@autorag/librarian 2.10.5 → 2.10.6

package/dist/memory/context.d.ts

@@ -1,4 +1,4 @@
-import type { Embedder } from "../embedding-runtime/gateway-embedder.js";
+import type { Embedder } from "../embedding/embedder.js";
 import type { JudgedEvidenceRecord } from "./judged-evidence.js";
 import type { RetrievalInsight, RetrievalMemory } from "./memory.js";
 import { type SimilarQuestion } from "./similar-queries.js";

package/dist/memory/similar-queries.d.ts

@@ -1,4 +1,4 @@
-import type { Embedder } from "../embedding-runtime/gateway-embedder.js";
+import type { Embedder } from "../embedding/embedder.js";
 import type { JudgedEvidenceRecord } from "./judged-evidence.js";
 /** How many similar past questions memory brings back for one question. */
 export declare const SIMILAR_QUESTION_LIMIT = 25;

package/dist/minsync/binary.d.ts

@@ -0,0 +1,21 @@
+import { type EnsureMinSyncBinaryOptions } from "./installer.js";
+export interface ResolveMinSyncBinaryOptions {
+    /** Workspace project root; the cached binary lives under `<root>/.autorag/bin`. */
+    readonly root: string;
+    /** Test/embedded override; CLI config intentionally does not persist this. */
+    readonly binaryPath?: string;
+    readonly autoInstall?: boolean;
+    readonly installer?: Omit<EnsureMinSyncBinaryOptions, "root">;
+    /** False for query paths: they use an existing binary or fail, and never wait on `cargo install`. */
+    readonly install?: boolean;
+}
+/** Resolve the `minsync` executable from PATH directories. Returns the first match or undefined. */
+export declare function lookupMinSyncInPath(env: NodeJS.ProcessEnv): string | undefined;
+/**
+ * Resolve the MinSync CLI from the user's global PATH first, then the workspace cache, and finally the
+ * verified auto-install fallback. Throws {@link MinSyncRequiredError}: MinSync is required, so there is
+ * no "no binary" value to degrade on.
+ */
+export declare function resolveMinSyncBinary(options: ResolveMinSyncBinaryOptions): Promise<string>;
+/** True when no `minsync` is resolvable right now without installing one. */
+export declare function isMinSyncBinaryMissing(options: Pick<ResolveMinSyncBinaryOptions, "root" | "binaryPath">): boolean;

package/dist/minsync/client.d.ts

@@ -1,18 +1,9 @@
-import type { EnsuredRuntime } from "../embedding-runtime/index.js";
 import type { MinSyncEmbedderConfig, MinSyncQueryHit, MinSyncSyncResult } from "./types.js";
-export declare const MINSYNC_OLLAMA_MIGRATION_MESSAGE = "This workspace uses the legacy 768-dimensional Ollama/TEI embedding path. Reindex explicitly, or pin an explicit Ollama/TEI embedder config before using the new default runtime.";
-export interface MinSyncRuntime {
-    ensureRuntime(options?: {
-        readonly profileId?: MinSyncEmbedderConfig["profile"];
-        readonly cachedOnly?: boolean;
-    }): Promise<EnsuredRuntime>;
-}
 export interface MinSyncClientOptions {
     readonly binaryPath: string;
     readonly workspacePath: string;
     readonly embedder?: MinSyncEmbedderConfig;
     readonly maxChunkSize?: number;
-    readonly runtime?: MinSyncRuntime;
 }
 /** MinSync's native query modes: vector, BM25, and hybrid. */
 export type MinSyncQueryMode = "vector" | "bm25" | "hybrid";
@@ -31,8 +22,12 @@ export declare class MinSyncClient {
     private readonly workspacePath;
     private readonly embedder;
     private readonly maxChunkSize;
-    private readonly runtime;
     constructor(options: MinSyncClientOptions);
+    /**
+     * What MinSync is told to embed with. No `id`, or a `native:` id, is MinSync's own Qwen3 engine, which
+     * MinSync provisions and runs (in-process, or through a `minsync serve` daemon it finds on its own).
+     * Anything else is an endpoint the operator configured, passed through untouched.
+     */
     private effectiveEmbedder;
     private readIdentity;
     private writeIdentity;
@@ -50,4 +45,10 @@ export declare class MinSyncClient {
     private checkAndSync;
     query(text: string, topK: number, mode?: MinSyncQueryMode, signal?: AbortSignal): Promise<readonly MinSyncQueryHit[]>;
     private spawn;
+    /**
+     * MinSync's `openai:` embedder reads its bearer token from `OPENAI_API_KEY` only. The operator names the
+     * variable that holds their key with `apiKeyEnv`, so that value is what the child receives as
+     * `OPENAI_API_KEY`. A local `native:` embedder needs no key and the environment is left as it is.
+     */
+    private childEnv;
 }

package/dist/minsync/index.d.ts

@@ -1,4 +1,4 @@
-export { MINSYNC_OLLAMA_MIGRATION_MESSAGE, MinSyncClient, type MinSyncClientOptions, MinSyncQueryError, type MinSyncQueryMode, } from "./client.js";
+export { MinSyncClient, type MinSyncClientOptions, MinSyncQueryError, type MinSyncQueryMode, } from "./client.js";
 export { configuredVectorDimension, DEFAULT_MINSYNC_EMBEDDER_CONFIG, DEFAULT_MINSYNC_EMBEDDER_DIMENSION, DEFAULT_MINSYNC_EMBEDDER_ID, MINSYNC_CONFIG_DIR, MINSYNC_CONFIG_FILE, type MinSyncEmbeddingIdentity, minSyncConfigPath, minSyncEmbeddingIdentityPath, rewriteEmbedderConfig, } from "./embedder-config.js";
 export { MinSyncRequiredError } from "./errors.js";
 export { type EnsureMinSyncBinaryOptions, ensureMinSyncBinary, executableName, fetchLatestMinSyncRelease, type InstalledMinSyncBinary, MINSYNC_VERSION, type MinSyncRelease, type MinSyncReleaseAsset, MinSyncReleaseError, selectReleaseAsset, } from "./installer.js";

package/dist/minsync/installer.d.ts

@@ -1,4 +1,4 @@
-export declare const MINSYNC_VERSION = "0.4.10";
+export declare const MINSYNC_VERSION = "0.4.11";
 export declare const MINSYNC_CARGO_INSTALL_TIMEOUT_MS: number;
 export interface MinSyncReleaseAsset {
     readonly name: string;

package/dist/minsync/method.d.ts

@@ -1,7 +1,6 @@
 import type { RetrievalMethod, RetrievalMethodDescriptor, RetrievalOptions, RetrievalResult } from "../retrieval/types.js";
 import type { MinSyncQueryMode } from "./client.js";
-import { type MinSyncRuntime } from "./client.js";
-import { type EnsureMinSyncBinaryOptions } from "./installer.js";
+import type { EnsureMinSyncBinaryOptions } from "./installer.js";
 import type { MinSyncEmbedderConfig, MinSyncSyncResult } from "./types.js";
 export interface MinSyncVectorMethodOptions {
     readonly root: string;
@@ -13,7 +12,6 @@ export interface MinSyncVectorMethodOptions {
     readonly embedder?: MinSyncEmbedderConfig;
     readonly maxChunkSize?: number;
     readonly mode?: MinSyncQueryMode;
-    readonly runtime?: MinSyncRuntime;
     /** Default `topK` used when the caller omits one. Default 50. */
     readonly defaultTopK?: number;
     /** Fetch cap applied when a scope narrows the query. Default 100. */
@@ -29,7 +27,6 @@ export declare class MinSyncVectorMethod implements RetrievalMethod {
     private readonly embedder;
     private readonly maxChunkSize;
     private readonly mode;
-    private readonly runtime;
     private readonly defaultTopK;
     private readonly scopedTopK;
     constructor(options: MinSyncVectorMethodOptions);
@@ -45,13 +42,8 @@ export declare class MinSyncVectorMethod implements RetrievalMethod {
     retrieve(query: string, options: RetrievalOptions): Promise<RetrievalResult[]>;
     private retrieveUnlocked;
     /**
-     * Resolve the MinSync CLI from the user's global PATH first, then the
-     * workspace cache, and finally the verified auto-install fallback. The
-     * auto-install runs only for `sync` (refresh); queries pass
-     * `install: false` so a user's turn never waits on `cargo install`.
-     *
-     * Returns the executable path, or throws {@link MinSyncRequiredError}:
-     * MinSync is required, so there is no "no binary" value to degrade on.
+     * Queries pass `install: false` so a user's turn never waits on `cargo install`; only `sync` (refresh)
+     * may install. See {@link resolveMinSyncBinary}.
      */
     protected resolveBinary(options?: {
         readonly install?: boolean;

package/dist/minsync/process.d.ts

@@ -8,4 +8,5 @@ export interface ProcessResult {
 export declare function spawnProcess(command: string, args: readonly string[], cwd: string, options?: {
     readonly timeoutMs?: number;
     readonly signal?: AbortSignal;
+    readonly env?: NodeJS.ProcessEnv;
 }): Promise<ProcessResult>;

package/dist/minsync/types.d.ts

@@ -1,8 +1,6 @@
 /** Embedder configuration for MinSync vector indexing. Shared by config and method layers. */
 import type { MinSyncLexicalLanguage } from "./lexical-language.js";
 export interface MinSyncEmbedderConfig {
-    /** Select an AutoRAG-owned embedding runtime profile. */
-    readonly profile?: "qwen3-embedding-0.6b";
     readonly id?: string;
     readonly baseUrl?: string;
     /** Environment variable name whose value holds the embedder API key. /^[A-Za-z_][A-Za-z0-9_]*$/ */

package/dist/retrieval/rerank.d.ts

@@ -24,6 +24,13 @@ export declare const DEFAULT_RERANK_API_KEY_ENV = "OPENROUTER_API_KEY";
  * chat LLMs, so the project's 2026+ model rule does not cover them.
  */
 export declare const DEFAULT_DAZZI_RERANK_MODEL = "qwen/qwen3-reranker-8b";
+/**
+ * Default per-request rerank timeout. A reranker that never answers would stall
+ * the fast answer. Measured against the hosted reranker, a 172-document pool
+ * takes ~15s and a 300-document pool ~45s, so 60s clears the largest baseline
+ * pool (3 x 100 candidates) with headroom.
+ */
+export declare const DEFAULT_RERANK_TIMEOUT_MS = 60000;
 /** Default number of merged chunks kept after reranking. */
 export declare const DEFAULT_RERANK_TOP_N = 25;
 /** Rerank providers {@link createReranker} can build. */
@@ -65,7 +72,7 @@ export interface Reranker {
      */
     rerank(query: string, results: readonly RetrievalResult[], options?: RerankOptions): Promise<RetrievalResult[]>;
 }
-/** Minimal rerank client shape, so tests can inject a fake without the SDK. */
+/** Minimal rerank client shape, so tests can inject a fake without network access. */
 export interface RerankClient {
     rerank(request: {
         requestBody: {
@@ -78,7 +85,7 @@ export interface RerankClient {
         signal?: AbortSignal;
     }): Promise<RerankResponse>;
 }
-/** Response body returned by OpenRouter's rerank router. */
+/** Rerank response as {@link OpenRouterReranker} consumes it. */
 export interface RerankResponse {
     readonly results: ReadonlyArray<{
         readonly index: number;
@@ -123,7 +130,7 @@ export interface CreateRerankerOptions extends OpenRouterRerankerOptions {
  */
 export declare function createReranker(config: CreateRerankerOptions | false | undefined): Reranker | undefined;
 /**
- * Reranker backed by an OpenRouter-compatible rerank router.
+ * Reranker backed by an OpenRouter-compatible rerank endpoint.
  *
  * OpenRouter is the default provider; the hosted Dazzi Models plan serves the
  * same wire shape at `<root>/v1/rerank`, so {@link createReranker} reuses this

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@autorag/librarian",
-  "version": "2.10.5",
+  "version": "2.10.6",
   "description": "Self-evolving librarian agent on Pi framework with pluggable retrieval methods",
   "license": "Apache-2.0",
   "repository": {
@@ -44,7 +44,6 @@
     "check": "biome check --write .",
     "lint": "biome check .",
     "typecheck": "tsc --noEmit",
-    "check:embedding-manifest": "node scripts/check-embedding-manifest.mjs",
     "supply-chain": "bun scripts/supply-chain/evaluate.ts gate --sbom sbom.local.cdx.json"
   },
   "engines": {
@@ -57,8 +56,8 @@
     "@earendil-works/pi-tui": "1.2.0",
     "@llblab/pi-telegram": "0.57.2",
     "@modelcontextprotocol/server": "^2.3.1",
-    "@openrouter/sdk": "^1.4.25",
     "chardet": "^2.2.0",
+    "discord.js": "14.27.0",
     "fast-xml-parser": "^5.11.2",
     "iconv-lite": "^0.7.3",
     "jev-use": "^0.8.0",
@@ -66,6 +65,7 @@
     "kordoc": "^4.21.30",
     "mailparser": "^3.9.38",
     "pdfjs-dist": "^6.4.299",
+    "pi-discord-remote": "0.2.4",
     "puppeteer-core": "^25.13.0",
     "slides-grab": "^1.5.2",
     "smol-toml": "1.9.1",

package/skills/autorag/SKILL.md

@@ -142,8 +142,6 @@ autorag index rebuild --yes --json
 autorag index reset --method parsed --yes --json
 autorag memory inspect --json
 autorag tui
-autorag serve --force
-autorag p2p policy list --json
 ```
 
 Prefer a full refresh so parsed mirrors, MinSync, Jikji, configured
@@ -153,9 +151,7 @@ aligned. `--method` accepts
 a `--method` name. Use `--method` only for deliberate narrowing. Scheduled
 maintenance should use non-daemon `autorag watch --once`, typically every
 1 hour, with the same config used by search and no overlapping runs.
-`autorag tui` is the shipped beta terminal UI. `autorag serve` / `autorag p2p`
-are opt-in SimpleX peer sharing (disabled until `p2p.enabled` is true, unless
-`--force`).
+`autorag tui` is the shipped beta terminal UI.
 
 Reset and rebuild commands remove only selected workspace `.autorag` indexes.
 Never target source documents. `memory inspect` is read-only and path-opaque.
@@ -169,8 +165,7 @@ Never target source documents. `memory inspect` is read-only and path-opaque.
 - Preserve real source mapping.
 - Prefer `--json --debug` when another agent consumes search output.
 - Do not invent CLI commands. `autorag --help` is the command list of record,
-  including the shipped `setup`, `gateway`, `models`, `serve`, `p2p`, `tui`, and `lite`
+  including the shipped `setup`, `models`, `tui`, and `lite`
   commands. `autorag <command> --help` prints that command's own flags, and
   `autorag --version` prints the installed version. See
-  `docs/embedding-runtime.md` for the local embedding runtime
-  and gateway setup.
+  `docs/minsync-setup.md#embedding-backends` for MinSync setup and embedding-backend selection.

package/skills/autorag-doctor/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: autorag-doctor
-description: Diagnose and repair a broken or half-working AutoRAG install so every configured source is both indexed and searchable. Checks AutoRAG, MinSync, Jikji, the embedding gateway, and every CLI-backed datasource (lazykatok, discrawl, slacrawl, wacrawl, telecrawl, notcrawl, qmd, mailcrawl, rclone, Spotlight), then fixes orphan locks, orphan processes, embedding dimension or identity mismatches, stale indexes, and missing setup. Use when search returns nothing or too little, refresh hangs or fails, a datasource disappeared from results, indexes look stale, the gateway will not start, or the user asks to check, diagnose, verify, or repair AutoRAG.
+description: Diagnose and repair a broken or half-working AutoRAG install so every configured source is both indexed and searchable. Checks AutoRAG, MinSync, Jikji, the embedding backend, and every CLI-backed datasource (lazykatok, discrawl, slacrawl, wacrawl, telecrawl, notcrawl, qmd, mailcrawl, rclone, Spotlight), then fixes orphan locks, orphan processes, embedding dimension or identity mismatches, stale indexes, and missing setup. Use when search returns nothing or too little, refresh hangs or fails, a datasource disappeared from results, indexes look stale, the embedder is unavailable, or the user asks to check, diagnose, verify, or repair AutoRAG.
 license: Apache-2.0
 ---
 
@@ -29,7 +29,7 @@ Always finish with the status table in [Report](#report).
 ```bash
 autorag status --json
 autorag health --json
-autorag gateway status --format json
+autorag setup --format json
 ```
 
 - `status` reports `state`, `stale`, `diagnostics`, and per-component state for
@@ -38,8 +38,12 @@ autorag gateway status --format json
 - `health` resolves the single search model and does one live completion probe.
   `--skip-probes` only proves config shape, never live access; do not claim a
   healthy model from it.
-- The gateway is on-demand: `stopped` is normal when nothing is embedding.
-  `unavailable` **while** a refresh is running is a real failure.
+- `setup --format json` reports `embedding.backend` (`remote`, `daemon`, or
+  `in-process`) and the resolved `provider`/`model`/`dimension` when a backend
+  is alive, or `embedding.reason` and a `remediation` when none is. It only
+  probes; it never starts a daemon. Semantic search needs the daemon (or a
+  remote endpoint) to be running: a backend that is absent is a real
+  `embedding-unavailable` state, not normal.
 
 Config lives at `--config`, `AUTORAG_CONFIG`, `$AUTORAG_HOME/config.json`, or
 `~/.autorag/config.json`. Read `searchPaths`, `workspacePath`, `minSync`,
@@ -129,19 +133,30 @@ autorag search "summarize the collection" --top-k 3 --json --debug
 
 ## 4. Repair playbook
 
-### Orphan lock
+### Stuck embedding daemon
 
-The embedding runtime keeps `embedding-runtime.lock`, `embedding-runtime.pid`,
-and `embedding-runtime.port` in `AUTORAG_HOME`. It reclaims them automatically
-when the recorded PID is dead; a `lock-conflict` means the PID is **alive**.
+AutoRAG owns no lock or pid file for embeddings. A running `minsync serve`
+publishes `host`, `port`, and `pid` under `~/.cache/minsync/serve/<instance>/`
+and removes them on a clean stop. A leftover file from a crashed daemon is
+harmless: AutoRAG only adopts a daemon whose `GET /health` answers as a loopback
+server of the wanted model, so a stale entry is skipped.
 
 ```bash
-autorag gateway status --format json
-autorag gateway stop
+ls ~/.cache/minsync/serve/* 2>/dev/null
+autorag setup --format json
 ```
 
-Only when `gateway stop` cannot clear it, and the recorded PID is confirmed
-dead (`kill -0 PID` fails), remove the three files by hand and retry.
+To diagnose an embedding backend that is not alive, `setup --format json` and
+`status` say so (`embedding.reason`, and the `embedding-unavailable`
+diagnostic). AutoRAG never starts the daemon: run `minsync serve
+--idle-timeout 0` by hand (or as a launchd/systemd service, see
+`docs/minsync-setup.md`) and read its own error, or press Ctrl+Shift+E in the
+TUI. When no daemon answers and the in-process addon
+(`@nomadamas/minsync-native`) is missing, that is the `embedder-unavailable`
+state. The daemon stops itself after `--idle-timeout` seconds (default 300), so
+a semantic failure after a quiet spell usually means it timed out. Stop an
+orphaned daemon by the pid from `~/.cache/minsync/serve/<instance>/pid` after
+confirming no live refresh holds it.
 
 Index locks are owned by their engines: MinSync/tantivy locks under
 `<workspace>/.autorag/`, and per-CLI locks such as
@@ -152,11 +167,12 @@ that holds it.
 ### Orphan process
 
 ```bash
-pgrep -fl 'autorag|autorag-gateway|minsync' | grep -v pgrep
+pgrep -fl 'autorag|minsync' | grep -v pgrep
 ```
 
-A refresh that was killed mid-run can leave the gateway or a `minsync` child
-alive. Stop the gateway with `autorag gateway stop` first; only kill a PID
+A refresh that was killed mid-run can leave a `minsync serve` daemon or a
+`minsync` child alive. A daemon idle-stops itself; stop a confirmed orphan by
+the pid from `~/.cache/minsync/serve/<instance>/pid`, and only kill a PID
 directly when it is confirmed orphaned (no parent CLI, no live refresh).
 Re-run `autorag status --json` afterwards to confirm `inFlight: false`.
 
@@ -167,16 +183,21 @@ were built with a different embedder than the configured one. Vectors of two
 different dimensions can never be compared, so the index must be rebuilt:
 
 ```bash
-autorag models prefetch --profile qwen3-embedding-0.6b
 autorag index rebuild --method minsync
 ```
 
-The current default is the local `native:Qwen/Qwen3-Embedding-0.6B` runtime at
-1024 dimensions. A workspace still pinned to the legacy 768-dimension
-Ollama/TEI path must be reindexed explicitly, or pinned to an explicit profile.
-Changing `embedder.dimension` in config **without** a rebuild leaves retrieval
-silently empty. Keep embeddings local; do not point the embedder at a remote
-endpoint to work around a local failure.
+The default is MinSync's local `Qwen/Qwen3-Embedding-0.6B` engine at 1024
+dimensions, reached through a running `minsync serve` daemon or, failing that,
+MinSync's in-process engine. Both report one identity (`provider: minsync`), so
+a store never re-embeds because the daemon appeared or went away. Vectors
+written by the removed local Q8_0 gateway engine are **not** compatible: their
+identity changed, so the memory store (`memory.json.vectors.db`), chat-sidecar
+semantic DBs, and the gist `vectors.json` re-embed on the next refresh, while a
+MinSync workspace created through the old gateway profile needs
+`autorag index rebuild --method minsync` (or
+`autorag index reset --method minsync --yes` then refresh). Changing the
+embedder's `dimension` in config **without** a rebuild leaves retrieval silently
+empty.
 
 ### Stale index
 
@@ -208,7 +229,7 @@ to retry those files once without re-indexing healthy ones.
   (`cargo install minsync`) or leave `minSync.autoInstall` on, then retry.
 - `jikji-unavailable`: the binary is missing and auto-install failed. It
   installs through cargo; verify the Rust toolchain, then `autorag refresh --method jikji` to retry.
-- `auth-error` / `rate-limited`: model or datasource credentials. Report the
+- `datasource-auth-error` / `datasource-rate-limited`: datasource credentials or quota. Report the
   missing environment-variable **name** and let the user supply it.
 - A configured datasource that returns nothing is a native store, connector, or
   query problem — every configured connection is searchable. Run its native
@@ -227,13 +248,12 @@ to retry those files once without re-indexing healthy ones.
 | `jikji-unavailable` | Jikji binary missing or install failed | Check cargo, retry refresh |
 | `everything-index-failed` | Windows Everything instance could not start or index; message carries ES exit code and stderr | Fix the reported cause, `autorag refresh --method everything --json` |
 | `embedding-identity-mismatch` | Indexed vectors use a different embedder | `autorag index rebuild --method minsync` |
-| `embedder-unavailable` | Embedding gateway or runtime down | `autorag gateway status --format json` |
-| `lock-conflict` | Another runtime holds the lock | `autorag gateway stop` |
+| `embedder-unavailable` | No remote endpoint answered, no daemon could start, and the in-process addon is missing or failed | Run `minsync serve` by hand to read its error, install MinSync (`cargo install minsync`) or `@nomadamas/minsync-native`, or configure a remote `minSync.embedder` |
 | `datasource-index-failed` | Backing CLI failed to index | Run that CLI's own doctor |
 | `datasource-empty` | Store has no matching content | Re-sync with the CLI, or accept as empty |
 | `unknown-datasource-skill` | Config name is not a known template | Fix the name or add `"type"` |
 | `retrieval-method-failed` | One method errored during the query | Read `--debug` diagnostics |
-| `auth-error` | Credentials missing or rejected | Report the env var name |
+| `datasource-auth-error` | Datasource credentials missing or rejected | Report the env var name |
 | `query-route-fallback` | Jev routing unavailable (often `OPENROUTER_API_KEY` unset); searched local with the original question | `test -n "$OPENROUTER_API_KEY"`; report the env var name, never its value |
 | `query-decomposition-failed` | Decomposition model call failed; searched the original question | Check `queryDecomposition.model` resolves (`autorag models list --provider openrouter`) |
 | `follow-up-check-fallback` | Jev post-fast-answer check unavailable; the run verified | Same as `query-route-fallback` |

package/skills/autorag-lite-setup/SKILL.md

@@ -168,14 +168,14 @@ AUTORAG_CONFIG=/absolute/path/to/.autorag/config.json autorag lite refresh --jso
   MinSync and Jikji auto-install during refresh (never during a query). If they
   are missing or broken, run a full refresh or return to setup rather than
   silently degrading to lexical-only search.
-- MinSync's default embedder is in-process native Qwen3 embeddings
-  (`native:Qwen/Qwen3-Embedding-0.6B`, 1024 dimensions, MinSync 0.4.10): no
-  embedder flags, no API key, and no external daemon (such as Ollama) are
-  needed, and no corpus text leaves the machine. For gateway-profiled
-  embeddings, prefetch with `autorag models prefetch --profile qwen3-embedding-0.6b`.
-  The legacy Ollama/TEI adapter path (EmbeddingGemma, 768 dimensions) is for
-  manual QA only. Override the embedder only when intentionally using a
-  different provider.
+- AutoRAG owns no embedding model; MinSync provisions
+  `Qwen/Qwen3-Embedding-0.6B` (1024 dimensions) in `~/.cache/minsync/models`. A
+  configured remote `minSync.embedder` is used first; otherwise a running
+  `minsync serve` daemon (started by `refresh`/`watch`, never by a search), and
+  failing that MinSync's in-process engine via the optional
+  `@nomadamas/minsync-native` addon. The local default needs no embedder flags
+  and no API key. Override the embedder only when intentionally using a remote
+  provider.
 - Exact duplicate exclusion during refresh is enabled by default via the
   external `dupey` CLI. Install dupey during setup when it is missing
   (`command -v dupey || cargo install dupey --locked`) and tell the user the feature is

package/skills/autorag-setup/SKILL.md

@@ -189,20 +189,28 @@ is impossible, refresh continues without this optimization and the user is
 told duplicate exclusion is off. Set `"excludeExactDuplicates": false` to
 index every copy.
 
-MinSync's default embedder is in-process native Qwen3 embeddings
-(`native:Qwen/Qwen3-Embedding-0.6B`, 1024 dimensions, MinSync 0.4.10). The
-default needs no embedder flags, no API key, and no external daemon (such as
-Ollama) — all embeddings run in-process locally and privately. For workspaces
-using the loopback llama-server gateway, prefetch the verified model with
-`autorag models prefetch --profile qwen3-embedding-0.6b` (or verify with
-`autorag models verify --profile qwen3-embedding-0.6b`).
-
-The legacy Ollama/TEI adapter path (EmbeddingGemma, 768 dimensions) is
-supported only for existing legacy workspaces and manual QA; do not use it as a
-fresh install default.
-
-Override the embedder only when intentionally using a different, for example
-remote, provider:
+AutoRAG owns no embedding model. MinSync provisions the embedder
+(`Qwen/Qwen3-Embedding-0.6B`, 1024 dimensions) in its own cache
+(`~/.cache/minsync/models`); there is no local model download and no bundled
+runtime. One shared embedder serves everything that embeds inside
+AutoRAG: MinSync indexing/query, memory similar-question search, the
+chat-crawler sidecars, the GitHub Gist vector index, and the endpoint external
+CLIs are pointed at. The default needs no embedder flags and no API key. It
+resolves in order:
+
+1. A configured remote endpoint (`minSync.embedder`), used for everything; no
+   daemon is started and no local model is loaded.
+2. A running `minsync serve` daemon (loopback only; MinSync 0.4.11+).
+   AutoRAG never starts it: run `minsync serve --idle-timeout 0` yourself, as
+   a launchd/systemd service (see `docs/minsync-setup.md`), or press
+   Ctrl+Shift+E in `autorag tui`. `autorag setup` only reports whether it is
+   alive.
+3. MinSync's in-process engine, through the optional Node-API addon
+   `@nomadamas/minsync-native` (same engine and weights, so bit-for-bit
+   identical vectors; not bundled). Without a daemon and without the addon,
+   embedding fails with an error naming the missing package.
+
+Override the embedder only when intentionally using a remote provider:
 
 ```bash
 autorag init \
@@ -213,9 +221,14 @@ autorag init \
   --embedder-batch-size 64
 ```
 
-Only store the environment-variable name, never its value. Dimension and batch
-size must be positive integers, and the dimension must match the embedder
-(default Qwen3 is 1024; legacy EmbeddingGemma is 768; `voyageai/voyage-4-lite` via OpenRouter is 1024).
+A remote `minSync.embedder` names an `id` (`openai:<model>`, `tei:<model>`, or
+any id together with a `baseUrl`) plus optional `apiKeyEnv` (the **name** of the
+environment variable holding the bearer token), `dimension`, `batchSize`, and
+`timeoutMs`. OpenAI-protocol endpoints POST `{baseUrl}/v1/embeddings`; a `tei:`
+endpoint POSTs `{baseUrl}/embed`; `openai:` with no `baseUrl` means
+`https://api.openai.com`. Only store the environment-variable name, never its
+value. `dimension` and `batchSize` must be positive integers, and `dimension`
+must match the endpoint (the local default is 1024).
 
 ### Jev routing and question decomposition (on by default)
 
@@ -336,7 +349,7 @@ keeps its mirrors and reports `search-root-unavailable`; see
 
 ## Probe and configure datasource skills (setup wizard)
 
-Use `autorag setup` to probe the local runtime, model profile, and known
+Use `autorag setup` to probe the embedding backend, MinSync, and known
 datasources automatically:
 
 ```bash
@@ -393,13 +406,15 @@ aliases with `"type": "<template>"`. Unknown names are skipped with an
 Tokens are environment-variable names, not raw secrets. CLI-backed connectors
 keep authentication in their external tool configuration.
 
-Mailcrawl must be installed separately (`@nomadamas/mailcrawl@0.2.0` or newer)
+Mailcrawl must be installed separately (`@nomadamas/mailcrawl@0.2.2` or newer)
 and configured through its own Himalaya account. AutoRAG runs its local `sync`
 and `index` lifecycle, then uses the mailcrawl CLI for BM25, semantic, or
 hybrid search. Do not use 0.1.3 or earlier: a no-op sync followed by `index`
-fails with `text array must be non-empty`. 0.2.0 defaults to the in-process
-native `Qwen/Qwen3-Embedding-0.6B` embedder and keeps vectors in LanceDB, so a
-cold cache makes the first `index` download ONNX weights and run for minutes.
+fails with `text array must be non-empty`. 0.2.2 embeds through the same
+backend as the rest of AutoRAG (a remote `minSync.embedder`, else the
+`minsync serve` daemon) instead of its own in-process model, and keeps vectors
+in LanceDB. Switching from the old native model rebuilds the semantic index on
+the next `index`. kakao needs `lazykatok` 0.3.3 or newer for the same reason.
 Use mailcrawl for Gmail, IMAP, and Maildir retrieval. The former Gmail REST
 datasource is removed.
 
@@ -415,7 +430,8 @@ autorag refresh --json
 autorag search "summarize the collection" --top-k 3 --json --debug
 ```
 
-- `setup` probes runtime health, model profile, and datasource readiness.
+- `setup` probes the embedding backend (`embedding.backend`), MinSync, and
+  datasource readiness.
 - `status` is model-free and path-opaque.
 - `health` resolves the single model, checks credential presence, and normally
   performs one live completion probe.

package/dist/agent/peer-target-tool.d.ts

@@ -1,19 +0,0 @@
-import type { AgentTool } from "@earendil-works/pi-agent-core";
-import { Type } from "typebox";
-export declare const RECOMMEND_PEER_TARGETS_TOOL_NAME = "recommend_peer_targets";
-declare const recommendPeerTargetsSchema: Type.TObject<{
-    query: Type.TString;
-}>;
-export interface PeerTargetRecommendation {
-    readonly alias: string;
-    readonly matchedTerms: readonly string[];
-    readonly displayName?: string;
-    readonly description?: string;
-}
-export interface RecommendPeerTargetsDetails {
-    readonly method: "recommend_peer_targets";
-    readonly resultCount: number;
-    readonly matches: readonly PeerTargetRecommendation[];
-}
-export declare function createRecommendPeerTargetsTool(workspacePath: string): AgentTool<typeof recommendPeerTargetsSchema, RecommendPeerTargetsDetails>;
-export {};

package/dist/agent/query-peer-tool.d.ts

@@ -1,51 +0,0 @@
-import type { AgentTool } from "@earendil-works/pi-agent-core";
-import { Type } from "typebox";
-import { type SimplexQueryState } from "../p2p/simplex-query-store.js";
-import type { SimplexTransport } from "../p2p/simplex-transport.js";
-import type { PeerQueryResponse } from "../p2p/wire.js";
-import { EvidenceLedger } from "./evidence-ledger.js";
-export declare const QUERY_PEER_AGENT_TOOL_NAME = "query_peer_agent";
-declare const queryPeerAgentSchema: Type.TObject<{
-    alias: Type.TString;
-    query: Type.TString;
-    topK: Type.TOptional<Type.TInteger>;
-    scope: Type.TOptional<Type.TString>;
-}>;
-export interface QueryPeerAgentResult {
-    readonly number: number;
-    readonly title: string;
-    readonly summary: string;
-    readonly source: string;
-    readonly excerpt: string;
-}
-export interface QueryPeerAgentDetails {
-    readonly method: "query_peer_agent";
-    readonly ok: boolean;
-    readonly alias: string;
-    readonly contactId?: number;
-    readonly pendingId?: string;
-    readonly expiresAt?: string;
-    readonly status?: "ok" | "rejected" | "pending";
-    readonly answer?: string;
-    readonly results?: readonly QueryPeerAgentResult[];
-    readonly diagnostics?: PeerQueryResponse["diagnostics"];
-    readonly fileCount?: number;
-    readonly message?: string;
-}
-export interface QueryPeerAgentToolOptions {
-    readonly workspacePath: string;
-    readonly openTransport: () => Promise<SimplexTransport>;
-    readonly fastTimeoutMs?: number;
-    readonly autoStart?: boolean;
-    readonly sessionId?: () => string | undefined;
-    readonly onResponse?: (state: SimplexQueryState, response: PeerQueryResponse) => void;
-    readonly onExpired?: (state: SimplexQueryState) => void;
-    /** Run evidence ledger; each peer result is registered so the final emit can cite it by id. */
-    readonly ledger?: EvidenceLedger;
-}
-export interface QueryPeerAgentTool extends AgentTool<typeof queryPeerAgentSchema, QueryPeerAgentDetails> {
-    readonly ready: Promise<void>;
-    close(): Promise<void>;
-}
-export declare function createQueryPeerAgentTool(options: QueryPeerAgentToolOptions): QueryPeerAgentTool;
-export {};

package/dist/cli/commands/gateway.d.ts

@@ -1,7 +0,0 @@
-import { type RuntimeStatus } from "../../embedding-runtime/index.js";
-import type { CommandContext } from "./types.js";
-export interface GatewayCommandDeps {
-    runtimeStatus?: () => Promise<RuntimeStatus>;
-    stopRuntime?: () => Promise<void>;
-}
-export declare function runGateway(ctx: CommandContext, deps?: GatewayCommandDeps): Promise<number>;

package/dist/cli/commands/p2p-policy.d.ts

@@ -1,2 +0,0 @@
-import type { CommandContext } from "./types.js";
-export declare function runP2pPolicy(ctx: CommandContext): number;

package/dist/cli/commands/p2p.d.ts

@@ -1,7 +0,0 @@
-import type { CommandContext } from "./types.js";
-/** Resolve the workspace used for P2P identity/policy, matching `autorag serve`. */
-export declare function resolveP2pWorkspace(ctx: CommandContext): string;
-/**
- * `autorag p2p` — SimpleX peer trust management.
- */
-export declare function runP2p(ctx: CommandContext): Promise<number>;