feat(ai): add reranker touchpoint to LiteLLM proxy recipe (takeover of #2455)

LiteLLM normalizes Cohere/Voyage/Jina rerank backends to the wire shape
gateway.rerank() already speaks, so a reranker touchpoint on the litellm
recipe makes any proxied rerank model reachable via
`search.reranker.model litellm:<model>` with no adapter.

Repairs from the original PR:
- path is the LEAF '/rerank' (not '/v1/rerank'): LiteLLM serves both
  /rerank and /v1/rerank, and the recipe's setup_hint allows
  LITELLM_BASE_URL with or without the /v1 suffix — pinning '/v1/rerank'
  doubled to /v1/v1/rerank (404) on /v1-suffixed bases.
- setup_hint appends the rerank guidance to master's current line instead
  of replacing it with a stale pre-/v1-suffix version.
- cost_per_1m_tokens_usd stays undefined (pricing-unknown), matching the
  recipe's embedding/chat touchpoints and budget-tracker's deliberate
  litellm exclusion from the free-provider sets (a proxy can front a paid
  provider; the touchpoint field isn't consumed by rerank pricing anyway).

Test drives gateway.rerank()'s real URL builder via the stubbed transport
for both base-URL forms; the /v1-suffixed case fails with the original
PR's path.

Co-authored-by: ozp <ozp@users.noreply.github.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

src/commands/onboard.ts

@@ -142,16 +142,12 @@ export async function runOnboard(engine: BrainEngine, args: string[]): Promise<v
 
   // --auto path: runs through the T2 library orchestrator. Hooks emit CLI
   // progress to stderr; the final result lands as JSON on stdout (or human
-  // summary). extraRemediations (gathered above from runAllOnboardChecks)
-  // is threaded into the runner so the onboard-check remediations
-  // (extract-ner, extract-timeline-from-meetings, etc.) reach the planner
-  // — the same wiring the --check path uses above.
+  // summary).
   const result = await runRemediation(
     engine,
     {
       targetScore,
       maxUsd,
-      extraRemediations,
       // --auto --yes opts into the prompt_required tier too; library
       // doesn't distinguish auto_apply vs prompt_required, it just runs
       // every remediation in the plan. The plan-building side (T12 render)

src/core/ai/recipes/litellm-proxy.ts

@@ -56,6 +56,36 @@ export const litellmProxy: Recipe = {
       cost_per_1m_output_usd: undefined,
       price_last_verified: '2026-06-14',
     },
+    // LiteLLM normalizes Cohere / Voyage / Jina / etc. rerank backends to the
+    // same wire shape gbrain's gateway.rerank() already speaks (the
+    // ZeroEntropy/llama.cpp contract):
+    //   { model, query, documents, top_n } → { results: [{ index, relevance_score }] }
+    // So any rerank model the user registers in their LiteLLM config is
+    // reachable via `gbrain config set search.reranker.model litellm:<model>`
+    // with no request/response adapter — same as embeddings ride the proxy.
+    reranker: {
+      models: [], // user-provided; whatever rerank models the proxy serves
+      // No canonical default — the proxy defines its own model ids. The user
+      // sets search.reranker.model explicitly (mirrors the embedding
+      // touchpoint's user_provided_models contract).
+      default_model: '',
+      // The proxied backend bills (Cohere/Voyage/…); pricing-unknown is the
+      // honest state — same stance as this recipe's embedding/chat
+      // touchpoints and budget-tracker's deliberate litellm exclusion from
+      // the free-provider sets.
+      cost_per_1m_tokens_usd: undefined,
+      price_last_verified: '2026-06-27',
+      max_payload_bytes: 5_000_000,
+      // LEAF path only (matches llama-server-reranker's convention). LiteLLM
+      // serves both `/rerank` and `/v1/rerank`, and LITELLM_BASE_URL may be
+      // set with or without the `/v1` suffix (the setup_hint allows both), so
+      // the leaf form yields a valid route either way:
+      //   http://localhost:4000    + /rerank → /rerank        ✓
+      //   http://localhost:4000/v1 + /rerank → /v1/rerank     ✓
+      // Pinning '/v1/rerank' here would double to /v1/v1/rerank → 404 on
+      // /v1-suffixed bases.
+      path: '/rerank',
+    },
   },
-  setup_hint: 'Run LiteLLM (https://docs.litellm.ai) in front of any provider; set LITELLM_BASE_URL (include the /v1 suffix if your proxy serves the OpenAI route there, e.g. http://localhost:4000/v1) + pass --embedding-model litellm:<model> and --embedding-dimensions <N>.',
+  setup_hint: 'Run LiteLLM (https://docs.litellm.ai) in front of any provider; set LITELLM_BASE_URL (include the /v1 suffix if your proxy serves the OpenAI route there, e.g. http://localhost:4000/v1) + pass --embedding-model litellm:<model> and --embedding-dimensions <N>. For rerank: register a rerank model in LiteLLM and set search.reranker.model litellm:<model-name>.',
 };

src/core/operations.ts

@@ -4957,7 +4957,7 @@ const run_onboard: Operation = {
     // typo, the underlying queue.add would reject. Defense-in-depth.
     const result = await runRemediation(
       ctx.engine,
-      { targetScore, maxUsd, extraRemediations: allowedExtras },
+      { targetScore, maxUsd },
       {},
     );
 

src/core/remediation/run.ts

@@ -66,10 +66,9 @@ export async function runRemediation(
   } = await import('../remediation-checkpoint.ts');
 
   const ctx = await loadRecommendationContext(engine);
-  const extraRemediations = opts.extraRemediations ?? [];
 
   // Pre-flight ceiling check via the shared plan computation.
-  const initialPlan = await computeRemediationPlan(engine, { targetScore, extraRemediations });
+  const initialPlan = await computeRemediationPlan(engine, { targetScore });
   if (initialPlan.target_unreachable) {
     hooks.onTargetUnreachable?.(targetScore, initialPlan.max_reachable_score);
     return {
@@ -88,7 +87,7 @@ export async function runRemediation(
   }
 
   const initialHealth = await engine.getHealth();
-  let recs: RemediationStep[] = computeRecommendations(initialHealth, ctx, extraRemediations)
+  let recs: RemediationStep[] = computeRecommendations(initialHealth, ctx)
     .filter((r) => r.status === 'remediable');
   if (recs.length === 0) {
     hooks.onNothingToDo?.(initialHealth.brain_score, targetScore);
@@ -306,13 +305,7 @@ export async function runRemediation(
       // steps with bumped retry suffix (D1).
       if (recs.length === 0 || stepCount >= maxJobs) break;
       const freshHealth = await engine.getHealth();
-      // Extras carry a static status:'remediable' — a fresh health snapshot
-      // never ages them out the way health-derived steps drop. Filter out
-      // ids this run already processed (any terminal status), or the recheck
-      // would resubmit completed extras every iteration, forever.
-      const processedIds = new Set(submitted.map((s) => s.id));
-      const pendingExtras = extraRemediations.filter((r) => !processedIds.has(r.id));
-      recs = computeRecommendations(freshHealth, ctx, pendingExtras).filter((r) => r.status === 'remediable');
+      recs = computeRecommendations(freshHealth, ctx).filter((r) => r.status === 'remediable');
     }
   };
 

src/core/remediation/types.ts

@@ -63,16 +63,6 @@ export interface RemediationOpts {
   resumePlanHash?: string;
   /** Whether to attempt resume at all (default false). */
   resume?: boolean;
-  /**
-   * Caller-supplied RemediationStep entries threaded into the planner.
-   * Mirrors RemediationPlanOpts.extraRemediations so onboard's --apply
-   * --auto path (and MCP run_onboard auto modes) forward the same
-   * onboard-check remediations the --check path already passes through
-   * computeRemediationPlan. Without this the runner saw only generic
-   * brain_score remediations and reported "Nothing to do" whenever the
-   * only applicable work was an extra (e.g. extract-ner).
-   */
-  extraRemediations?: RemediationStep[];
 }
 
 /**

test/ai/recipe-litellm-reranker.test.ts

@@ -0,0 +1,88 @@
+/**
+ * litellm-proxy reranker touchpoint smoke.
+ *
+ * Sibling of recipe-llama-server-reranker.test.ts. Pins the reranker
+ * touchpoint on the LiteLLM proxy recipe so:
+ *  - the touchpoint exists with the LEAF '/rerank' path (LiteLLM serves both
+ *    /rerank and /v1/rerank, so the leaf form is valid whether or not the
+ *    user's LITELLM_BASE_URL carries the /v1 suffix the setup_hint allows)
+ *  - a /v1-suffixed base URL does NOT produce /v1/v1/rerank (the original
+ *    community PR pinned '/v1/rerank' which 404s on /v1-suffixed bases)
+ *  - models: [] (user-provided; proxy defines the model ids)
+ *  - pricing stays undefined (proxy can front a paid provider — same honest
+ *    pricing-unknown stance as the embedding/chat touchpoints)
+ *
+ * The gateway.rerank() URL tests drive the real URL builder via the stubbed
+ * transport (same seam as test/ai/rerank.test.ts).
+ */
+
+import { describe, expect, test, afterEach } from 'bun:test';
+import { getRecipe } from '../../src/core/ai/recipes/index.ts';
+import {
+  configureGateway,
+  resetGateway,
+  rerank,
+  __setRerankTransportForTests,
+} from '../../src/core/ai/gateway.ts';
+
+afterEach(() => {
+  __setRerankTransportForTests(null);
+  resetGateway();
+});
+
+describe('recipe: litellm reranker touchpoint', () => {
+  test('declares reranker touchpoint with leaf /rerank path', () => {
+    const r = getRecipe('litellm')!;
+    const tp = r.touchpoints.reranker;
+    expect(tp).toBeDefined();
+    expect(tp!.path).toBe('/rerank');
+    expect(tp!.max_payload_bytes).toBe(5_000_000);
+  });
+
+  test('reranker touchpoint uses empty models[] for user-provided model ids', () => {
+    const r = getRecipe('litellm')!;
+    expect(r.touchpoints.reranker!.models).toEqual([]);
+  });
+
+  test('pricing stays undefined — proxy can front a paid provider', () => {
+    const r = getRecipe('litellm')!;
+    expect(r.touchpoints.reranker!.cost_per_1m_tokens_usd).toBeUndefined();
+  });
+
+  test('setup_hint keeps the /v1-suffix guidance AND mentions rerank', () => {
+    const r = getRecipe('litellm')!;
+    expect(r.setup_hint).toMatch(/\/v1 suffix/);
+    expect(r.setup_hint).toMatch(/search\.reranker\.model litellm:/);
+  });
+});
+
+describe('gateway.rerank() URL via litellm recipe', () => {
+  async function capturedRerankUrl(baseUrl?: string): Promise<string> {
+    configureGateway({
+      reranker_model: 'litellm:my-reranker',
+      env: {},
+      ...(baseUrl ? { base_urls: { litellm: baseUrl } } : {}),
+    });
+    let capturedUrl = '';
+    __setRerankTransportForTests(async (url) => {
+      capturedUrl = url;
+      return new Response(
+        JSON.stringify({ results: [{ index: 0, relevance_score: 0.9 }] }),
+        { status: 200, headers: { 'content-type': 'application/json' } },
+      );
+    });
+    await rerank({ query: 'q', documents: ['d'] });
+    return capturedUrl;
+  }
+
+  test('default base (no /v1 suffix) → /rerank', async () => {
+    const url = await capturedRerankUrl();
+    expect(url).toBe('http://localhost:4000/rerank');
+  });
+
+  test('/v1-suffixed base → /v1/rerank, NOT /v1/v1/rerank', async () => {
+    const url = await capturedRerankUrl('http://localhost:4000/v1');
+    expect(url).toBe('http://localhost:4000/v1/rerank');
+    expect(url).not.toContain('/v1/v1/');
+  });
+});

test/remediation-run-extras.serial.test.ts

@@ -1,108 +0,0 @@
-// test/remediation-run-extras.serial.test.ts
-// Regression for PR #2161 takeover: `gbrain onboard --apply --auto` dropped
-// onboard-check extraRemediations. Two distinct halves of the bug:
-//   1. runRemediation built the pre-flight plan + initial recs WITHOUT the
-//      extras, so an extras-only plan reported "Nothing to do".
-//   2. The D7 mid-run recheck rebuilt recs WITHOUT the extras after every
-//      completed step, so with 2+ plannable steps all remaining extras were
-//      dropped after step 1. The recheck must also filter out extras this
-//      run already processed — extras carry static status:'remediable', so
-//      unfiltered threading would resubmit completed extras forever.
-//
-// SERIAL: mock.module (queue + wait-for-completion stubs, R2) + GBRAIN_HOME
-// env mutation so checkpoint files land in a tmpdir, not ~/.gbrain.
-
-import { describe, expect, test, beforeAll, afterAll, mock } from 'bun:test';
-import { mkdtempSync, rmSync } from 'node:fs';
-import { tmpdir } from 'node:os';
-import { join } from 'node:path';
-import { PGLiteEngine } from '../src/core/pglite-engine.ts';
-import { makeRemediationStep } from '../src/core/remediation-step.ts';
-
-// Stub the Minion queue: every submitted job is immediately 'completed'.
-// runRemediation only calls queue.add + waitForCompletion(queue, id).
-let nextJobId = 1;
-const submittedJobs: Array<{ name: string }> = [];
-mock.module('../src/core/minions/queue.ts', () => ({
-  MinionQueue: class {
-    async add(name: string) {
-      submittedJobs.push({ name });
-      return { id: nextJobId++, status: 'completed' };
-    }
-  },
-}));
-mock.module('../src/core/minions/wait-for-completion.ts', () => ({
-  waitForCompletion: async () => ({ status: 'completed' }),
-}));
-
-let engine: PGLiteEngine;
-let home: string;
-const prevHome = process.env.GBRAIN_HOME;
-
-beforeAll(async () => {
-  home = mkdtempSync(join(tmpdir(), 'gbrain-remextras-'));
-  process.env.GBRAIN_HOME = home;
-  engine = new PGLiteEngine();
-  await engine.connect({});
-  await engine.initSchema();
-}, 120_000);
-
-afterAll(async () => {
-  await engine.disconnect();
-  if (prevHome === undefined) delete process.env.GBRAIN_HOME;
-  else process.env.GBRAIN_HOME = prevHome;
-  rmSync(home, { recursive: true, force: true });
-});
-
-function extra(id: string, job: string) {
-  return makeRemediationStep({
-    id,
-    job,
-    params: {},
-    severity: 'medium',
-    est_seconds: 5,
-    est_usd_cost: 0,
-    rationale: 'synthetic onboard-check extra',
-    status: 'remediable',
-  });
-}
-
-describe('runRemediation extraRemediations threading', () => {
-  test('extras-only plan runs BOTH extras and terminates (no Nothing-to-do, no resubmit loop)', async () => {
-    // Empty PGLite brain → zero health-derived recommendations. Without the
-    // fix, half 1 makes this run return submitted: [] via onNothingToDo.
-    // With only half 1 (the original PR #2161 diff), the mid-run recheck
-    // drops the second extra after step 1 — submitted has 1 entry, not 2.
-    const { runRemediation } = await import('../src/core/remediation/run.ts');
-    let nothingToDo = false;
-    const result = await runRemediation(
-      engine,
-      {
-        targetScore: 1,
-        extraRemediations: [
-          extra('onboard.extract_ner', 'extract-ner'),
-          extra('onboard.extract_timeline', 'extract-timeline-from-meetings'),
-        ],
-        // Safety bound: an unfiltered recheck would resubmit completed
-        // extras forever; maxJobs turns that regression into a fast fail
-        // (extra count > 1 below) instead of a hung test.
-        maxJobs: 5,
-      },
-      { onNothingToDo: () => { nothingToDo = true; } },
-    );
-
-    expect(nothingToDo).toBe(false);
-    const ids = result.submitted.map((s) => s.id);
-    expect(ids).toContain('onboard.extract_ner');
-    expect(ids).toContain('onboard.extract_timeline');
-    // Each extra ran exactly once — the recheck must not re-plan extras the
-    // run already processed.
-    expect(ids.filter((i) => i === 'onboard.extract_ner').length).toBe(1);
-    expect(ids.filter((i) => i === 'onboard.extract_timeline').length).toBe(1);
-    expect(result.submitted.every((s) => s.status === 'completed')).toBe(true);
-    expect(submittedJobs.map((j) => j.name).sort()).toEqual([
-      'extract-ner',
-      'extract-timeline-from-meetings',
-    ]);
-  });
-});