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/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/postgres-engine.ts

@@ -5565,78 +5565,30 @@ export class PostgresEngine implements BrainEngine {
     });
   }
 
-  /**
-   * perf (#1694 by @Omerbahari): process-lifetime config cache. A single
-   * search fires ~85 getConfig() reads (loadConfigWithEngine x2, plus
-   * mode/cache/intent/rerank/graph-signals resolvers). On a remote pooler
-   * each read is a round-trip; serial they dominate query latency and can
-   * push the op handler past cli.ts's 10s disconnect force-exit, truncating
-   * stdout. First read batch-loads the whole `config` table into this Map
-   * (inside the same connRetry posture as the per-key read — #1603/#1891);
-   * setConfig/unsetConfig write through. TTL bounds staleness for
-   * multi-writer processes; GBRAIN_CONFIG_CACHE_TTL_MS=0 disables.
-   * Only present keys are stored — Map.has() distinguishes known-absent.
-   */
-  private _configCache: Map<string, string> | null = null;
-  private _configCacheLoadedAt = 0;
-  private _configCacheLoad: Promise<void> | null = null;
-  private get _configCacheTtlMs(): number {
-    const raw = process.env.GBRAIN_CONFIG_CACHE_TTL_MS;
-    if (raw !== undefined) {
-      const n = parseInt(raw, 10);
-      if (Number.isFinite(n) && n >= 0) return n;
-    }
-    return 30_000;
-  }
-
   async getConfig(key: string): Promise<string | null> {
     // #1603: a transient pooler drop on this read used to throw / fall through
     // to defaults silently — which on remote Postgres surfaces as the wrong
-    // search mode/knobs and empty-stdout queries. Both the batch load and the
-    // cache-off per-key read keep the connRetry reconnect posture.
-    const ttl = this._configCacheTtlMs;
-    if (ttl === 0) {
-      return this.connRetry(async () => {
-        const rows = await this.sql`SELECT value FROM config WHERE key = ${key}`;
-        return rows.length > 0 ? (rows[0].value as string) : null;
-      });
-    }
-    if (this._configCache === null || Date.now() - this._configCacheLoadedAt >= ttl) {
-      // Single-flight: concurrent cold reads share one batch load.
-      this._configCacheLoad ??= this.connRetry(async () => {
-        const rows = await this.sql`SELECT key, value FROM config` as unknown as
-          Array<{ key: string; value: string | null }>;
-        const map = new Map<string, string>();
-        for (const r of rows) if (r.value != null) map.set(r.key, r.value);
-        this._configCache = map;
-        this._configCacheLoadedAt = Date.now();
-      }).finally(() => {
-        this._configCacheLoad = null;
-      });
-      await this._configCacheLoad;
-    }
-    return this._configCache!.has(key) ? this._configCache!.get(key)! : null;
+    // search mode/knobs and empty-stdout queries.
+    return this.connRetry(async () => {
+      const rows = await this.sql`SELECT value FROM config WHERE key = ${key}`;
+      return rows.length > 0 ? (rows[0].value as string) : null;
+    });
   }
 
   async setConfig(key: string, value: string): Promise<void> {
-    await this.connRetry(async () => {
+    return this.connRetry(async () => {
       await this.sql`
         INSERT INTO config (key, value) VALUES (${key}, ${value})
         ON CONFLICT (key) DO UPDATE SET value = EXCLUDED.value
       `;
     });
-    // Write-through so a long-lived process never serves stale config.
-    this._configCache?.set(key, value);
   }
 
   async unsetConfig(key: string): Promise<number> {
-    const count = await this.connRetry(async () => {
+    return this.connRetry(async () => {
       const result = await this.sql`DELETE FROM config WHERE key = ${key}` as unknown as { count: number };
       return result.count ?? 0;
     });
-    // Write-through: known-absent, so the cache doesn't serve a stale value.
-    this._configCache?.delete(key);
-    return count;
   }
 
   async listConfigKeys(prefix: string): Promise<string[]> {

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/e2e/helpers.ts

@@ -29,13 +29,6 @@ if (existsSync(envPath)) {
   }
 }
 
-// E2E suites seed/rewrite the config table via raw SQL and expect engine
-// reads to see it immediately; disable the process-lifetime config cache
-// (#1694) so read semantics match pre-cache behavior. Spawned CLI
-// subprocesses inherit this. Cache semantics are pinned by
-// test/postgres-engine-config-cache.test.ts.
-process.env.GBRAIN_CONFIG_CACHE_TTL_MS ??= '0';
-
 const DATABASE_URL = process.env.DATABASE_URL;
 const FIXTURES_DIR = resolve(import.meta.dir, 'fixtures');
 

test/postgres-engine-config-cache.test.ts

@@ -1,112 +0,0 @@
-/**
- * Process-lifetime config cache (#1694 takeover, by @Omerbahari).
- *
- * A single search fires ~85 getConfig() reads; on a remote pooler each is a
- * round-trip. The first read now batch-loads the whole `config` table into a
- * Map; setConfig/unsetConfig write through; TTL bounds multi-writer
- * staleness; GBRAIN_CONFIG_CACHE_TTL_MS=0 restores per-key reads.
- *
- * Pure: stubs `_sql` with a call-counting fake; no real DB.
- */
-
-import { describe, it, expect } from 'bun:test';
-import { PostgresEngine } from '../src/core/postgres-engine.ts';
-import { withEnv } from './helpers/with-env.ts';
-
-const FAST_RETRY = { maxRetries: 3, delayMs: 1, delayMaxMs: 1, jitter: 'none' as const };
-
-/** Engine whose `sql` records every query's template strings and returns `rows`. */
-function makeEngine(rows: unknown[]) {
-  const e = new PostgresEngine();
-  const calls: string[] = [];
-  (e as unknown as { _connectionStyle: string })._connectionStyle = 'instance';
-  (e as unknown as { _bulkRetryOptsCache: unknown })._bulkRetryOptsCache = FAST_RETRY;
-  (e as unknown as { _sql: unknown })._sql = (strings: TemplateStringsArray) => {
-    calls.push(strings.join('?'));
-    return Promise.resolve(rows);
-  };
-  return { engine: e, calls };
-}
-
-/** Run `fn` with GBRAIN_CONFIG_CACHE_TTL_MS set (or cleared when undefined). */
-const withTtl = (ttl: string | undefined, fn: () => Promise<void>) =>
-  withEnv({ GBRAIN_CONFIG_CACHE_TTL_MS: ttl }, fn);
-
-describe('PostgresEngine config cache (#1694)', () => {
-  it('batch-loads once and serves repeat reads from the cache', () => withTtl(undefined, async () => {
-    const { engine, calls } = makeEngine([
-      { key: 'search.mode', value: 'balanced' },
-      { key: 'embedding_multimodal', value: 'true' },
-    ]);
-    expect(await engine.getConfig('search.mode')).toBe('balanced');
-    expect(await engine.getConfig('embedding_multimodal')).toBe('true');
-    expect(await engine.getConfig('search.mode')).toBe('balanced');
-    // One SELECT total — this is the whole point of the fix.
-    expect(calls.length).toBe(1);
-    expect(calls[0]).toContain('SELECT key, value FROM config');
-  }));
-
-  it('returns null for a known-absent key without an extra round-trip', () => withTtl(undefined, async () => {
-    const { engine, calls } = makeEngine([{ key: 'a', value: '1' }]);
-    expect(await engine.getConfig('missing.key')).toBeNull();
-    expect(await engine.getConfig('missing.key')).toBeNull();
-    expect(calls.length).toBe(1);
-  }));
-
-  it('setConfig writes through so subsequent reads see the new value', () => withTtl(undefined, async () => {
-    const { engine, calls } = makeEngine([{ key: 'k', value: 'old' }]);
-    expect(await engine.getConfig('k')).toBe('old');
-    await engine.setConfig('k', 'new');
-    expect(await engine.getConfig('k')).toBe('new');
-    expect(calls.length).toBe(2); // batch load + upsert; no re-read
-  }));
-
-  it('unsetConfig writes through so subsequent reads see absence', () => withTtl(undefined, async () => {
-    const { engine } = makeEngine([{ key: 'k', value: 'v' }]);
-    expect(await engine.getConfig('k')).toBe('v');
-    await engine.unsetConfig('k');
-    expect(await engine.getConfig('k')).toBeNull();
-  }));
-
-  it('concurrent cold reads share a single batch load (single-flight)', () => withTtl(undefined, async () => {
-    const { engine, calls } = makeEngine([{ key: 'k', value: 'v' }]);
-    const [a, b, c] = await Promise.all([
-      engine.getConfig('k'),
-      engine.getConfig('k'),
-      engine.getConfig('other'),
-    ]);
-    expect([a, b, c]).toEqual(['v', 'v', null]);
-    expect(calls.length).toBe(1);
-  }));
-
-  it('GBRAIN_CONFIG_CACHE_TTL_MS=0 disables the cache (per-key reads)', () => withTtl('0', async () => {
-    const { engine, calls } = makeEngine([{ value: 'v' }]);
-    expect(await engine.getConfig('k')).toBe('v');
-    expect(await engine.getConfig('k')).toBe('v');
-    expect(calls.length).toBe(2);
-    expect(calls[0]).toContain('SELECT value FROM config WHERE key =');
-  }));
-
-  it('an expired TTL reloads from the database', () => withTtl('1', async () => {
-    const { engine, calls } = makeEngine([{ key: 'k', value: 'v' }]);
-    expect(await engine.getConfig('k')).toBe('v');
-    await new Promise((r) => setTimeout(r, 5));
-    expect(await engine.getConfig('k')).toBe('v');
-    expect(calls.length).toBe(2); // two batch loads
-  }));
-
-  it('the batch load keeps the connRetry reconnect posture (#1603/#1891)', () => withTtl(undefined, async () => {
-    const e = new PostgresEngine();
-    (e as unknown as { _connectionStyle: string })._connectionStyle = 'instance';
-    (e as unknown as { _sql: unknown })._sql = null; // torn-down pool → retryable
-    (e as unknown as { _bulkRetryOptsCache: unknown })._bulkRetryOptsCache = FAST_RETRY;
-    let reconnects = 0;
-    (e as unknown as { reconnect: () => Promise<void> }).reconnect = async () => {
-      reconnects++;
-      (e as unknown as { _sql: unknown })._sql = () =>
-        Promise.resolve([{ key: 'k', value: 'v' }]);
-    };
-    expect(await e.getConfig('k')).toBe('v');
-    expect(reconnects).toBe(1);
-  }));
-});

test/postgres-engine-config-reconnect.test.ts

@@ -43,9 +43,7 @@ function makeTornDownEngine(poolResult: unknown): { engine: PostgresEngine; reco
 
 describe('PostgresEngine non-batch config accessors self-heal (PR #1891 takeover)', () => {
   it('getConfig reconnects + retries a null instance pool, then returns the value', async () => {
-    // Rows carry `key` too: getConfig's default cached path batch-loads
-    // `SELECT key, value FROM config` (#1694) through the same connRetry.
-    const { engine, reconnects } = makeTornDownEngine([{ key: 'some.key', value: 'live-value' }]);
+    const { engine, reconnects } = makeTornDownEngine([{ value: 'live-value' }]);
     expect(await engine.getConfig('some.key')).toBe('live-value');
     expect(reconnects()).toBe(1); // exactly one reconnect closed the gap
   });