuC,CAC1C,CAAC;IAEF,+FAA+F;IAC/F,+FAA+F;IAC/F,4FAA4F;IAC5F,kBAAkB;IAClB,uCAAuC,CACrC,YAAY,EACZ,SAAS,EACT,UAAU,CAAC,OAAO,EAClB,eAAe,CAAC,OAAO,CACxB,CAAC;IAEF,MAAM,eAAe,GACnB,OAAO;QACP,SAAS,EAAE,cAAc,EAAE,CAAC,mBAAmB,CAAC,OAAO,CAAC;QACxD,yBAAyB,CAAC;IAC5B,MAAM,gBAAgB,GACpB,QAAQ,IAAI,SAAS,EAAE,cAAc,EAAE,CAAC,mBAAmB,CAAC,QAAQ,CAAC,IAAI,aAAa,CAAC;IAEzF,OAAO,IAAI,cAAc,CAAC;QACxB,GAAG,UAAU;QACb,MAAM,EAAE,gBAAgB;QACxB,KAAK,EAAE,SAAS,CAAC,KAAK,IAAI,uBAAuB,CAAC,YAAY,CAAC;QAC/D,OAAO,EAAE,eAAe;QACxB,QAAQ,EAAE,gBAAgB;QAC1B,GAAG,eAAe;KACnB,CAAC,CAAC;AACL,CAAC;AAED,SAAS,SAAS,CAAC,SAAkC;IACnD,MAAM,UAAU,GAAG,SAAS,CAAC,yBAA+C,CAAC;IAC7E,IAAI,UAAU,IAAI,GAAG,CAAC,UAAU,CAAC,EAAE,CAAC;QAClC,OAAO,GAAG,CAAC,UAAU,CAAC,CAAC;IACzB,CAAC;IACD,OAAO,CACJ,SAAS,CAAC,MAA6B,IAAI,GAAG,CAAC,mBAAmB,IAAI,GAAG,CAAC,kBAAkB,CAC9F,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,IAAI,CAAC,cAAsB,EAAE,KAAK,GAAG,KAAK,EAAE,KAAc;IACxE,yDAAyD;IACzD,IAAI,CAAC,cAAc,CAAC,QAAQ,CAAC,OAAO,CAAC,EAAE,CAAC;QACtC,MAAM,IAAI,KAAK,CAAC,gCAAgC,CAAC,CAAC;IACpD,CAAC;IAED,2BAA2B,CAAC,cAAc,EAAE,sBAAsB,CAAC,YAAY,EAAE,KAAK,CAAC,EAAE,KAAK,CAAC,CAAC;IAChG,cAAc,CACZ,yBAAyB,cAAc,uBAAuB;QAC5D,qDAAqD,CACxD,CAAC;AACJ,CAAC"}

package/dist/providers/vertexai.js

@@ -13,6 +13,9 @@
 import { displayWarning } from '#src/utils/consoleUtils.js';
 import { writeConfigFileWithMessages } from '#src/utils/fileUtils.js';
 import { buildInitConfigContent, getCuratedFallbackModel } from '#src/providers/modelDiscovery.js';
+import { applyGeminiToolSchemaSanitizer } from '#src/providers/geminiSchemaSanitizer.js';
+import { applyGeminiThoughtSummaries } from '#src/providers/geminiThinking.js';
+import { NATIVE_CLIENT_REASON, warnUnusedConfiguration, } from '#src/providers/configurationPassthrough.js';
 export function init(configFileName, force = false, model) {
     // Determine which content to use based on file extension
     if (!configFileName.endsWith('.json')) {
@@ -31,6 +34,21 @@ export async function processJsonConfig(llmConfig) {
     };
     delete configFields.type;
     delete configFields.apiKeyEnvironmentVariable;
-    return new ChatGoogle(configFields);
+    // `ChatGoogle` is a native client for the Gemini API, so nothing in a `configuration` block
+    // reaches it — say so before dropping it.
+    warnUnusedConfiguration({
+        provider: 'vertexai',
+        configuration: llmConfig.configuration,
+        consumedPaths: [],
+        reason: NATIVE_CLIENT_REASON,
+        guidance: 'ChatGoogle talks to Vertex AI through its own client instead: set "customHeaders", ' +
+            '"endpoint", "apiVersion" or "location" as top-level fields of the "llm" block beside ' +
+            '"model".',
+    });
+    // GS2-58: normalise every tool's JSON-Schema at the ChatGoogle boundary so Gemini's OpenAPI-3.0
+    // subset accepts built-in, custom, and MCP tools alike (see geminiSchemaSanitizer).
+    // CFG-33: Vertex serves the same Gemini models through the same ChatGoogle class, so it has the
+    // same silently-discarded thinking; ask for the summaries here too (see geminiThinking).
+    return applyGeminiThoughtSummaries(applyGeminiToolSchemaSanitizer(new ChatGoogle(configFields)));
 }
 //# sourceMappingURL=vertexai.js.map

package/dist/providers/vertexai.js.map

@@ -1 +1 @@
-{"version":3,"file":"vertexai.js","sourceRoot":"","sources":["../../src/providers/vertexai.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AACH,OAAO,EAAE,cAAc,EAAE,MAAM,4BAA4B,CAAC;AAI5D,OAAO,EAAE,2BAA2B,EAAE,MAAM,yBAAyB,CAAC;AACtE,OAAO,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,MAAM,kCAAkC,CAAC;AAEnG,MAAM,UAAU,IAAI,CAAC,cAAsB,EAAE,KAAK,GAAG,KAAK,EAAE,KAAc;IACxE,yDAAyD;IACzD,IAAI,CAAC,cAAc,CAAC,QAAQ,CAAC,OAAO,CAAC,EAAE,CAAC;QACtC,MAAM,IAAI,KAAK,CAAC,gCAAgC,CAAC,CAAC;IACpD,CAAC;IAED,2BAA2B,CAAC,cAAc,EAAE,sBAAsB,CAAC,UAAU,EAAE,KAAK,CAAC,EAAE,KAAK,CAAC,CAAC;IAC9F,cAAc,CACZ,+GAA+G,CAChH,CAAC;AACJ,CAAC;AAED,mEAAmE;AACnE,MAAM,CAAC,KAAK,UAAU,iBAAiB,CACrC,SAAmF;IAEnF,MAAM,EAAE,UAAU,EAAE,GAAG,MAAM,MAAM,CAAC,wBAAwB,CAAC,CAAC;IAC9D,MAAM,YAAY,GAAG;QACnB,GAAG,SAAS;QACZ,KAAK,EAAE,SAAS,CAAC,KAAK,IAAI,uBAAuB,CAAC,UAAU,CAAC;QAC7D,QAAQ,EAAE,IAAI;KACf,CAAC;IACF,OAAO,YAAY,CAAC,IAAI,CAAC;IACzB,OAAO,YAAY,CAAC,yBAAyB,CAAC;IAC9C,OAAO,IAAI,UAAU,CAAC,YAAY,CAAC,CAAC;AACtC,CAAC"}
+{"version":3,"file":"vertexai.js","sourceRoot":"","sources":["../../src/providers/vertexai.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AACH,OAAO,EAAE,cAAc,EAAE,MAAM,4BAA4B,CAAC;AAI5D,OAAO,EAAE,2BAA2B,EAAE,MAAM,yBAAyB,CAAC;AACtE,OAAO,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,MAAM,kCAAkC,CAAC;AACnG,OAAO,EAAE,8BAA8B,EAAE,MAAM,yCAAyC,CAAC;AACzF,OAAO,EAAE,2BAA2B,EAAE,MAAM,kCAAkC,CAAC;AAC/E,OAAO,EACL,oBAAoB,EACpB,uBAAuB,GACxB,MAAM,4CAA4C,CAAC;AAEpD,MAAM,UAAU,IAAI,CAAC,cAAsB,EAAE,KAAK,GAAG,KAAK,EAAE,KAAc;IACxE,yDAAyD;IACzD,IAAI,CAAC,cAAc,CAAC,QAAQ,CAAC,OAAO,CAAC,EAAE,CAAC;QACtC,MAAM,IAAI,KAAK,CAAC,gCAAgC,CAAC,CAAC;IACpD,CAAC;IAED,2BAA2B,CAAC,cAAc,EAAE,sBAAsB,CAAC,UAAU,EAAE,KAAK,CAAC,EAAE,KAAK,CAAC,CAAC;IAC9F,cAAc,CACZ,+GAA+G,CAChH,CAAC;AACJ,CAAC;AAED,mEAAmE;AACnE,MAAM,CAAC,KAAK,UAAU,iBAAiB,CACrC,SAAmF;IAEnF,MAAM,EAAE,UAAU,EAAE,GAAG,MAAM,MAAM,CAAC,wBAAwB,CAAC,CAAC;IAC9D,MAAM,YAAY,GAAG;QACnB,GAAG,SAAS;QACZ,KAAK,EAAE,SAAS,CAAC,KAAK,IAAI,uBAAuB,CAAC,UAAU,CAAC;QAC7D,QAAQ,EAAE,IAAI;KACf,CAAC;IACF,OAAO,YAAY,CAAC,IAAI,CAAC;IACzB,OAAO,YAAY,CAAC,yBAAyB,CAAC;IAC9C,4FAA4F;IAC5F,0CAA0C;IAC1C,uBAAuB,CAAC;QACtB,QAAQ,EAAE,UAAU;QACpB,aAAa,EAAG,SAAyC,CAAC,aAAa;QACvE,aAAa,EAAE,EAAE;QACjB,MAAM,EAAE,oBAAoB;QAC5B,QAAQ,EACN,qFAAqF;YACrF,uFAAuF;YACvF,UAAU;KACb,CAAC,CAAC;IACH,gGAAgG;IAChG,oFAAoF;IACpF,gGAAgG;IAChG,yFAAyF;IACzF,OAAO,2BAA2B,CAAC,8BAA8B,CAAC,IAAI,UAAU,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC;AACnG,CAAC"}

package/dist/providers/xai.js

@@ -2,11 +2,31 @@ import { displayWarning } from '#src/utils/consoleUtils.js';
 import { env } from '#src/utils/systemUtils.js';
 import { writeConfigFileWithMessages } from '#src/utils/fileUtils.js';
 import { buildInitConfigContent, getCuratedFallbackModel } from '#src/providers/modelDiscovery.js';
+import { warnUnusedConfiguration } from '#src/providers/configurationPassthrough.js';
 // Function to process JSON config and create XAI LLM instance
 export async function processJsonConfig(llmConfig) {
     const { ChatXAI } = await import('@langchain/xai');
     // Use config value if available, otherwise use the environment variable
     const apiKey = llmConfig.apiKey || env.XAI_API_KEY;
+    // `ChatXAI` DOES build an OpenAI client, and still consumes none of the user's block: its
+    // constructor passes `configuration: { baseURL: fields.baseURL ?? <xAI default> }` to `super`,
+    // replacing whatever was set here. So the reason clause is xai's own — the shared native-client
+    // sentence would be a false explanation — and the replacements are the top-level fields that were
+    // measured to reach the client: `baseURL` (a declared `ChatXAIInput` field) and `timeout` (read
+    // by the OpenAI base class from the same spread). Headers reach it by no route at all.
+    warnUnusedConfiguration({
+        provider: 'xai',
+        configuration: llmConfig.configuration,
+        consumedPaths: [],
+        reason: 'builds its OpenAI client with a "configuration" block of its own, replacing anything set here',
+        // The escape hatch is only followable with the key hint: the "openai" provider reads
+        // OPENAI_API_KEY, so a user who moves to it for the headers alone hits a missing-key error.
+        guidance: 'Set "baseURL" or "timeout" as top-level fields of the "llm" block instead, beside "model". ' +
+            'Extra headers have no home on this provider at all — to send them, use the "openai" ' +
+            'provider with "configuration.baseURL" pointed at the xAI endpoint, and set ' +
+            '"apiKeyEnvironmentVariable" to XAI_API_KEY there, since that provider otherwise reads ' +
+            'OPENAI_API_KEY.',
+    });
     return new ChatXAI({
         ...llmConfig,
         apiKey,

package/dist/providers/xai.js.map

@@ -1 +1 @@
-{"version":3,"file":"xai.js","sourceRoot":"","sources":["../../src/providers/xai.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,cAAc,EAAE,MAAM,4BAA4B,CAAC;AAC5D,OAAO,EAAE,GAAG,EAAE,MAAM,2BAA2B,CAAC;AAOhD,OAAO,EAAE,2BAA2B,EAAE,MAAM,yBAAyB,CAAC;AACtE,OAAO,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,MAAM,kCAAkC,CAAC;AAEnG,8DAA8D;AAC9D,MAAM,CAAC,KAAK,UAAU,iBAAiB,CACrC,SAA6C;IAE7C,MAAM,EAAE,OAAO,EAAE,GAAG,MAAM,MAAM,CAAC,gBAAgB,CAAC,CAAC;IACnD,wEAAwE;IACxE,MAAM,MAAM,GAAG,SAAS,CAAC,MAAM,IAAI,GAAG,CAAC,WAAW,CAAC;IACnD,OAAO,IAAI,OAAO,CAAC;QACjB,GAAG,SAAS;QACZ,MAAM;QACN,KAAK,EAAE,SAAS,CAAC,KAAK,IAAI,uBAAuB,CAAC,KAAK,CAAC;KACzD,CAAC,CAAC;AACL,CAAC;AAED,MAAM,UAAU,IAAI,CAAC,cAAsB,EAAE,KAAK,GAAG,KAAK,EAAE,KAAc;IACxE,yDAAyD;IACzD,IAAI,CAAC,cAAc,CAAC,QAAQ,CAAC,OAAO,CAAC,EAAE,CAAC;QACtC,MAAM,IAAI,KAAK,CAAC,gCAAgC,CAAC,CAAC;IACpD,CAAC;IAED,2BAA2B,CAAC,cAAc,EAAE,sBAAsB,CAAC,KAAK,EAAE,KAAK,CAAC,EAAE,KAAK,CAAC,CAAC;IACzF,cAAc,CACZ,2BAA2B,cAAc,4BAA4B;QACnE,6CAA6C,CAChD,CAAC;AACJ,CAAC"}
+{"version":3,"file":"xai.js","sourceRoot":"","sources":["../../src/providers/xai.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,cAAc,EAAE,MAAM,4BAA4B,CAAC;AAC5D,OAAO,EAAE,GAAG,EAAE,MAAM,2BAA2B,CAAC;AAOhD,OAAO,EAAE,2BAA2B,EAAE,MAAM,yBAAyB,CAAC;AACtE,OAAO,EAAE,sBAAsB,EAAE,uBAAuB,EAAE,MAAM,kCAAkC,CAAC;AACnG,OAAO,EAAE,uBAAuB,EAAE,MAAM,4CAA4C,CAAC;AAErF,8DAA8D;AAC9D,MAAM,CAAC,KAAK,UAAU,iBAAiB,CACrC,SAA6C;IAE7C,MAAM,EAAE,OAAO,EAAE,GAAG,MAAM,MAAM,CAAC,gBAAgB,CAAC,CAAC;IACnD,wEAAwE;IACxE,MAAM,MAAM,GAAG,SAAS,CAAC,MAAM,IAAI,GAAG,CAAC,WAAW,CAAC;IACnD,0FAA0F;IAC1F,+FAA+F;IAC/F,gGAAgG;IAChG,kGAAkG;IAClG,gGAAgG;IAChG,uFAAuF;IACvF,uBAAuB,CAAC;QACtB,QAAQ,EAAE,KAAK;QACf,aAAa,EAAG,SAAyC,CAAC,aAAa;QACvE,aAAa,EAAE,EAAE;QACjB,MAAM,EACJ,+FAA+F;QACjG,qFAAqF;QACrF,4FAA4F;QAC5F,QAAQ,EACN,6FAA6F;YAC7F,sFAAsF;YACtF,6EAA6E;YAC7E,wFAAwF;YACxF,iBAAiB;KACpB,CAAC,CAAC;IACH,OAAO,IAAI,OAAO,CAAC;QACjB,GAAG,SAAS;QACZ,MAAM;QACN,KAAK,EAAE,SAAS,CAAC,KAAK,IAAI,uBAAuB,CAAC,KAAK,CAAC;KACzD,CAAC,CAAC;AACL,CAAC;AAED,MAAM,UAAU,IAAI,CAAC,cAAsB,EAAE,KAAK,GAAG,KAAK,EAAE,KAAc;IACxE,yDAAyD;IACzD,IAAI,CAAC,cAAc,CAAC,QAAQ,CAAC,OAAO,CAAC,EAAE,CAAC;QACtC,MAAM,IAAI,KAAK,CAAC,gCAAgC,CAAC,CAAC;IACpD,CAAC;IAED,2BAA2B,CAAC,cAAc,EAAE,sBAAsB,CAAC,KAAK,EAAE,KAAK,CAAC,EAAE,KAAK,CAAC,CAAC;IACzF,cAAc,CACZ,2BAA2B,cAAc,4BAA4B;QACnE,6CAA6C,CAChD,CAAC;AACJ,CAAC"}

package/dist/runtime/askStructured.d.ts

@@ -0,0 +1,105 @@
+/**
+ * @module runtime/askStructured
+ *
+ * A reusable, non-agentic "ask the LLM and get a schema-validated object back" primitive — the
+ * structured-output half the (later) `gth workflow` host calls. Deliberately mirrors the
+ * *mechanism* of the structured-evaluation calls ({@link judgeEvalCase} — the EVAL GRADER in
+ * `@gaunt-sloth/batch`'s `judge.ts` — and its in-core sibling {@link rateShellCommand}, the
+ * approvals AI rater in `core/shell/rater.ts`):
+ * `model.withStructuredOutput(schema)` for a single structured call, `.invoke([SystemMessage,
+ * HumanMessage])` raced against a wall-clock timeout via `Promise.race`, a defensive `safeParse`
+ * re-validation, `clearTimeout` in `finally`, and — crucially — it **never throws**, returning a
+ * failure object instead.
+ *
+ * Differences from those two: this one is **generic** over the Zod schema and takes the
+ * system/user strings from the caller (they hard-code a schema and build a rubric/safety
+ * prompt), and it reads the model from `config.llm` (like `runSingleShot`/`rateShellCommand`),
+ * so the workflow host can hand it the resolved {@link GthConfig} directly. `judgeEvalCase` could
+ * later be refactored to delegate to this primitive — out of scope here.
+ */
+import * as z from 'zod';
+import type { GthConfig } from '#src/config.js';
+/**
+ * Default wall-clock budget (ms) for the structured LLM call — same value as their
+ * `EVAL_JUDGE_DEFAULT_TIMEOUT_MS` / `JUDGE_DEFAULT_TIMEOUT_MS`, kept as this module's own constant
+ * since the primitive is conceptually independent of them.
+ */
+export declare const ASK_STRUCTURED_DEFAULT_TIMEOUT_MS = 30000;
+/**
+ * The exact `error` text {@link askStructured} returns when its wall-clock budget fires — i.e. the
+ * provider produced **no answer at all** within the budget.
+ *
+ * Exported as a builder rather than left as an inline literal because the three failure classes here
+ * (no response · an answer that does not fit the schema · a call that failed outright) are the
+ * question a reader of a red run has to answer first, and they are only distinguishable by this text.
+ * A caller that re-types the sentence to recognise it stops recognising it the day the wording moves,
+ * and silently re-files a stall as a rejection — which is precisely the confusion this exists to end.
+ */
+export declare function structuredCallTimedOutError(timeoutMs: number): string;
+/**
+ * The exact `error` text {@link askStructured} returns when the provider **did** answer and the
+ * answer failed the caller's own schema. The counterpart of {@link structuredCallTimedOutError};
+ * see that doc for why both are exported rather than written inline.
+ */
+export declare const STRUCTURED_CALL_UNPARSEABLE_ERROR = "Model returned unparseable output.";
+/**
+ * Which of {@link askStructured}'s three failure paths a result came from.
+ *
+ * - `timeout` — the budget fired and the provider had returned nothing at all.
+ * - `unparseable` — the provider answered, and the answer failed the caller's schema.
+ * - `call-failed` — the call threw before any answer arrived: a provider rejection (OpenAI's 400 on
+ *   a strict `json_schema` is the canonical one), an auth failure, a transport failure.
+ */
+export type StructuredFailureKind = 'timeout' | 'unparseable' | 'call-failed';
+/**
+ * Name which failure happened, given the `error` string and the budget the call was made with.
+ *
+ * The single place the three classes are told apart. A stall and a rejection are the two failures a
+ * reader has to distinguish first and the two that look most alike from the outside — a red that
+ * cannot say which happened sends the reader to re-run a real defect, or to investigate a provider
+ * hiccup. Every caller that wants to say which one it was calls this rather than matching the prose
+ * itself, so the wording can only be got wrong in one place, and that place has a test.
+ *
+ * @param error The `error` from a `{ ok: false }` {@link AskStructuredResult}.
+ * @param timeoutMs The budget passed to that same {@link askStructured} call — the timeout text
+ *   carries it, so a different value here reports a genuine timeout as `call-failed`.
+ */
+export declare function classifyStructuredFailure(error: string, timeoutMs: number): StructuredFailureKind;
+/** Inputs to {@link askStructured}. The caller supplies the model (via config), the two message
+ * texts, and an optional timeout — the Zod schema is a separate positional argument so `<T>` can
+ * be inferred from it. */
+export interface AskStructuredOptions {
+    config: GthConfig;
+    /** System-message text (instructions). May be empty. */
+    system: string;
+    /** Human-message text (the actual content/question). */
+    user: string;
+    /** Wall-clock budget in ms. Default {@link ASK_STRUCTURED_DEFAULT_TIMEOUT_MS} (30_000). */
+    timeoutMs?: number;
+}
+/** Discriminated result of {@link askStructured}: the parsed value on success, an error string on
+ * any failure (unusable model, timeout, unparseable output, or a thrown error). Never throws. */
+export type AskStructuredResult<T> = {
+    ok: true;
+    value: T;
+} | {
+    ok: false;
+    error: string;
+};
+/**
+ * Ask the configured model for a single schema-validated object — a non-agentic structured-output
+ * call that mirrors their mechanism (see the module doc) and never throws.
+ *
+ * - No usable model (`config.llm` missing or lacking `withStructuredOutput`) →
+ *   `{ ok: false, error: 'No usable model configured.' }`.
+ * - Timeout → `ok: false` with the error {@link structuredCallTimedOutError} builds for that budget.
+ * - Output that fails `schema.safeParse` → `{ ok: false, error: {@link STRUCTURED_CALL_UNPARSEABLE_ERROR} }`.
+ * - Any thrown error → `{ ok: false, error: <message> }` — a provider rejection (e.g. OpenAI's 400 on
+ *   a strict `json_schema`), an auth or transport failure. Distinct from the two above BY CODE PATH,
+ *   so a caller can classify a failure by comparing against those two exported texts.
+ * - Success → `{ ok: true, value }` with the parsed data.
+ *
+ * @param schema The Zod schema the model output must satisfy; `<T>` is inferred from it.
+ * @param opts The model (via `config.llm`), the system/user message texts, and an optional timeout.
+ */
+export declare function askStructured<T>(schema: z.ZodType<T>, opts: AskStructuredOptions): Promise<AskStructuredResult<T>>;