fix: capture path drops embed timeoutMs (#30)

Capture built its own EmbedClientConfig literal instead of using search.ts's
EngineConfig-to-client-config mapping, so EMBED_TIMEOUT_MS never reached the
capture path and it kept timing out at embed-client.ts's 10s default.

Extract the mapping into a single shared engine-client-config.ts used by
search.ts and capture.ts (and re-exported from both for existing imports),
so every construction site stays in sync with EngineConfig.

Author: TheGreatAxios

src/core/engine-client-config.ts

@@ -0,0 +1,61 @@
+/**
+ * Maps `EngineConfig`'s embed/rerank sub-objects to the client configs
+ * `embed-client.ts`/`rerank-client.ts` dispatch on. This is the ONLY place
+ * that mapping happens — every construction site (search, capture, and any
+ * future caller) must go through these functions so operator overrides like
+ * EMBED_TIMEOUT_MS / RERANK_TIMEOUT_MS reach every code path uniformly.
+ */
+import type { EngineConfig } from "../config.ts";
+import type { EmbedClientConfig } from "./embed-client.ts";
+import type { RerankClientConfig } from "./rerank-client.ts";
+
+const VALID_EMBED_API_STYLES = new Set(["openai", "tei", "ollama"]);
+
+// The engine's `EngineConfig.embed.apiStyle` is a plain, operator-set string
+// (config.ts has no arktype gate on it); `EmbedClientConfig` requires the
+// literal union `embed-client.ts` dispatches on. Validated here, once, at
+// the trust boundary between config and the client — an invalid value is an
+// operator misconfiguration and must fail loudly, not silently degrade.
+// Built from the engine's own operator-configured embed endpoint — a trusted
+// URL, the same as KNOWLEDGE_DATABASE_URL.
+export function toEmbedClientConfig(
+  embed: EngineConfig["embed"],
+): EmbedClientConfig {
+  if (!VALID_EMBED_API_STYLES.has(embed.apiStyle)) {
+    throw new Error(
+      `Invalid EMBED_API_STYLE "${embed.apiStyle}" — must be one of: ${[...VALID_EMBED_API_STYLES].join(", ")}`,
+    );
+  }
+  return {
+    baseUrl: embed.baseUrl,
+    modelId: embed.model,
+    apiStyle: embed.apiStyle as EmbedClientConfig["apiStyle"],
+    ...(embed.apiKey !== undefined ? { apiKey: embed.apiKey } : {}),
+    ...(embed.timeoutMs !== undefined ? { timeoutMs: embed.timeoutMs } : {}),
+  };
+}
+
+// `EngineConfig.rerank` carries no `apiStyle` field — the engine currently
+// wires only a TEI-compatible cross-encoder endpoint (the locked default
+// model, `bge-reranker-v2-m3`, is TEI-servable); rerank apiStyle is hardcoded
+// `"tei"` below. Absent `baseUrl` => rerank is unconfigured => `undefined`,
+// same degrade-soft precedent as the embed config being absent upstream.
+// Built from the engine's own operator-configured rerank endpoint — a trusted
+// URL, the same as KNOWLEDGE_DATABASE_URL.
+export function toRerankClientConfig(
+  rerank: EngineConfig["rerank"],
+): RerankClientConfig | undefined {
+  if (!rerank.baseUrl) return undefined;
+  return {
+    baseUrl: rerank.baseUrl,
+    apiStyle: "tei",
+    ...(rerank.model !== undefined ? { model: rerank.model } : {}),
+    ...(rerank.apiKey !== undefined ? { apiKey: rerank.apiKey } : {}),
+    ...(rerank.maxDocChars !== undefined
+      ? { maxDocChars: rerank.maxDocChars }
+      : {}),
+    ...(rerank.timeoutMs !== undefined
+      ? { timeoutMs: rerank.timeoutMs }
+      : {}),
+  };
+}

src/services/capture.test.ts

@@ -0,0 +1,39 @@
+import { describe, expect, it } from "bun:test";
+import { toEmbedClientConfig } from "./capture.ts";
+import type { EngineConfig } from "../config.ts";
+
+// Regression for the capture path silently timing out at embed-client.ts's
+// default (10000ms) even when EMBED_TIMEOUT_MS was set: capture.ts used to
+// build its own EmbedClientConfig literal instead of going through the same
+// mapping search.ts used, and dropped timeoutMs. capture.ts now re-exports
+// the one shared mapping (engine-client-config.ts) — assert it carries
+// EngineConfig.embed.timeoutMs through on the capture path specifically.
+describe("capture path embed client config", () => {
+  it("carries EngineConfig.embed.timeoutMs through to the capture-path EmbedClientConfig", () => {
+    const embed: EngineConfig["embed"] = {
+      baseUrl: "http://embed.example",
+      model: "test-model",
+      apiStyle: "openai",
+      apiKey: undefined,
+      timeoutMs: 5000,
+    };
+
+    const embedClientConfig = toEmbedClientConfig(embed);
+
+    expect(embedClientConfig.timeoutMs).toBe(5000);
+  });
+
+  it("leaves timeoutMs undefined (so embed-client.ts's own default applies) when EngineConfig doesn't set one", () => {
+    const embed: EngineConfig["embed"] = {
+      baseUrl: "http://embed.example",
+      model: "test-model",
+      apiStyle: "openai",
+      apiKey: undefined,
+      timeoutMs: undefined,
+    };
+
+    const embedClientConfig = toEmbedClientConfig(embed);
+
+    expect(embedClientConfig.timeoutMs).toBeUndefined();
+  });
+});

src/services/capture.ts

@@ -1,4 +1,3 @@
-import { type } from "arktype";
 import { createHash } from "node:crypto";
 import { and, desc, eq } from "drizzle-orm";
 import type { Db, RawSql } from "../db/client.ts";
@@ -28,8 +27,9 @@ import type {
 import type { KnowledgeEdgeHint } from "../core/schemas/entity-edge.ts";
 import { createRawSqlClient } from "../core/embed-sql.ts";
 import { activateEmbedModel } from "../core/embed-model-registry.ts";
-import { EmbedClientConfigSchema, type EmbedClientConfig } from "../core/embed-client.ts";
+import type { EmbedClientConfig } from "../core/embed-client.ts";
 import { embedChunks, type EmbeddableChunk } from "../core/embed-worker.ts";
+import { toEmbedClientConfig } from "../core/engine-client-config.ts";
 
 type Tx = Parameters<Parameters<Db["transaction"]>[0]>[0];
 
@@ -462,21 +462,10 @@ async function captureInTransaction(
   );
 }
 
-// Built from the engine's own operator-configured embed endpoint — a trusted
-// URL, the same as KNOWLEDGE_DATABASE_URL.
-function toEmbedClientConfig(embed: EngineConfig["embed"]): EmbedClientConfig {
-  const candidate = {
-    baseUrl: embed.baseUrl,
-    modelId: embed.model,
-    apiStyle: embed.apiStyle,
-    ...(embed.apiKey !== undefined ? { apiKey: embed.apiKey } : {}),
-  };
-  const parsed = EmbedClientConfigSchema(candidate);
-  if (parsed instanceof type.errors) {
-    throw new Error(`Invalid embed client config: ${parsed.summary}`);
-  }
-  return parsed;
-}
+// Re-exported so tests can assert the capture path resolves its embed
+// client config through the one shared mapping (see engine-client-config.ts)
+// rather than a capture-local duplicate that could drop fields like timeoutMs.
+export { toEmbedClientConfig };
 
 // Embeds a version's freshly-inserted chunks and stores their vectors, after
 // the derivation transaction has already committed. Best-effort in the

src/services/search.ts

@@ -15,6 +15,10 @@ import {
   resolveActiveEmbedTable,
 } from "../core/embed-model-registry.ts";
 import { embedTexts, type EmbedClientConfig } from "../core/embed-client.ts";
+import {
+  toEmbedClientConfig,
+  toRerankClientConfig,
+} from "../core/engine-client-config.ts";
 import {
   rerankDocuments,
   RerankConfigError,
@@ -652,54 +656,11 @@ function applyBoosts(
   });
 }
 
-const VALID_EMBED_API_STYLES = new Set(["openai", "tei", "ollama"]);
-
-// The engine's `EngineConfig.embed.apiStyle` is a plain, operator-set string
-// (config.ts has no arktype gate on it); `EmbedClientConfig` requires the
-// literal union `embed-client.ts` dispatches on. Validated here, once, at
-// the trust boundary between config and the client — an invalid value is an
-// operator misconfiguration and must fail loudly, not silently degrade.
-// Built from the engine's own operator-configured embed endpoint — a trusted
-// URL, the same as KNOWLEDGE_DATABASE_URL.
-function toEmbedClientConfig(embed: EngineConfig["embed"]): EmbedClientConfig {
-  if (!VALID_EMBED_API_STYLES.has(embed.apiStyle)) {
-    throw new Error(
-      `Invalid EMBED_API_STYLE "${embed.apiStyle}" — must be one of: ${[...VALID_EMBED_API_STYLES].join(", ")}`,
-    );
-  }
-  return {
-    baseUrl: embed.baseUrl,
-    modelId: embed.model,
-    apiStyle: embed.apiStyle as EmbedClientConfig["apiStyle"],
-    ...(embed.apiKey !== undefined ? { apiKey: embed.apiKey } : {}),
-    ...(embed.timeoutMs !== undefined ? { timeoutMs: embed.timeoutMs } : {}),
-  };
-}
-
-// `EngineConfig.rerank` carries no `apiStyle` field — the engine currently
-// wires only a TEI-compatible cross-encoder endpoint (the locked default
-// model, `bge-reranker-v2-m3`, is TEI-servable); rerank apiStyle is hardcoded
-// `"tei"` below. Absent `baseUrl` => rerank is unconfigured => `undefined`,
-// same degrade-soft precedent as the embed config being absent upstream.
-// Built from the engine's own operator-configured rerank endpoint — a trusted
-// URL, the same as KNOWLEDGE_DATABASE_URL.
-export function toRerankClientConfig(
-  rerank: EngineConfig["rerank"],
-): RerankClientConfig | undefined {
-  if (!rerank.baseUrl) return undefined;
-  return {
-    baseUrl: rerank.baseUrl,
-    apiStyle: "tei",
-    ...(rerank.model !== undefined ? { model: rerank.model } : {}),
-    ...(rerank.apiKey !== undefined ? { apiKey: rerank.apiKey } : {}),
-    ...(rerank.maxDocChars !== undefined
-      ? { maxDocChars: rerank.maxDocChars }
-      : {}),
-    ...(rerank.timeoutMs !== undefined
-      ? { timeoutMs: rerank.timeoutMs }
-      : {}),
-  };
-}
+// Single shared EngineConfig -> client-config mapping, used by every
+// construction site (search, capture) so operator overrides like
+// EMBED_TIMEOUT_MS reach every code path uniformly. Re-exported here since
+// this is the module memory.ts already imports `toRerankClientConfig` from.
+export { toEmbedClientConfig, toRerankClientConfig };
 
 export interface HybridSearchDeps {
   db: Db;