fix(embed): support hosted Perplexity embeddings (pplx-embed-v1-*) (#1046)

Adds a `perplexity` embedding recipe (OpenAI-compatible at
https://api.perplexity.ai/v1, auth via PERPLEXITY_API_KEY only — never an
OPENAI_API_KEY fallback) covering pplx-embed-v1-0.6b and pplx-embed-v1-4b.

Perplexity's /embeddings endpoint diverges from OpenAI's wire shape in two
places that break the AI SDK adapter, handled by a new perplexityCompatFetch
shim (mirrors the Voyage/ZeroEntropy pattern incl. the two-layer OOM caps):
- encoding_format only accepts base64_int8/base64_binary; the SDK's 'float'
  default is forced to 'base64_int8' outbound.
- The response embedding is base64-encoded signed int8 components (natively
  quantized); decoded to number[] inbound so the SDK's Zod schema validates.
  Cosine similarity is scale-invariant, so raw int8 components rank correctly.

Flexible dims (Matryoshka-style 128..native max: 1024 for 0.6b, 2560 for 4b)
validate fail-loud in dims.ts + the init preflight; `dimensions` is
Perplexity's native field so no wire translation is needed. default_dims is
1024 (works on a plain vector column for both models); the 4b model's full
2560 width rides the existing halfvec (>2000 dims) storage/ANN path. Pricing
entries land in embedding-pricing.ts.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

src/core/ai/anthropic-key.ts

@@ -18,22 +18,12 @@
 import { loadConfig } from '../config.ts';
 
 export function hasAnthropicKey(): boolean {
-  return resolveAnthropicKey() !== undefined;
-}
-
-/**
- * Resolve the actual key value: env first, then the gbrain config file.
- * Callers constructing an Anthropic client directly (e.g. the legacy
- * subagent path) must pass this as `apiKey` — a bare `new Anthropic()`
- * only sees env, so launchd/MCP workers with config-stored keys fail.
- */
-export function resolveAnthropicKey(): string | undefined {
-  if (process.env.ANTHROPIC_API_KEY) return process.env.ANTHROPIC_API_KEY;
+  if (process.env.ANTHROPIC_API_KEY) return true;
   try {
     const cfg = loadConfig();
-    if (cfg?.anthropic_api_key) return cfg.anthropic_api_key;
+    if (cfg?.anthropic_api_key) return true;
   } catch {
     // loadConfig may throw on first-run installs; treat as no key available.
   }
-  return undefined;
+  return false;
 }

src/core/ai/dims.ts

@@ -90,6 +90,30 @@ export function isValidOpenAITextEmbedding3Dim(modelId: string, dims: number): b
   return Number.isInteger(dims) && dims >= 1 && dims <= max;
 }
 
+// Perplexity hosted embeddings (#1046): Matryoshka-style flexible dims,
+// any integer from 128 up to the model's native size. `dimensions` is the
+// native wire field (no translation needed); output encoding divergence
+// (base64 int8) is handled by perplexityCompatFetch in gateway.ts.
+const PERPLEXITY_EMBEDDING_MAX_DIMS: Record<string, number> = {
+  'pplx-embed-v1-0.6b': 1024,
+  'pplx-embed-v1-4b': 2560,
+};
+export const PERPLEXITY_MIN_DIMS = 128;
+
+export function isPerplexityEmbeddingModel(modelId: string): boolean {
+  return modelId in PERPLEXITY_EMBEDDING_MAX_DIMS;
+}
+
+export function maxPerplexityEmbeddingDim(modelId: string): number | undefined {
+  return PERPLEXITY_EMBEDDING_MAX_DIMS[modelId];
+}
+
+export function isValidPerplexityDim(modelId: string, dims: number): boolean {
+  const max = PERPLEXITY_EMBEDDING_MAX_DIMS[modelId];
+  if (max === undefined) return false;
+  return Number.isInteger(dims) && dims >= PERPLEXITY_MIN_DIMS && dims <= max;
+}
+
 // NVIDIA NIM hosted embedding models use asymmetric input_type values. Most
 // emit fixed natural dimensions, but llama-nemotron-embed-1b-v2 accepts
 // Matryoshka-style dimension overrides (e.g. matching an existing 1280d
@@ -226,6 +250,23 @@ export function dimsProviderOptions(
           },
         };
       }
+      // Perplexity pplx-embed-v1-* — flexible dims via the native
+      // `dimensions` field. Fail-loud when the configured dim is outside
+      // the model's range (same rationale as the Voyage/ZE guards: the
+      // upstream HTTP 400 misroutes as a transient network error).
+      // Symmetric retrieval — inputType is never emitted.
+      if (isPerplexityEmbeddingModel(modelId)) {
+        if (!isValidPerplexityDim(modelId, dims)) {
+          const max = maxPerplexityEmbeddingDim(modelId)!;
+          throw new AIConfigError(
+            `Perplexity model "${modelId}" supports embedding_dimensions in ` +
+            `${PERPLEXITY_MIN_DIMS}..${max}, got ${dims}.`,
+            `Set \`embedding_dimensions\` to a value between ${PERPLEXITY_MIN_DIMS} and ${max} ` +
+            `in your gbrain config.`,
+          );
+        }
+        return { openaiCompatible: { dimensions: dims } };
+      }
       // NVIDIA NIM hosted embeddings are OpenAI-compatible but require
       // asymmetric input_type. Use passage for indexing/document-side vectors
       // and query for search-side vectors. Only llama-nemotron-embed-1b-v2

src/core/ai/gateway.ts

@@ -263,6 +263,18 @@ export class ZeroEntropyResponseTooLargeError extends Error {
   }
 }
 
+/** Perplexity twin of the Voyage/ZE OOM caps (#1046). Int8 components are
+ * 1 byte each, so a real response (512 texts × 2560 dims) is ~1.3 MB —
+ * anything near this cap is unambiguously not legitimate. */
+const MAX_PERPLEXITY_RESPONSE_BYTES = 256 * 1024 * 1024;
+
+export class PerplexityResponseTooLargeError extends Error {
+  constructor(message: string) {
+    super(message);
+    this.name = 'PerplexityResponseTooLargeError';
+  }
+}
+
 // ---- Unified auth resolution (D12=A) ----
 //
 // Pre-v0.32, openai-compatible auth was duplicated across instantiateEmbedding,
@@ -1292,6 +1304,103 @@ const openAICompatAsymmetricFetch = (async (input: RequestInfo | URL, init?: Req
   return fetch(typeof input === 'string' ? input : input.toString(), baseInit);
 }) as unknown as typeof fetch;
 
+/**
+ * Perplexity compatibility shim (#1046). Perplexity's `/v1/embeddings`
+ * endpoint is OpenAI-shaped but diverges on two points that break the AI
+ * SDK's openai-compatible adapter:
+ *   - `encoding_format` only accepts 'base64_int8' (default) or
+ *     'base64_binary'; the SDK sends 'float', which Perplexity rejects.
+ *     Force 'base64_int8' on the wire.
+ *   - The response `embedding` is a base64 string encoding SIGNED INT8
+ *     components (natively quantized output). The SDK schema expects
+ *     `number[]` — decode Int8Array → number[] here. Cosine similarity is
+ *     scale-invariant, so the raw int8 components rank correctly.
+ * `dimensions` is Perplexity's native field name — no translation needed
+ * (dims.ts emits it directly). Layer 1/Layer 2 OOM caps mirror the Voyage
+ * pattern.
+ *
+ * Exported for tests (behavioral coverage of the int8 decode); not part of
+ * the public gateway API.
+ */
+export const perplexityCompatFetch = (async (input: RequestInfo | URL, init?: RequestInit) => {
+  // OUTBOUND: force the encoding Perplexity actually accepts.
+  if (init?.body && typeof init.body === 'string') {
+    try {
+      const parsed = JSON.parse(init.body);
+      if (parsed && typeof parsed === 'object' && parsed.encoding_format !== 'base64_int8') {
+        parsed.encoding_format = 'base64_int8';
+        // Drop Content-Length so fetch recomputes from the new body.
+        const headers = new Headers(init.headers ?? {});
+        headers.delete('content-length');
+        init = { ...init, body: JSON.stringify(parsed), headers };
+      }
+    } catch {
+      // Body wasn't JSON — pass through untouched.
+    }
+  }
+
+  const resp = await fetch(input as any, init);
+  if (!resp.ok) return resp;
+  const ct = resp.headers.get('content-type') ?? '';
+  if (!ct.toLowerCase().includes('application/json')) return resp;
+
+  // Layer 1: Content-Length pre-check BEFORE the body is parsed.
+  const contentLengthHeader = resp.headers.get('content-length');
+  if (contentLengthHeader) {
+    const len = parseInt(contentLengthHeader, 10);
+    if (Number.isFinite(len) && len > MAX_PERPLEXITY_RESPONSE_BYTES) {
+      throw new PerplexityResponseTooLargeError(
+        `Perplexity response Content-Length=${len} exceeds ${MAX_PERPLEXITY_RESPONSE_BYTES} bytes — ` +
+        `likely compromised endpoint or misconfiguration`,
+      );
+    }
+  }
+
+  // INBOUND: decode base64 int8 embeddings to number[] so the SDK's Zod
+  // schema validates.
+  try {
+    const json: any = await resp.clone().json();
+    if (!json || typeof json !== 'object') return resp;
+    let modified = false;
+    if (Array.isArray(json.data)) {
+      for (const item of json.data) {
+        if (item && typeof item.embedding === 'string') {
+          // Layer 2: per-embedding cap for chunked responses that skipped
+          // Layer 1. base64 → bytes is the canonical 0.75 ratio.
+          const estDecoded = Math.ceil(item.embedding.length * 0.75);
+          if (estDecoded > MAX_PERPLEXITY_RESPONSE_BYTES) {
+            throw new PerplexityResponseTooLargeError(
+              `Perplexity embedding base64 exceeds ${MAX_PERPLEXITY_RESPONSE_BYTES} bytes ` +
+              `(estimated ${estDecoded} bytes from ${item.embedding.length} base64 chars)`,
+            );
+          }
+          // base64_int8: one signed int8 per component.
+          const bytes = Buffer.from(item.embedding, 'base64');
+          item.embedding = Array.from(new Int8Array(bytes.buffer, bytes.byteOffset, bytes.byteLength));
+          modified = true;
+        }
+      }
+    }
+    if (json.usage && typeof json.usage === 'object' && json.usage.prompt_tokens === undefined) {
+      json.usage.prompt_tokens = typeof json.usage.total_tokens === 'number'
+        ? json.usage.total_tokens
+        : 0;
+      modified = true;
+    }
+    if (!modified) return resp;
+    return new Response(JSON.stringify(json), {
+      status: resp.status,
+      statusText: resp.statusText,
+      headers: resp.headers,
+    });
+  } catch (err) {
+    // OOM-cap throws MUST propagate; anything else falls back to the
+    // original response (same contract as voyageCompatFetch).
+    if (err instanceof PerplexityResponseTooLargeError) throw err;
+    return resp;
+  }
+}) as unknown as typeof fetch;
+
 async function resolveEmbeddingProvider(modelStr: string): Promise<{ model: any; recipe: Recipe; modelId: string }> {
   const { parsed, recipe } = resolveRecipe(modelStr);
   assertTouchpoint(recipe, 'embedding', parsed.modelId, getExtendedModelsForProvider(parsed.providerId));
@@ -1360,6 +1469,8 @@ function instantiateEmbedding(recipe: Recipe, modelId: string, cfg: AIGatewayCon
           ? zeroEntropyCompatFetch
           : recipe.id === 'nvidia'
           ? nvidiaCompatFetch
+          : recipe.id === 'perplexity'
+          ? perplexityCompatFetch
           : openAICompatAsymmetricFetch);
       const client = createOpenAICompatible({
         name: recipe.id,

src/core/ai/recipes/index.ts

@@ -26,6 +26,7 @@ import { llamaServerReranker } from './llama-server-reranker.ts';
 import { moonshot } from './moonshot.ts';
 import { mistral } from './mistral.ts';
 import { nvidia } from './nvidia.ts';
+import { perplexity } from './perplexity.ts';
 
 const ALL: Recipe[] = [
   openai,
@@ -48,6 +49,7 @@ const ALL: Recipe[] = [
   moonshot,
   mistral,
   nvidia,
+  perplexity,
 ];
 
 /** Map from `provider:id` key to recipe. */

src/core/ai/recipes/perplexity.ts

@@ -0,0 +1,54 @@
+import type { Recipe } from '../types.ts';
+
+/**
+ * Perplexity's hosted embeddings API (#1046). OpenAI-shaped at
+ * `POST {base}/embeddings` but diverges on the wire:
+ *   - `encoding_format` only accepts 'base64_int8' (default) or
+ *     'base64_binary' — the AI SDK's 'float' default is rejected.
+ *   - The response `embedding` is a base64 string encoding SIGNED INT8
+ *     components (natively quantized output), not a float array.
+ * Both divergences are handled by perplexityCompatFetch in gateway.ts
+ * (force 'base64_int8' outbound; decode Int8Array → number[] inbound).
+ * Cosine similarity is scale-invariant, so the raw int8 components store
+ * and rank correctly as floats.
+ *
+ * Models (per docs.perplexity.ai/api-reference/embeddings-post, 2026-07):
+ *   - pplx-embed-v1-0.6b: dims 128..1024 (default 1024)
+ *   - pplx-embed-v1-4b:   dims 128..2560 (default 2560)
+ * The flexible-dim range validation lives in src/core/ai/dims.ts
+ * (PERPLEXITY_EMBEDDING_MAX_DIMS). default_dims is pinned at 1024 so both
+ * models work out of the box on a plain vector(N) column; users who want
+ * the 4b model's full 2560 width set `embedding_dimensions: 2560` and the
+ * existing halfvec path (dims > 2000) covers storage + ANN.
+ *
+ * Auth is PERPLEXITY_API_KEY only — deliberately NO OPENAI_API_KEY
+ * fallback (a Perplexity brain must never silently bill/route through
+ * OpenAI). If your key lives in PPLX_API_KEY, re-export it.
+ */
+export const perplexity: Recipe = {
+  id: 'perplexity',
+  name: 'Perplexity',
+  tier: 'openai-compat',
+  implementation: 'openai-compatible',
+  base_url_default: 'https://api.perplexity.ai/v1',
+  auth_env: {
+    required: ['PERPLEXITY_API_KEY'],
+    setup_url: 'https://www.perplexity.ai/settings/api',
+  },
+  touchpoints: {
+    embedding: {
+      models: ['pplx-embed-v1-0.6b', 'pplx-embed-v1-4b'],
+      default_dims: 1024,
+      cost_per_1m_tokens_usd: 0.03, // pplx-embed-v1-4b; 0.6b is $0.004/M
+      price_last_verified: '2026-07-21',
+      // Perplexity enforces 120K combined tokens (and 512 texts) per
+      // request. Same pre-split posture as Voyage: assume a dense
+      // tokenizer (1 char ≈ 1 token) at 0.5 utilization; the gateway's
+      // recursive halving is the runtime safety net.
+      max_batch_tokens: 120_000,
+      chars_per_token: 1,
+      safety_factor: 0.5,
+    },
+  },
+  setup_hint: 'Get an API key at https://www.perplexity.ai/settings/api, then `export PERPLEXITY_API_KEY=...` (re-export PPLX_API_KEY if that is where your key lives).',
+};

src/core/embedding-dim-check.ts

@@ -32,6 +32,10 @@ import {
   nvidiaEmbeddingDim,
   nvidiaEmbeddingDimOptions,
   supportsNvidiaEmbeddingDimension,
+  isPerplexityEmbeddingModel,
+  isValidPerplexityDim,
+  maxPerplexityEmbeddingDim,
+  PERPLEXITY_MIN_DIMS,
 } from './ai/dims.ts';
 
 /**
@@ -462,6 +466,15 @@ function isCustomDimValidForProvider(
         `(allowed: ${ZEROENTROPY_VALID_DIMS.join(', ')}).`,
     };
   }
+  if (recipe.id === 'perplexity' && isPerplexityEmbeddingModel(modelId)) {
+    if (isValidPerplexityDim(modelId, requestedDims)) return { valid: true, error: '' };
+    return {
+      valid: false,
+      error:
+        `Perplexity ${modelId} accepts dimensions ${PERPLEXITY_MIN_DIMS}..${maxPerplexityEmbeddingDim(modelId)}, ` +
+        `got ${requestedDims}.`,
+    };
+  }
   if (recipe.id === 'openai' && isOpenAITextEmbedding3Model(modelId)) {
     if (isValidOpenAITextEmbedding3Dim(modelId, requestedDims)) return { valid: true, error: '' };
     const maxDim = maxOpenAITextEmbedding3Dim(modelId);

src/core/embedding-pricing.ts

@@ -40,6 +40,9 @@ export const EMBEDDING_PRICING: Record<string, EmbeddingPricing> = {
   // Mistral (https://mistral.ai/pricing/api/, verified 2026-07-19)
   'mistral:mistral-embed':         { pricePerMTok: 0.10 },
   'mistral:mistral-embed-2312':    { pricePerMTok: 0.10 },
+  // Perplexity (https://docs.perplexity.ai/getting-started/pricing, verified 2026-07-21)
+  'perplexity:pplx-embed-v1-0.6b': { pricePerMTok: 0.004 },
+  'perplexity:pplx-embed-v1-4b':   { pricePerMTok: 0.03 },
 };
 
 export type PriceLookupResult =

src/core/minions/handlers/subagent.ts

@@ -48,7 +48,6 @@ import {
   logSubagentHeartbeat,
 } from './subagent-audit.ts';
 import { resolveModel, isAnthropicProvider, TIER_DEFAULTS } from '../../model-config.ts';
-import { resolveAnthropicKey } from '../../ai/anthropic-key.ts';
 import { buildSystemPrompt, DEFAULT_SUBAGENT_SYSTEM } from '../system-prompt.ts';
 import { toolLoop as gatewayToolLoop } from '../../ai/gateway.ts';
 import type { ChatToolDef, ChatMessage, ChatBlock, ChatResult, ToolHandler } from '../../ai/gateway.ts';
@@ -187,10 +186,7 @@ export function makeSubagentHandler(deps: SubagentDeps) {
   // lives at sdk.messages.create. Assigning sdk.messages directly gets the
   // right object; JS method-call semantics preserve `this` at the call
   // site (subagent.ts invokes client.create(...) with client === sdk.messages).
-  // Resolve the key env-first, then config (anthropic_api_key) — a bare
-  // new Anthropic() only reads env, so launchd/MCP workers whose key lives
-  // in the gbrain config file would fail auth (#2048).
-  const makeAnthropic = deps.makeAnthropic ?? (() => new Anthropic({ apiKey: resolveAnthropicKey() }));
+  const makeAnthropic = deps.makeAnthropic ?? (() => new Anthropic());
   const client: MessagesClient = deps.client ?? makeAnthropic().messages;
   const config = deps.config ?? loadConfig() ?? ({ engine: 'postgres' } as GBrainConfig);
   const rateLeaseKey = deps.rateLeaseKey ?? DEFAULT_RATE_KEY;

src/core/operations.ts

@@ -4562,8 +4562,7 @@ const list_schema_packs: Operation = {
     const { existsSync, readdirSync } = await import('node:fs');
     const { join } = await import('node:path');
     const { gbrainPath } = await import('./config.ts');
-    const { BUNDLED_PACK_NAMES } = await import('./schema-pack/bundled.ts');
-    const bundled = [...BUNDLED_PACK_NAMES];
+    const bundled = ['gbrain-base', 'gbrain-recommended'];
     const installedDir = gbrainPath('schema-packs');
     const installed: string[] = [];
     if (existsSync(installedDir)) {

src/core/schema-pack/bundled.ts

@@ -1,24 +0,0 @@
-// Bundled schema-pack registry — single source of truth for the packs that
-// ship in src/core/schema-pack/base/. Keep every bundled-pack consumer
-// (CLI/MCP inspection, active-pack loading, mutation guards, upgrade
-// discovery) on this one list so they cannot drift.
-//
-// v0.39 T8 — gbrain-base + gbrain-recommended.
-// v0.41 T4 — lens packs: creator, investor, engineer, everything (meta-pack).
-// v0.42 type-unification — gbrain-base-v2, the 15-type canonical successor.
-
-export const BUNDLED_PACK_NAMES = [
-  'gbrain-base',
-  'gbrain-recommended',
-  'gbrain-creator',
-  'gbrain-investor',
-  'gbrain-engineer',
-  'gbrain-everything',
-  'gbrain-base-v2',
-] as const;
-
-export type BundledPackName = typeof BUNDLED_PACK_NAMES[number];
-
-export function isBundledPackName(name: string): name is BundledPackName {
-  return (BUNDLED_PACK_NAMES as readonly string[]).includes(name);
-}

src/core/schema-pack/load-active.ts

@@ -37,7 +37,6 @@ import {
   type ResolutionInput,
   type ResolutionResult,
 } from './registry.ts';
-import { isBundledPackName } from './bundled.ts';
 
 /**
  * Inputs the caller (operations.ts handler / engine query path) provides.
@@ -93,7 +92,28 @@ export function _resetPackLocatorForTests(): void {
  * throwing UnknownPackError with a paste-ready install hint.
  */
 function defaultPackLocator(name: string): string | null {
-  if (isBundledPackName(name)) {
+  // v0.39 T8 — bundled packs registry. gbrain-base + gbrain-recommended
+  // ship in src/core/schema-pack/base/. Add a new entry here to bundle
+  // additional canonical packs.
+  //
+  // v0.41 T4 — lens packs join the bundle: creator (atoms + concepts +
+  // extract_atoms/synthesize_concepts phases), investor (theses + bet
+  // resolution + 3 calibration domains), engineer (gstack-learnings bridge
+  // + 3 calibration domains), everything (meta-pack stacking all three
+  // via extends + borrow_from). Each ships as a real YAML at base/<name>.yaml.
+  const BUNDLED: ReadonlyArray<string> = [
+    'gbrain-base',
+    'gbrain-recommended',
+    'gbrain-creator',
+    'gbrain-investor',
+    'gbrain-engineer',
+    'gbrain-everything',
+    // v0.42 type-unification: 15-type canonical successor to gbrain-base.
+    // Ships as install default (Lane E T17) + via gbrain onboard pack
+    // upgrade flow (the unify-types Minion handler).
+    'gbrain-base-v2',
+  ];
+  if (BUNDLED.includes(name)) {
     // Resolve bundled YAML relative to this source file. Works in both
     // direct-bun execution and bun --compile binaries.
     const here = dirname(fileURLToPath(import.meta.url));

src/core/schema-pack/loader.ts

@@ -159,29 +159,6 @@ export function parseYamlMini(content: string): unknown {
     return parseMapping(baseIndent);
   }
 
-  function parseBlockScalar(parentIndent: number, folded: boolean): string {
-    const contentIndent = parentIndent + 2;
-    const out: string[] = [];
-    while (i < lines.length) {
-      const raw = lines[i];
-      // Inside a block scalar everything is literal content — '#' is NOT a
-      // comment here, so use the raw line (no stripComment / isBlank).
-      if (raw.trim() === '') {
-        out.push('');
-        i++;
-        continue;
-      }
-      const indent = indentOf(raw);
-      if (indent <= parentIndent) break;
-      out.push(raw.slice(Math.min(contentIndent, indent)));
-      i++;
-    }
-    if (folded) {
-      return out.join(' ').replace(/\s+$/u, '');
-    }
-    return out.join('\n').replace(/\n+$/u, '');
-  }
-
   function parseSequence(baseIndent: number): unknown[] {
     const result: unknown[] = [];
     while (i < lines.length) {
@@ -250,10 +227,6 @@ export function parseYamlMini(content: string): unknown {
           i++;
           if (rest2 === '') {
             map[key2] = parseBlock(nextIndent + 2);
-          } else if (rest2 === '|' || rest2 === '|-' || rest2 === '|+') {
-            map[key2] = parseBlockScalar(nextIndent, false);
-          } else if (rest2 === '>' || rest2 === '>-' || rest2 === '>+') {
-            map[key2] = parseBlockScalar(nextIndent, true);
           } else {
             map[key2] = parseScalar(rest2);
           }
@@ -284,10 +257,6 @@ export function parseYamlMini(content: string): unknown {
       i++;
       if (rest === '') {
         result[key] = parseBlock(indent + 2);
-      } else if (rest === '|' || rest === '|-' || rest === '|+') {
-        result[key] = parseBlockScalar(indent, false);
-      } else if (rest === '>' || rest === '>-' || rest === '>+') {
-        result[key] = parseBlockScalar(indent, true);
       } else {
         result[key] = parseScalar(rest);
       }

src/core/schema-pack/mutate.ts

@@ -65,7 +65,6 @@ import { invalidateQueryCache } from './query-cache-invalidator.ts';
 import { logMutationFailure, logMutationSuccess, type MutationActor, type MutationOp } from './mutate-audit.ts';
 import { runFilePlaneLintRules } from './lint-rules.ts';
 import { withPackLock, type PackLockOpts } from './pack-lock.ts';
-import { BUNDLED_PACK_NAMES as BUNDLED_PACK_NAME_LIST } from './bundled.ts';
 import type { BrainEngine } from '../engine.ts';
 
 export type PackFileFormat = 'json' | 'yaml';
@@ -94,7 +93,7 @@ export class SchemaPackMutationError extends Error {
   }
 }
 
-export const BUNDLED_PACK_NAMES = new Set<string>(BUNDLED_PACK_NAME_LIST);
+export const BUNDLED_PACK_NAMES = new Set(['gbrain-base', 'gbrain-recommended', 'gbrain-base-v2']);
 
 export interface MutateResult {
   /** Pack name that was mutated. */

test/ai/anthropic-key.test.ts

@@ -10,7 +10,7 @@ import { mkdtempSync, mkdirSync, writeFileSync, rmSync } from 'node:fs';
 import { tmpdir } from 'node:os';
 import { join } from 'node:path';
 import { withEnv } from '../helpers/with-env.ts';
-import { hasAnthropicKey, resolveAnthropicKey } from '../../src/core/ai/anthropic-key.ts';
+import { hasAnthropicKey } from '../../src/core/ai/anthropic-key.ts';
 
 const tmpDirs: string[] = [];
 function freshHome(withConfig?: Record<string, unknown>): string {
@@ -62,35 +62,3 @@ describe('hasAnthropicKey', () => {
     );
   });
 });
-
-describe('resolveAnthropicKey (#2048 — subagent config-key auth)', () => {
-  test('env wins over config', async () => {
-    const home = freshHome({ anthropic_api_key: 'sk-from-config' });
-    await withEnv(
-      { ANTHROPIC_API_KEY: 'sk-from-env', GBRAIN_HOME: home, DATABASE_URL: undefined, GBRAIN_DATABASE_URL: undefined },
-      async () => {
-        expect(resolveAnthropicKey()).toBe('sk-from-env');
-      },
-    );
-  });
-
-  test('config key returned when env unset', async () => {
-    const home = freshHome({ anthropic_api_key: 'sk-from-config' });
-    await withEnv(
-      { ANTHROPIC_API_KEY: undefined, GBRAIN_HOME: home, DATABASE_URL: undefined, GBRAIN_DATABASE_URL: undefined },
-      async () => {
-        expect(resolveAnthropicKey()).toBe('sk-from-config');
-      },
-    );
-  });
-
-  test('neither → undefined', async () => {
-    const home = freshHome();
-    await withEnv(
-      { ANTHROPIC_API_KEY: undefined, GBRAIN_HOME: home, DATABASE_URL: undefined, GBRAIN_DATABASE_URL: undefined },
-      async () => {
-        expect(resolveAnthropicKey()).toBeUndefined();
-      },
-    );
-  });
-});

test/ai/recipe-perplexity.test.ts

@@ -0,0 +1,142 @@
+/**
+ * #1046 — Perplexity hosted embeddings (pplx-embed-v1-*).
+ *
+ * Covers the three seams the recipe touches:
+ *  - recipe registration + auth (PERPLEXITY_API_KEY only, never OPENAI_API_KEY)
+ *  - flexible-dim validation (128..native max) in dims.ts + the init
+ *    preflight (resolveSchemaEmbeddingDim), incl. the >2000-dim 4b case
+ *  - perplexityCompatFetch: forces encoding_format=base64_int8 outbound and
+ *    decodes the base64 int8 embedding payload to number[] inbound
+ */
+import { afterEach, describe, expect, test } from 'bun:test';
+import {
+  dimsProviderOptions,
+  isPerplexityEmbeddingModel,
+  isValidPerplexityDim,
+  maxPerplexityEmbeddingDim,
+} from '../../src/core/ai/dims.ts';
+import { getRecipe, RECIPES } from '../../src/core/ai/recipes/index.ts';
+import { perplexity } from '../../src/core/ai/recipes/perplexity.ts';
+import { defaultResolveAuth, perplexityCompatFetch } from '../../src/core/ai/gateway.ts';
+import { AIConfigError } from '../../src/core/ai/errors.ts';
+import { resolveSchemaEmbeddingDim } from '../../src/core/embedding-dim-check.ts';
+import { lookupEmbeddingPrice } from '../../src/core/embedding-pricing.ts';
+
+describe('recipe: perplexity', () => {
+  test('registered as an OpenAI-compatible embedding provider', () => {
+    expect(RECIPES.has('perplexity')).toBe(true);
+    expect(getRecipe('perplexity')).toBe(perplexity);
+    expect(perplexity.tier).toBe('openai-compat');
+    expect(perplexity.implementation).toBe('openai-compatible');
+    expect(perplexity.base_url_default).toBe('https://api.perplexity.ai/v1');
+    const e = perplexity.touchpoints.embedding!;
+    expect(e.models).toEqual(['pplx-embed-v1-0.6b', 'pplx-embed-v1-4b']);
+    expect(e.default_dims).toBe(1024);
+    expect(e.max_batch_tokens).toBe(120_000);
+  });
+
+  test('auth is PERPLEXITY_API_KEY bearer — no OPENAI_API_KEY fallback', () => {
+    expect(perplexity.resolveAuth).toBeUndefined();
+    expect(perplexity.auth_env?.required).toEqual(['PERPLEXITY_API_KEY']);
+    expect(defaultResolveAuth(perplexity, { PERPLEXITY_API_KEY: 'fake-pplx' }, 'embedding')).toEqual({
+      headerName: 'Authorization',
+      token: 'Bearer fake-pplx',
+    });
+    // An OPENAI_API_KEY in the env must NOT satisfy Perplexity auth.
+    expect(() => defaultResolveAuth(perplexity, { OPENAI_API_KEY: 'sk-test' }, 'embedding')).toThrow(AIConfigError);
+  });
+
+  test('dims: 128..native-max range per model', () => {
+    expect(isPerplexityEmbeddingModel('pplx-embed-v1-4b')).toBe(true);
+    expect(maxPerplexityEmbeddingDim('pplx-embed-v1-4b')).toBe(2560);
+    expect(maxPerplexityEmbeddingDim('pplx-embed-v1-0.6b')).toBe(1024);
+    expect(isValidPerplexityDim('pplx-embed-v1-4b', 2560)).toBe(true);
+    expect(isValidPerplexityDim('pplx-embed-v1-4b', 128)).toBe(true);
+    expect(isValidPerplexityDim('pplx-embed-v1-4b', 64)).toBe(false);
+    expect(isValidPerplexityDim('pplx-embed-v1-0.6b', 2560)).toBe(false);
+  });
+
+  test('dimsProviderOptions emits native `dimensions`, fails loud out of range', () => {
+    expect(dimsProviderOptions('openai-compatible', 'pplx-embed-v1-4b', 2560)).toEqual({
+      openaiCompatible: { dimensions: 2560 },
+    });
+    // Symmetric provider — inputType never emitted.
+    expect(dimsProviderOptions('openai-compatible', 'pplx-embed-v1-4b', 1024, 'query')).toEqual({
+      openaiCompatible: { dimensions: 1024 },
+    });
+    expect(() => dimsProviderOptions('openai-compatible', 'pplx-embed-v1-0.6b', 2560)).toThrow(AIConfigError);
+  });
+
+  test('init preflight accepts the 4b model at its native 2560 dims (halfvec territory)', () => {
+    const res = resolveSchemaEmbeddingDim({
+      embedding_model: 'perplexity:pplx-embed-v1-4b',
+      embedding_dimensions: 2560,
+    });
+    expect(res).toEqual({
+      ok: true,
+      dim: 2560,
+      model: 'perplexity:pplx-embed-v1-4b',
+      provider: 'perplexity',
+      recipeDefault: 1024,
+    });
+    const bad = resolveSchemaEmbeddingDim({
+      embedding_model: 'perplexity:pplx-embed-v1-4b',
+      embedding_dimensions: 4096,
+    });
+    expect(bad.ok).toBe(false);
+  });
+
+  test('embedding pricing table knows both models', () => {
+    expect(lookupEmbeddingPrice('perplexity:pplx-embed-v1-4b')).toMatchObject({ kind: 'known', pricePerMTok: 0.03 });
+    expect(lookupEmbeddingPrice('perplexity:pplx-embed-v1-0.6b')).toMatchObject({ kind: 'known', pricePerMTok: 0.004 });
+  });
+});
+
+describe('perplexityCompatFetch — int8 wire shim', () => {
+  const realFetch = globalThis.fetch;
+  afterEach(() => {
+    globalThis.fetch = realFetch;
+  });
+
+  test('forces encoding_format=base64_int8 outbound and decodes int8 base64 inbound', async () => {
+    const int8 = new Int8Array([3, -7, 127, -128]);
+    const b64 = Buffer.from(int8.buffer).toString('base64');
+    let sentBody: any;
+    globalThis.fetch = (async (_input: any, init?: RequestInit) => {
+      sentBody = JSON.parse(init!.body as string);
+      return new Response(
+        JSON.stringify({
+          object: 'list',
+          model: 'pplx-embed-v1-4b',
+          data: [{ object: 'embedding', index: 0, embedding: b64 }],
+          usage: { prompt_tokens: 4, total_tokens: 4 },
+        }),
+        { status: 200, headers: { 'content-type': 'application/json' } },
+      );
+    }) as any;
+
+    const resp = await (perplexityCompatFetch as any)('https://api.perplexity.ai/v1/embeddings', {
+      method: 'POST',
+      headers: { 'content-type': 'application/json' },
+      // The AI SDK sends encoding_format:'float' — Perplexity rejects it.
+      body: JSON.stringify({ model: 'pplx-embed-v1-4b', input: ['hi'], encoding_format: 'float', dimensions: 4 }),
+    });
+
+    expect(sentBody.encoding_format).toBe('base64_int8');
+    expect(sentBody.dimensions).toBe(4); // native field, untouched
+    const json = await resp.json();
+    expect(json.data[0].embedding).toEqual([3, -7, 127, -128]);
+    expect(json.usage.prompt_tokens).toBe(4);
+  });
+
+  test('non-JSON and error responses pass through untouched', async () => {
+    globalThis.fetch = (async () =>
+      new Response('nope', { status: 401, headers: { 'content-type': 'text/plain' } })) as any;
+    const resp = await (perplexityCompatFetch as any)('https://api.perplexity.ai/v1/embeddings', {
+      method: 'POST',
+      body: JSON.stringify({ model: 'pplx-embed-v1-4b', input: ['hi'] }),
+    });
+    expect(resp.status).toBe(401);
+    expect(await resp.text()).toBe('nope');
+  });
+});

test/lens-pack-manifests.test.ts

@@ -55,13 +55,13 @@ describe('v0.41 T4: all 4 bundled lens packs parse cleanly', () => {
 });
 
 describe('v0.41 T4: bundled registry includes lens packs', () => {
-  test('BUNDLED_PACK_NAMES includes the 4 lens pack names', async () => {
-    // The bundled list moved from load-active.ts to bundled.ts (the
-    // single source of truth); assert the array directly instead of
-    // grepping source text.
-    const { BUNDLED_PACK_NAMES } = await import('../src/core/schema-pack/bundled.ts');
+  test('load-active.ts BUNDLED array source includes the 4 lens pack names', () => {
+    const loadActiveSrc = readFileSync(
+      join(here, '..', 'src', 'core', 'schema-pack', 'load-active.ts'),
+      'utf-8',
+    );
     for (const name of PACK_NAMES) {
-      expect(BUNDLED_PACK_NAMES).toContain(name);
+      expect(loadActiveSrc).toContain(`'${name}'`);
     }
   });
 });

test/operations-schema-pack.test.ts

@@ -149,9 +149,6 @@ describe('list_schema_packs', () => {
       seedPack('mine');
       const result = await operationsByName.list_schema_packs!.handler(ctxOf(), {}) as { bundled: string[]; installed: string[] };
       expect(result.bundled).toContain('gbrain-base');
-      expect(result.bundled).toContain('gbrain-recommended');
-      expect(result.bundled).toContain('gbrain-base-v2');
-      expect(result.bundled).toContain('gbrain-investor');
       expect(result.installed).toContain('mine');
     });
   });

test/schema-cli.test.ts

@@ -64,14 +64,11 @@ describe('gbrain schema CLI (Phase C)', () => {
     expect(r.stdout + r.stderr).toMatch(/schema|active|list|show|validate|use/i);
   });
 
-  test('schema list shows all bundled packs', () => {
+  test('schema list shows gbrain-base bundled', () => {
     const r = gbrain(['schema', 'list']);
     expect(r.code).toBe(0);
     expect(r.stdout).toContain('Bundled packs:');
     expect(r.stdout).toContain('gbrain-base');
-    expect(r.stdout).toContain('gbrain-recommended');
-    expect(r.stdout).toContain('gbrain-base-v2');
-    expect(r.stdout).toContain('gbrain-investor');
   });
 
   test('schema show gbrain-base prints manifest details', () => {
@@ -100,40 +97,6 @@ describe('gbrain schema CLI (Phase C)', () => {
     expect(r.stdout).toContain('valid manifest');
   });
 
-  test('schema show/validate exposes bundled gbrain-recommended', () => {
-    const show = gbrain(['schema', 'show', 'gbrain-recommended']);
-    expect(show.code).toBe(0);
-    expect(show.stdout).toContain('gbrain-recommended v1.0.0');
-    expect(show.stdout).toContain('Page types (');
-    expect(show.stdout).toContain('meeting :: temporal');
-
-    const validate = gbrain(['schema', 'validate', 'gbrain-recommended']);
-    expect(validate.code).toBe(0);
-    expect(validate.stdout).toContain('valid manifest');
-  });
-
-  test('schema show exposes bundled gbrain-base-v2 successor pack', () => {
-    const r = gbrain(['schema', 'show', 'gbrain-base-v2']);
-    expect(r.code).toBe(0);
-    expect(r.stdout).toContain('gbrain-base-v2 v1.0.0');
-    expect(r.stdout).toContain('Page types (');
-    expect(r.stdout).toContain('Link verbs (14)');
-  });
-
-  test('schema active loads configured gbrain-recommended with real types', () => {
-    const home = mkdtempSync(join(tmpdir(), 'gbrain-schema-active-recommended-'));
-    try {
-      mkdirSync(join(home, '.gbrain'), { recursive: true });
-      writeFileSync(join(home, '.gbrain', 'config.json'), JSON.stringify({ schema_pack: 'gbrain-recommended' }), 'utf-8');
-      const r = gbrain(['schema', 'active'], { GBRAIN_HOME: home });
-      expect(r.code).toBe(0);
-      expect(r.stdout).toContain('Active pack: gbrain-recommended');
-      expect(r.stdout).not.toContain('Page types: 0');
-    } finally {
-      rmSync(home, { recursive: true, force: true });
-    }
-  });
-
   test('schema active reports default resolution', () => {
     const r = gbrain(['schema', 'active']);
     expect(r.code).toBe(0);

test/schema-pack-loader.test.ts

@@ -345,34 +345,6 @@ describe('YAML mini-parser', () => {
     expect(result.types[1].weight).toBe(2);
   });
 
-  test('parses block scalar without swallowing following keys', () => {
-    const yaml = `name: blocky
-description: |
-  First line.
-  Second line.
-page_types:
-  - name: meeting
-    primitive: temporal
-    path_prefixes:
-      - meetings/
-    aliases: []
-    extractable: true
-    expert_routing: false`;
-    const result = parseYamlMini(yaml) as { description: string; page_types: Array<Record<string, unknown>> };
-    expect(result.description).toBe('First line.\nSecond line.');
-    expect(result.page_types).toHaveLength(1);
-    expect(result.page_types[0].name).toBe('meeting');
-  });
-
-  test('block scalar keeps # as literal content, not a comment', () => {
-    const yaml = `description: |
-  See issue #2029 for context.
-name: hashy`;
-    const result = parseYamlMini(yaml) as Record<string, unknown>;
-    expect(result.description).toBe('See issue #2029 for context.');
-    expect(result.name).toBe('hashy');
-  });
-
   test('strips comments', () => {
     const result = parseYamlMini('# top comment\nname: value # inline comment') as Record<string, unknown>;
     expect(result.name).toBe('value');
@@ -402,27 +374,6 @@ extends: null`;
     const pack = loadPackFromString(json, 'fixture.json');
     expect(pack.name).toBe('json-pack');
   });
-
-  test('loads block-scalar pack descriptions without losing page types', () => {
-    const pack = loadPackFromString(`api_version: gbrain-schema-pack-v1
-name: recommended-fixture
-version: 1.0.0
-extends: gbrain-base
-description: |
-  Operational starter pack.
-page_types:
-  - name: meeting
-    primitive: temporal
-    path_prefixes:
-      - meetings/
-    aliases: []
-    extractable: true
-    expert_routing: false
-link_types: []`, 'fixture.yaml');
-    expect(pack.name).toBe('recommended-fixture');
-    expect(pack.extends).toBe('gbrain-base');
-    expect(pack.page_types.map((t) => t.name)).toContain('meeting');
-  });
 });
 
 describe('ReDoS guard', () => {

test/schema-pack-mutate.test.ts

@@ -103,10 +103,7 @@ describe('locateMutablePackFile — bundled guard', () => {
     expect(BUNDLED_PACK_NAMES.has('gbrain-recommended')).toBe(true);
     // v0.42 (T22): gbrain-base-v2 joins the bundled set.
     expect(BUNDLED_PACK_NAMES.has('gbrain-base-v2')).toBe(true);
-    // Derived from the single bundled registry — the lens packs (creator,
-    // investor, engineer, everything) are read-only too.
-    expect(BUNDLED_PACK_NAMES.has('gbrain-investor')).toBe(true);
-    expect(BUNDLED_PACK_NAMES.size).toBe(7);
+    expect(BUNDLED_PACK_NAMES.size).toBe(3);
   });
 
   it('rejects gbrain-base-v2 with PACK_READONLY (bundled guard)', () => {