feat(llm): recognize orcarouter/ gateway model refs alongside openrouter

README.md

@@ -683,6 +683,22 @@ Notes for `llm.auth: "oauth"`:
 - `auth login` snapshots the previous api-key `llm` config next to the OAuth file, and `auth logout` restores that snapshot when available.
 - Switching from `api-key` to `oauth` does not automatically carry over `llm.baseURL`. Set it manually in OAuth mode only when you intentionally want a custom ChatGPT/Codex-compatible backend.
 
+**Gateway (OpenAI-compatible) `llm` config — e.g. [OrcaRouter](https://www.orcarouter.ai):**
+
+The plugin's LLM client accepts any OpenAI-compatible gateway via `llm.baseURL` + `llm.apiKey`. When pointing at a gateway, use its namespaced model id — for OrcaRouter that is `orcarouter/auto` (the adaptive auto-router) or a provider-prefixed id such as `anthropic/claude-sonnet-4.6`. A bare model name is rejected by OrcaRouter, so keep the `orcarouter/` or vendor prefix intact:
+
+```json
+{
+  "llm": {
+    "auth": "api-key",
+    "apiKey": "${ORCAROUTER_API_KEY}",
+    "baseURL": "https://api.orcarouter.ai/v1",
+    "model": "orcarouter/auto",
+    "timeoutMs": 30000
+  }
+}
+```
+
 </details>
 
 <details>

dist/index.js

@@ -763,6 +763,8 @@ export function inferProviderFromBaseURL(baseURL) {
             return "openai";
         if (hostname.endsWith(".anthropic.com"))
             return "anthropic";
+        if (hostname.endsWith(".orcarouter.ai"))
+            return "orcarouter";
         return undefined;
     }
     catch {

dist/src/admission-control.js

@@ -480,17 +480,31 @@ function parseBatchUtilityResponse(response, expectedCount) {
  * Strip that literal "openrouter/" prefix so both forms reach this plugin's
  * direct client correctly; a bare "<vendor>/<model>" or an "@preset/<name>"
  * alias already work against OpenRouter unchanged, so they pass through.
+ *
+ * The same mirror applies to the "orcarouter/" gateway prefix: a namespaced
+ * "orcarouter/<vendor>/<model>" id strips to "<vendor>/<model>", while the
+ * auto-router "orcarouter/auto" keeps its prefix (OrcaRouter rejects the
+ * bare "auto" id with 503 model_not_found).
  */
 export function normalizeAdmissionModelRef(modelRef) {
     const trimmed = modelRef.trim();
     const idx = trimmed.indexOf("/");
     if (idx <= 0)
         return trimmed;
     const provider = trimmed.slice(0, idx).trim().toLowerCase();
-    if (provider !== "openrouter")
-        return trimmed;
-    const rest = trimmed.slice(idx + 1).trim();
-    return rest || trimmed;
+    if (provider === "openrouter") {
+        const rest = trimmed.slice(idx + 1).trim();
+        return rest || trimmed;
+    }
+    if (provider === "orcarouter") {
+        const rest = trimmed.slice(idx + 1).trim();
+        // OrcaRouter requires a namespaced model id. A remainder that still
+        // carries a "<vendor>/" prefix (e.g. orcarouter/anthropic/claude-...)
+        // can drop the gateway prefix; a bare remainder (e.g. "auto") must keep
+        // "orcarouter/" or OrcaRouter returns 503 model_not_found.
+        return rest.includes("/") ? rest : trimmed;
+    }
+    return trimmed;
 }
 /**
  * Resolves which LLM model an admission call should use, in order:

dist/src/llm-client.js

@@ -7,18 +7,30 @@ import { buildOauthEndpoint, extractOutputTextFromSse, loadOAuthSession, needsRe
 /**
  * Strips a core-style provider prefix (e.g. "openrouter/anthropic/claude-...")
  * down to the bare "<vendor>/<model>" form a direct OpenRouter-compatible API
- * needs. Any other prefix, or a string with no "/", passes through unchanged.
+ * needs. Also recognizes the "orcarouter/" gateway prefix: a namespaced
+ * "orcarouter/<vendor>/<model>" id strips to "<vendor>/<model>", while the
+ * auto-router "orcarouter/auto" keeps its prefix (OrcaRouter rejects the bare
+ * "auto" id). Any other prefix, or a string with no "/", passes through.
  */
 export function normalizeDirectModelRef(modelRef) {
     const trimmed = modelRef.trim();
     const idx = trimmed.indexOf("/");
     if (idx <= 0)
         return trimmed;
     const provider = trimmed.slice(0, idx).trim().toLowerCase();
-    if (provider !== "openrouter")
-        return trimmed;
-    const rest = trimmed.slice(idx + 1).trim();
-    return rest || trimmed;
+    if (provider === "openrouter") {
+        const rest = trimmed.slice(idx + 1).trim();
+        return rest || trimmed;
+    }
+    if (provider === "orcarouter") {
+        const rest = trimmed.slice(idx + 1).trim();
+        // OrcaRouter requires a namespaced model id. A remainder that still
+        // carries a "<vendor>/" prefix (e.g. orcarouter/anthropic/claude-...)
+        // can drop the gateway prefix; a bare remainder (e.g. "auto") must keep
+        // "orcarouter/" or OrcaRouter returns 503 model_not_found.
+        return rest.includes("/") ? rest : trimmed;
+    }
+    return trimmed;
 }
 const DEFAULT_SYSTEM_PROMPT = "You are a memory extraction assistant. Always respond with valid JSON only.";
 /**

index.ts

@@ -1206,6 +1206,7 @@ export function inferProviderFromBaseURL(baseURL: string | undefined): string |
     if (hostname.endsWith(".minimax.io")) return "minimax-portal";
     if (hostname.endsWith(".openai.com")) return "openai";
     if (hostname.endsWith(".anthropic.com")) return "anthropic";
+    if (hostname.endsWith(".orcarouter.ai")) return "orcarouter";
     return undefined;
   } catch {
     return undefined;

src/admission-control.ts

@@ -688,15 +688,30 @@ export type AdmissionLane = "reflection" | "other";
  * Strip that literal "openrouter/" prefix so both forms reach this plugin's
  * direct client correctly; a bare "<vendor>/<model>" or an "@preset/<name>"
  * alias already work against OpenRouter unchanged, so they pass through.
+ *
+ * The same mirror applies to the "orcarouter/" gateway prefix: a namespaced
+ * "orcarouter/<vendor>/<model>" id strips to "<vendor>/<model>", while the
+ * auto-router "orcarouter/auto" keeps its prefix (OrcaRouter rejects the
+ * bare "auto" id with 503 model_not_found).
  */
 export function normalizeAdmissionModelRef(modelRef: string): string {
   const trimmed = modelRef.trim();
   const idx = trimmed.indexOf("/");
   if (idx <= 0) return trimmed;
   const provider = trimmed.slice(0, idx).trim().toLowerCase();
-  if (provider !== "openrouter") return trimmed;
-  const rest = trimmed.slice(idx + 1).trim();
-  return rest || trimmed;
+  if (provider === "openrouter") {
+    const rest = trimmed.slice(idx + 1).trim();
+    return rest || trimmed;
+  }
+  if (provider === "orcarouter") {
+    const rest = trimmed.slice(idx + 1).trim();
+    // OrcaRouter requires a namespaced model id. A remainder that still
+    // carries a "<vendor>/" prefix (e.g. orcarouter/anthropic/claude-...)
+    // can drop the gateway prefix; a bare remainder (e.g. "auto") must keep
+    // "orcarouter/" or OrcaRouter returns 503 model_not_found.
+    return rest.includes("/") ? rest : trimmed;
+  }
+  return trimmed;
 }
 
 /**

src/llm-client.ts

@@ -17,16 +17,29 @@ import {
 /**
  * Strips a core-style provider prefix (e.g. "openrouter/anthropic/claude-...")
  * down to the bare "<vendor>/<model>" form a direct OpenRouter-compatible API
- * needs. Any other prefix, or a string with no "/", passes through unchanged.
+ * needs. Also recognizes the "orcarouter/" gateway prefix: a namespaced
+ * "orcarouter/<vendor>/<model>" id strips to "<vendor>/<model>", while the
+ * auto-router "orcarouter/auto" keeps its prefix (OrcaRouter rejects the bare
+ * "auto" id). Any other prefix, or a string with no "/", passes through.
  */
 export function normalizeDirectModelRef(modelRef: string): string {
   const trimmed = modelRef.trim();
   const idx = trimmed.indexOf("/");
   if (idx <= 0) return trimmed;
   const provider = trimmed.slice(0, idx).trim().toLowerCase();
-  if (provider !== "openrouter") return trimmed;
-  const rest = trimmed.slice(idx + 1).trim();
-  return rest || trimmed;
+  if (provider === "openrouter") {
+    const rest = trimmed.slice(idx + 1).trim();
+    return rest || trimmed;
+  }
+  if (provider === "orcarouter") {
+    const rest = trimmed.slice(idx + 1).trim();
+    // OrcaRouter requires a namespaced model id. A remainder that still
+    // carries a "<vendor>/" prefix (e.g. orcarouter/anthropic/claude-...)
+    // can drop the gateway prefix; a bare remainder (e.g. "auto") must keep
+    // "orcarouter/" or OrcaRouter returns 503 model_not_found.
+    return rest.includes("/") ? rest : trimmed;
+  }
+  return trimmed;
 }
 
 export interface LlmClientConfig {

test/admission-lane-model-affinity.test.mjs

@@ -215,6 +215,44 @@ describe("admission lane model affinity", () => {
     );
   });
 
+  it("normalizes a namespaced orcarouter/<vendor>/<model> reflection model and keeps orcarouter/auto for the direct client", () => {
+    const harness = createPluginApiHarness({
+      resolveRoot: workspaceDir,
+      pluginConfig: baseConfig(workspaceDir, {
+        admissionControl: { enabled: true, modelAffinity: "lane" },
+        memoryReflection: { model: "orcarouter/anthropic/claude-sonnet-4.6" },
+      }),
+    });
+
+    memoryLanceDBProPlugin.register(harness.api);
+
+    assert.ok(
+      requestedModels.includes("anthropic/claude-sonnet-4.6"),
+      "orcarouter/<vendor>/<model> must strip to the bare <vendor>/<model> id a direct OrcaRouter call accepts",
+    );
+    assert.ok(
+      !requestedModels.includes("orcarouter/anthropic/claude-sonnet-4.6"),
+      "the raw core-style orcarouter ref must never reach a direct client",
+    );
+  });
+
+  it("keeps the orcarouter/auto router model prefixed for the direct client", () => {
+    const harness = createPluginApiHarness({
+      resolveRoot: workspaceDir,
+      pluginConfig: baseConfig(workspaceDir, {
+        admissionControl: { enabled: true, modelAffinity: "lane" },
+        memoryReflection: { model: "orcarouter/auto" },
+      }),
+    });
+
+    memoryLanceDBProPlugin.register(harness.api);
+
+    assert.ok(
+      requestedModels.includes("orcarouter/auto"),
+      "OrcaRouter rejects the bare auto id, so orcarouter/auto must reach the direct client intact",
+    );
+  });
+
   it("lets an explicit admissionControl.model override beat lane affinity on every admission lane", () => {
     const harness = createPluginApiHarness({
       resolveRoot: workspaceDir,

test/admission-model-resolution.test.mjs

@@ -101,6 +101,56 @@ describe("resolveAdmissionModel", () => {
     assert.equal(reflection, "anthropic/claude-opus-4-8");
   });
 
+  it("normalizes a core-style orcarouter/<vendor>/<model> reflection model to the bare <vendor>/<model> form the OrcaRouter-direct client needs", () => {
+    const admissionControl = normalizeAdmissionControlConfig({ enabled: true, modelAffinity: "lane" });
+
+    const reflection = resolveAdmissionModel({
+      admissionControl,
+      lane: "reflection",
+      globalModel: "global-model",
+      reflectionModel: "orcarouter/anthropic/claude-sonnet-4.6",
+    });
+
+    assert.equal(reflection, "anthropic/claude-sonnet-4.6");
+  });
+
+  it("keeps the orcarouter/auto router model prefixed (OrcaRouter rejects the bare auto id)", () => {
+    const admissionControl = normalizeAdmissionControlConfig({ enabled: true, modelAffinity: "lane" });
+
+    const reflection = resolveAdmissionModel({
+      admissionControl,
+      lane: "reflection",
+      globalModel: "global-model",
+      reflectionModel: "orcarouter/auto",
+    });
+
+    assert.equal(reflection, "orcarouter/auto");
+  });
+
+  it("normalizes an explicit orcarouter admissionControl.model override the same way as lane-resolved models", () => {
+    const admissionControl = normalizeAdmissionControlConfig({
+      enabled: true,
+      modelAffinity: "lane",
+      model: "orcarouter/anthropic/claude-sonnet-4.6",
+    });
+
+    const other = resolveAdmissionModel({
+      admissionControl,
+      lane: "other",
+      globalModel: "global-model",
+      reflectionModel: "reflection-model",
+    });
+    const reflection = resolveAdmissionModel({
+      admissionControl,
+      lane: "reflection",
+      globalModel: "global-model",
+      reflectionModel: "reflection-model",
+    });
+
+    assert.equal(other, "anthropic/claude-sonnet-4.6");
+    assert.equal(reflection, "anthropic/claude-sonnet-4.6");
+  });
+
   it("passes a bare <vendor>/<model> reflection model through unchanged", () => {
     const admissionControl = normalizeAdmissionControlConfig({ enabled: true, modelAffinity: "lane" });
 

test/infer-provider-from-baseurl.test.mjs

@@ -33,6 +33,11 @@ describe("inferProviderFromBaseURL - PR #713 regression", () => {
       const result = inferProviderFromBaseURL("https://api.anthropic.com");
       assert.strictEqual(result, "anthropic");
     });
+
+    it("baseURL with orcarouter.ai returns orcarouter", () => {
+      const result = inferProviderFromBaseURL("https://api.orcarouter.ai/v1");
+      assert.strictEqual(result, "orcarouter");
+    });
   });
 
   describe("edge cases", () => {

test/llm-api-key-client.test.mjs

@@ -4,7 +4,29 @@ import { afterEach, describe, it } from "node:test";
 import jitiFactory from "jiti";
 
 const jiti = jitiFactory(import.meta.url, { interopDefault: true });
-const { createLlmClient, shouldDisableReasoningForJson, stripReasoningTrace } = jiti("../src/llm-client.ts");
+const { createLlmClient, normalizeDirectModelRef, shouldDisableReasoningForJson, stripReasoningTrace } = jiti("../src/llm-client.ts");
+
+describe("normalizeDirectModelRef", () => {
+  it("strips a core-style openrouter/<vendor>/<model> ref to the bare <vendor>/<model> form", () => {
+    assert.equal(normalizeDirectModelRef("openrouter/anthropic/claude-opus-4-8"), "anthropic/claude-opus-4-8");
+  });
+
+  it("strips a namespaced orcarouter/<vendor>/<model> ref to the bare <vendor>/<model> form", () => {
+    assert.equal(normalizeDirectModelRef("orcarouter/anthropic/claude-sonnet-4.6"), "anthropic/claude-sonnet-4.6");
+  });
+
+  it("keeps orcarouter/auto prefixed (OrcaRouter rejects the bare auto id)", () => {
+    assert.equal(normalizeDirectModelRef("orcarouter/auto"), "orcarouter/auto");
+  });
+
+  it("passes a bare <vendor>/<model> ref through unchanged", () => {
+    assert.equal(normalizeDirectModelRef("anthropic/claude-sonnet-4.6"), "anthropic/claude-sonnet-4.6");
+  });
+
+  it("passes an unrelated provider prefix through unchanged", () => {
+    assert.equal(normalizeDirectModelRef("openai/gpt-4o"), "openai/gpt-4o");
+  });
+});
 
 describe("LLM api-key client", () => {
   let server;