mlx-bun 0.5.0 → 1.0.0

package/src/serve/prompt-response-trace.ts

@@ -1,161 +0,0 @@
-import { runtimeValue } from "../runtime-config";
-
-export const P2R_TRACE_PREFIX = "[p2r] ";
-export const P2R_TRACE_VERSION = 1;
-
-export type P2RTraceOutcome = "success" | "error" | "abort";
-
-export type P2RTracePhase =
-  | "request.body_parse"
-  | "request.prompt_prepare"
-  | "completion.total"
-  | "completion.placement"
-  | "engine.request_wait"
-  | "engine.admission_wait"
-  | "engine.mixed_forward"
-  | "cache.lookup_restore"
-  | "cache.prefetch"
-  | "prefill.total"
-  | "prefill.batch_setup"
-  | "prefill.chunk"
-  | "prefill.row_sync"
-  | "prefill.forward"
-  | "prefill.evaluate"
-  | "prefill.checkpoint"
-  | "prefill.project"
-  | "prefill.complete"
-  | "prefill.companion"
-  | "prefill.kv_maintenance"
-  | "token_zero.total"
-  | "token_zero.forward"
-  | "token_zero.head"
-  | "token_zero.sample"
-  | "response.token_route"
-  | "response.first_write"
-  | "response.final_write";
-
-export type P2RTraceAttributes = Readonly<Record<
-  string,
-  string | number | boolean | null
->>;
-
-export interface P2RTraceEvent {
-  phase: P2RTracePhase;
-  startMs: number;
-  durationMs: number;
-  attributes?: P2RTraceAttributes;
-}
-
-export interface P2RTraceRecord {
-  version: typeof P2R_TRACE_VERSION;
-  traceId: string;
-  requestId: string;
-  route: string;
-  clock: "monotonic-ms";
-  outcome: P2RTraceOutcome;
-  /** Process-local monotonic origin; aligns concurrent requests in one server. */
-  startedAtMs?: number;
-  totalMs: number;
-  events: P2RTraceEvent[];
-  attributes?: P2RTraceAttributes;
-}
-
-type Clock = () => number;
-type Emit = (record: P2RTraceRecord) => void;
-
-export interface PromptResponseTraceOptions {
-  traceId: string;
-  requestId: string;
-  route: string;
-  clock?: Clock;
-  emit?: Emit;
-}
-
-/** Request-local diagnostic trace. Callers pass it explicitly because the
- * continuous scheduler interleaves rows and cannot safely use ambient async
- * context. No instance is created unless MLX_BUN_P2R_TRACE=1. */
-export class PromptResponseTrace {
-  readonly #startedAt: number;
-  readonly #events: P2RTraceEvent[] = [];
-  readonly #clock: Clock;
-  readonly #emit: Emit;
-  #finished = false;
-
-  readonly traceId: string;
-  readonly requestId: string;
-  readonly route: string;
-
-  constructor(options: PromptResponseTraceOptions) {
-    this.traceId = options.traceId;
-    this.requestId = options.requestId;
-    this.route = options.route;
-    this.#clock = options.clock ?? performance.now.bind(performance);
-    this.#emit = options.emit ?? ((record) => {
-      console.error(`${P2R_TRACE_PREFIX}${JSON.stringify(record)}`);
-    });
-    this.#startedAt = this.#clock();
-  }
-
-  get finished(): boolean {
-    return this.#finished;
-  }
-
-  /** Start one span. The returned closer is idempotent so finally blocks and
-   * early exits cannot double-record a phase. */
-  begin(phase: P2RTracePhase, attributes?: P2RTraceAttributes): () => void {
-    const startedAt = this.#clock();
-    let closed = false;
-    return () => {
-      if (closed || this.#finished) return;
-      closed = true;
-      const endedAt = this.#clock();
-      this.#events.push({
-        phase,
-        startMs: startedAt - this.#startedAt,
-        durationMs: endedAt - startedAt,
-        ...(attributes ? { attributes } : {}),
-      });
-    };
-  }
-
-  mark(phase: P2RTracePhase, attributes?: P2RTraceAttributes): void {
-    if (this.#finished) return;
-    this.#events.push({
-      phase,
-      startMs: this.#clock() - this.#startedAt,
-      durationMs: 0,
-      ...(attributes ? { attributes } : {}),
-    });
-  }
-
-  finish(
-    outcome: P2RTraceOutcome,
-    attributes?: P2RTraceAttributes,
-  ): P2RTraceRecord | null {
-    if (this.#finished) return null;
-    this.#finished = true;
-    const totalMs = this.#clock() - this.#startedAt;
-    const record: P2RTraceRecord = {
-      version: P2R_TRACE_VERSION,
-      traceId: this.traceId,
-      requestId: this.requestId,
-      route: this.route,
-      clock: "monotonic-ms",
-      outcome,
-      startedAtMs: this.#startedAt,
-      totalMs,
-      events: [...this.#events].sort((a, b) => a.startMs - b.startMs),
-      ...(attributes ? { attributes } : {}),
-    };
-    this.#emit(record);
-    return record;
-  }
-}
-
-export function createPromptResponseTrace(
-  options: PromptResponseTraceOptions,
-): PromptResponseTrace | undefined {
-  return runtimeValue("MLX_BUN_P2R_TRACE") === "1"
-    ? new PromptResponseTrace(options)
-    : undefined;
-}

package/src/serve/response-lease.ts

@@ -1,24 +0,0 @@
-import type { DisposableResource } from "../contracts/resources";
-
-/** Keep host activity owned through response EOF/error/cancel, without buffering
- * the response. The parent additionally drains worker cleanup before GPU jobs. */
-export function retainResponseLease(response: Response, lease: DisposableResource): Response {
-  if (!response.body) { lease.dispose(); return response; }
-  const reader = response.body.getReader();
-  let closed = false;
-  const release = () => {
-    if (closed) return;
-    closed = true;
-    reader.releaseLock(); lease.dispose();
-  };
-  return new Response(new ReadableStream<Uint8Array>({
-    async pull(controller) {
-      try {
-        const item = await reader.read();
-        if (item.done) { release(); controller.close(); }
-        else controller.enqueue(item.value);
-      } catch (error) { release(); controller.error(error); }
-    },
-    async cancel(reason) { try { await reader.cancel(reason); } finally { release(); } },
-  }), { status: response.status, statusText: response.statusText, headers: response.headers });
-}

package/src/serve/session-search.ts

@@ -1,326 +0,0 @@
-// Full-text search across web-chat session message BODIES (plan §9 Phase 3,
-// beat-matrix Axis 10/11 "the full-text BEAT": Claude's own session search is
-// title-only — a widely-cited annoyance — and local storage removes any
-// server-cost excuse to skip real body search).
-//
-// Reads the SAME session directory pi-web.ts's PiWebSession uses
-// (`~/.mlx-bun/sessions`, pi's own JSONL-per-session format —
-// `join(homedir(), ".mlx-bun", "sessions")`, see pi-web.ts's `sessionDir`
-// field and SessionManager.list's own file resolution in
-// @earendil-works/pi-coding-agent). Deliberately NOT a SessionManager import:
-// SessionManager.list() builds the full session tree (labels, branches,
-// compaction) for the chat runtime; this scanner only needs a flat
-// "role + text" reading of each line for substring matching, so a small,
-// independent, read-only line parser is both simpler and can never
-// accidentally mutate/lock a live session file.
-//
-// v1 is a linear, case-insensitive substring scan over every session file on
-// every request — no index. That's the right trade at personal-chat-corpus
-// scale (a few hundred to a few thousand sessions, read once per keystroke's
-// debounced request); if the corpus ever grows enough to matter, the upgrade
-// path is a background-built inverted index (token -> {sessionPath, lineNo})
-// refreshed on file mtime change, not a rewrite of this module's contract.
-//
-// JSONL-tolerant: a corrupt or partially-written line (the active session's
-// file can be mid-append) is skipped, never thrown — matching
-// SessionManager's own `parseSessionEntryLine` tolerance.
-
-import { readdir, readFile } from "node:fs/promises";
-import { homedir } from "node:os";
-import { join, resolve } from "node:path";
-
-/** Same default session directory pi-web.ts's PiWebSession uses. Exported so
- *  server.ts/tests can override it (tests always pass an explicit dir —
- *  never touch the real `~/.mlx-bun/sessions`). */
-export function defaultSessionDir(): string {
-  return join(homedir(), ".mlx-bun", "sessions");
-}
-
-export interface SearchMatch {
-  /** ±60-char plain-text window around the match, NOT html-escaped and NOT
-   *  html-highlighted — matches are reported as plain-text [start,end)
-   *  ranges into `snippet` so the caller (server-api consumer) chooses how
-   *  to render highlighting (the frontend escapes then re-inserts a <mark>,
-   *  matching the rest of this codebase's escape-then-restore convention —
-   *  see markdown.ts's code-span handling for the same pattern applied to a
-   *  different problem). */
-  snippet: string;
-  /** [start,end) offsets of the matched substring within `snippet`, one pair
-   *  per occurrence actually contained in this snippet window (usually one —
-   *  more only if two hits land within 60 chars of each other). */
-  ranges: Array<[number, number]>;
-  role: string;
-}
-
-export interface SessionSearchResult {
-  sessionPath: string;
-  sessionTitle: string;
-  matches: SearchMatch[];
-}
-
-interface ParsedLine {
-  type?: string;
-  message?: { role?: string; content?: unknown; timestamp?: number };
-  name?: string;
-}
-
-/** Best-effort JSON.parse of one JSONL line; never throws. Mirrors
- *  pi-coding-agent's own parseSessionEntryLine tolerance (skip malformed
- *  lines rather than fail the whole file). */
-function parseLine(line: string): ParsedLine | null {
-  const trimmed = line.trim();
-  if (!trimmed) return null;
-  try {
-    return JSON.parse(trimmed) as ParsedLine;
-  } catch {
-    return null;
-  }
-}
-
-/** Flatten a message's content (string | content-block array) to plain text,
- *  same shape pi-coding-agent's own extractTextContent handles. */
-function contentText(content: unknown): string {
-  if (typeof content === "string") return content;
-  if (Array.isArray(content)) {
-    return content
-      .filter((p): p is { type: string; text: string } =>
-        !!p && (p as { type?: string }).type === "text" && typeof (p as { text?: unknown }).text === "string")
-      .map((p) => p.text)
-      .join(" ");
-  }
-  return "";
-}
-
-const MAX_MATCHES_PER_SESSION = 10;
-const MAX_TOTAL_MATCHES = 50;
-const SNIPPET_RADIUS = 60;
-
-/** Build one snippet window (±SNIPPET_RADIUS chars) around a single match
- *  index in `text`, plus the match's own [start,end) offsets translated into
- *  the snippet's local coordinate space. Pure — independently testable. */
-export function buildSnippet(text: string, matchStart: number, matchLen: number): { snippet: string; range: [number, number] } {
-  const start = Math.max(0, matchStart - SNIPPET_RADIUS);
-  const end = Math.min(text.length, matchStart + matchLen + SNIPPET_RADIUS);
-  const prefix = start > 0 ? "…" : "";
-  const suffix = end < text.length ? "…" : "";
-  const snippet = prefix + text.slice(start, end) + suffix;
-  const localStart = matchStart - start + prefix.length;
-  return { snippet, range: [localStart, localStart + matchLen] };
-}
-
-/** Build a case-folded copy of `text` PLUS a parallel index that maps every
- *  position in the folded string back to the position it came from in the
- *  original — needed because `String.prototype.toLowerCase()` is not
- *  length-preserving for every character (e.g. Turkish dotted capital I,
- *  U+0130, folds to a 2-code-unit "i̇": `"İ".length === 1` but
- *  `"İ".toLowerCase().length === 2`). Folding the WHOLE string at once (the
- *  original approach) desyncs any match offset that falls after such a
- *  character: `text.toLowerCase().indexOf(...)` returns an index into the
- *  now-longer folded string, but callers then slice the shorter original
- *  `text` at that same numeric offset. Folding one character at a time and
- *  recording, for each folded code unit, which original index it expanded
- *  from keeps the scan itself simple (still a plain indexOf loop) while
- *  making offset translation exact for every fold, not just the common
- *  1:1 case. */
-function foldWithIndex(text: string): { folded: string; index: number[] } {
-  let folded = "";
-  const index: number[] = [];
-  for (let i = 0; i < text.length; i++) {
-    const piece = text[i]!.toLowerCase();
-    for (let j = 0; j < piece.length; j++) index.push(i);
-    folded += piece;
-  }
-  return { folded, index };
-}
-
-/** Find every occurrence of `qLower` (already lowercased) in `textLower`,
- *  returning [start, len] pairs translated back into the ORIGINAL
- *  (pre-fold) string's coordinate space via `index` (see foldWithIndex) —
- *  callers slice the original `text`, not `textLower`, so both the start
- *  offset AND the matched span's length must live in that space. `len` is
- *  computed from the first and last folded code unit's original index
- *  (not `qLower.length`) so a match whose folded span itself contains a
- *  length-expanding character (e.g. the query matches text spanning a
- *  Turkish dotted İ) still yields the exact original substring, not a
- *  same-length-as-query slice that's now off by the expansion. Pure
- *  substring scan — no regex (query is user-controlled; escaping for
- *  RegExp is an unforced footgun v1 doesn't need since substring search is
- *  the stated v1 bar). */
-function findOccurrences(textLower: string, qLower: string, index: readonly number[]): Array<[number, number]> {
-  if (!qLower) return [];
-  const offsets: Array<[number, number]> = [];
-  let i = 0;
-  for (;;) {
-    const idx = textLower.indexOf(qLower, i);
-    if (idx === -1) break;
-    const start = index[idx] ?? idx;
-    const lastFoldedIdx = idx + qLower.length - 1;
-    const end = index[lastFoldedIdx] ?? lastFoldedIdx;
-    offsets.push([start, end - start + 1]);
-    i = idx + qLower.length;
-  }
-  return offsets;
-}
-
-/** Scan one session file's raw JSONL content for `query` (case-insensitive
- *  substring), returning up to MAX_MATCHES_PER_SESSION per session. Pure
- *  given the file content string — split out from the disk read so it's
- *  unit-testable without touching the filesystem. */
-export function searchSessionContent(content: string, query: string): SessionSearchResult | null {
-  const qLower = query.toLowerCase();
-  if (!qLower) return null;
-
-  const lines = content.split("\n");
-  let sessionPath = "";
-  let title = "";
-  let firstUserMessage = "";
-  const matches: SearchMatch[] = [];
-
-  for (const line of lines) {
-    const entry = parseLine(line);
-    if (!entry) continue;
-    if (entry.type === "session_info" && typeof entry.name === "string" && entry.name.trim()) {
-      title = entry.name.trim();
-      continue;
-    }
-    if (entry.type !== "message" || !entry.message) continue;
-    const role = entry.message.role;
-    if (role !== "user" && role !== "assistant") continue;
-    const text = contentText(entry.message.content);
-    if (!text) continue;
-    if (!firstUserMessage && role === "user") firstUserMessage = text;
-
-    if (matches.length >= MAX_MATCHES_PER_SESSION) continue;
-    // Fold with an index (not a plain text.toLowerCase()) so match offsets
-    // stay valid against the ORIGINAL `text` that buildSnippet slices below
-    // — a length-expanding fold (e.g. Turkish dotted İ -> 2-code-unit "i̇")
-    // would otherwise desync a whole-string lowercase's offsets from the
-    // original string's coordinates for any match after it (see
-    // foldWithIndex's comment).
-    const { folded: textLower, index } = foldWithIndex(text);
-    const offsets = findOccurrences(textLower, qLower, index);
-    for (const [off, len] of offsets) {
-      if (matches.length >= MAX_MATCHES_PER_SESSION) break;
-      const { snippet, range } = buildSnippet(text, off, len);
-      matches.push({ snippet, ranges: [range], role });
-    }
-  }
-
-  if (matches.length === 0) return null;
-  return {
-    sessionPath,
-    sessionTitle: (title || firstUserMessage || "New chat").slice(0, 80),
-    matches,
-  };
-}
-
-/** Scan every `.jsonl` file under `sessionDir` for `query`. Read-only,
- *  JSONL-tolerant (a corrupt/mid-write file contributes whatever valid lines
- *  it has rather than aborting the whole scan), capped at
- *  MAX_TOTAL_MATCHES results across all sessions. Sessions are visited in
- *  directory-listing order (pi names files `<timestamp>_<id>.jsonl`, so this
- *  is naturally oldest-first — good enough for v1; result ordering isn't a
- *  relevance rank, just "found in these sessions"). */
-export async function searchSessions(sessionDir: string, query: string): Promise<SessionSearchResult[]> {
-  const trimmed = query.trim();
-  if (!trimmed) return [];
-
-  let files: string[];
-  try {
-    files = (await readdir(sessionDir)).filter((f) => f.endsWith(".jsonl"));
-  } catch {
-    return []; // no session dir yet — never throw, just no results
-  }
-
-  const results: SessionSearchResult[] = [];
-  let total = 0;
-  for (const file of files) {
-    if (total >= MAX_TOTAL_MATCHES) break;
-    const path = join(sessionDir, file);
-    let content: string;
-    try {
-      content = await readFile(path, "utf8");
-    } catch {
-      continue; // deleted mid-scan, permissions, etc. — skip, don't throw
-    }
-    const result = searchSessionContent(content, trimmed);
-    if (!result) continue;
-    result.sessionPath = path;
-    const remaining = MAX_TOTAL_MATCHES - total;
-    if (result.matches.length > remaining) result.matches = result.matches.slice(0, remaining);
-    total += result.matches.length;
-    results.push(result);
-  }
-  return results;
-}
-
-/* ────────────────────────────────────────────────────────────────────
-   REST handlers — pure (URL) => Response, no `ctx` dependency, same
-   convention as src/memory/rest.ts / src/hub-rest.ts (dispatched from
-   server.ts by path+method; unit-tested directly in
-   tests/serve/session-search.test.ts without booting a model context).
-   ──────────────────────────────────────────────────────────────────── */
-
-function jsonOk<T extends object>(body: T, init?: ResponseInit): Response {
-  return Response.json({ ok: true, ...body }, init);
-}
-
-function jsonErr(error: string, status = 400): Response {
-  return Response.json({ ok: false, error }, { status });
-}
-
-/** True if `path` resolves under `sessionDir` — same guard shape as
- *  pi-web.ts's private `isUnderSessionDir` (root itself or a path prefixed
- *  by `root + "/"`), reimplemented here (not imported) since pi-web.ts's
- *  copy is a private method on PiWebSession, not a standalone export, and
- *  duplicating this one three-line check is cheaper than changing that
- *  class's visibility for two read-only REST routes. */
-export function isUnderSessionDir(path: string, sessionDir: string): boolean {
-  const root = resolve(sessionDir);
-  const p = resolve(path);
-  return p === root || p.startsWith(root + "/");
-}
-
-// ---- GET /api/sessions/search?q= ----------------------------------------
-
-/** `sessionDirOverride` lets tests point at a tmp fixture dir instead of the
- *  real `~/.mlx-bun/sessions` (never touched by tests). */
-export async function handleSessionsSearch(url: URL, sessionDirOverride?: string): Promise<Response> {
-  const q = (url.searchParams.get("q") ?? "").trim();
-  if (!q) return jsonErr("q is required");
-  const dir = sessionDirOverride ?? defaultSessionDir();
-  const results = await searchSessions(dir, q);
-  return jsonOk({ results });
-}
-
-// ---- GET /api/sessions/export?path= --------------------------------------
-
-/** Returns the raw session JSONL, parsed into an array of entries (one per
- *  line) — the frontend's non-open-session export path reads this to
- *  render Markdown/JSON client-side without a second, divergent parser.
- *  `path` is validated under sessionDir exactly like pi-web's
- *  isUnderSessionDir — a path outside the session directory 400s before
- *  touching the filesystem. */
-export async function handleSessionsExport(url: URL, sessionDirOverride?: string): Promise<Response> {
-  const path = (url.searchParams.get("path") ?? "").trim();
-  if (!path) return jsonErr("path is required");
-  const dir = sessionDirOverride ?? defaultSessionDir();
-  if (!isUnderSessionDir(path, dir)) return jsonErr("path must be under the session directory", 403);
-  let content: string;
-  try {
-    content = await readFile(path, "utf8");
-  } catch {
-    return jsonErr("session not found", 404);
-  }
-  const entries: unknown[] = [];
-  for (const line of content.split("\n")) {
-    const trimmed = line.trim();
-    if (!trimmed) continue;
-    try {
-      entries.push(JSON.parse(trimmed));
-    } catch {
-      // skip malformed lines — same JSONL tolerance as the search scanner
-    }
-  }
-  return jsonOk({ path, entries });
-}

package/src/serve/static-routes.ts

@@ -1,101 +0,0 @@
-import { readFileSync } from "node:fs";
-
-export interface StaticRouteAssets {
-  appPage: string;
-  appJs: string;
-  hljsJs: string;
-  hljsCss: string;
-  manifest: string;
-  iconSvg: string;
-  serviceWorker: string;
-  curvePage: string;
-}
-
-const ASSET_HEADERS = { "cache-control": "public, max-age=3600" } as const;
-const LEGACY_PATHS = new Set(["/status", "/chat", "/quantize", "/finetune", "/dataset"]);
-
-/** Serve the web shell and its static companion pages. Returns null when the
- * request belongs to another route family. */
-export function handleStaticRoute(
-  url: URL,
-  request: Request,
-  assets: StaticRouteAssets,
-): Response | null {
-  if (request.method !== "GET") return null;
-
-  switch (url.pathname) {
-    case "/":
-      return new Response(assets.appPage, {
-        headers: { "content-type": "text/html; charset=utf-8" },
-      });
-    case "/assets/hljs.js":
-      return new Response(assets.hljsJs, {
-        headers: { "content-type": "text/javascript; charset=utf-8", ...ASSET_HEADERS },
-      });
-    case "/assets/hljs.css":
-      return new Response(assets.hljsCss, {
-        headers: { "content-type": "text/css; charset=utf-8", ...ASSET_HEADERS },
-      });
-    case "/assets/app.js":
-      return new Response(assets.appJs, {
-        headers: { "content-type": "text/javascript; charset=utf-8", ...ASSET_HEADERS },
-      });
-    case "/manifest.webmanifest":
-      return new Response(assets.manifest, {
-        headers: { "content-type": "application/manifest+json; charset=utf-8", ...ASSET_HEADERS },
-      });
-    case "/assets/icon.svg":
-      return new Response(assets.iconSvg, {
-        headers: { "content-type": "image/svg+xml; charset=utf-8", ...ASSET_HEADERS },
-      });
-    case "/sw.js":
-      return new Response(assets.serviceWorker, {
-        headers: { "content-type": "text/javascript; charset=utf-8", "cache-control": "no-store" },
-      });
-    case "/curves": {
-      let html = assets.curvePage;
-      try {
-        html = readFileSync(new URL("../lab/curve/curve-designer.html", import.meta.url), "utf8");
-      } catch {
-        // Compiled binaries use the embedded copy.
-      }
-      return new Response(html, {
-        headers: { "content-type": "text/html; charset=utf-8", "cache-control": "no-store" },
-      });
-    }
-    case "/dag":
-      try {
-        const html = readFileSync(
-          new URL("./assets/training-inference-map.html", import.meta.url),
-          "utf8",
-        );
-        return new Response(html, {
-          headers: { "content-type": "text/html; charset=utf-8", "cache-control": "no-store" },
-        });
-      } catch {
-        return new Response(
-          "DAG map artifact not found",
-          { status: 404 },
-        );
-      }
-    case "/curve-terrain":
-      try {
-        const html = readFileSync(
-          new URL("./assets/curve-terrain.html", import.meta.url),
-          "utf8",
-        );
-        return new Response(html, {
-          headers: { "content-type": "text/html; charset=utf-8", "cache-control": "no-store" },
-        });
-      } catch {
-        return new Response(
-          "curve terrain artifact not found",
-          { status: 404 },
-        );
-      }
-  }
-
-  return LEGACY_PATHS.has(url.pathname)
-    ? Response.redirect(`/#${url.pathname}`, 302)
-    : null;
-}

package/src/serve/transcription-server.ts

@@ -1,64 +0,0 @@
-// Whisper-only server: `mlx-bun serve <whisper checkpoint>` starts this
-// instead of the chat server. Routes: the audio routes, /v1/models,
-// /v1, /health, /stats. No chat machinery, no prompt cache — the process
-// idles at a few tens of MB with the weights paged out (see
-// TranscriptionService). This is the footprint sotto wants from a
-// dictation backend.
-
-import type { Server } from "bun";
-import pkgJson from "../../package.json" with { type: "json" };
-import { handleAudioRoute } from "./audio-routes";
-import { TranscriptionService } from "./transcription-service";
-
-const pkgVersion = (pkgJson as { version: string }).version;
-
-export interface TranscriptionServerOptions {
-  modelDir: string;
-  modelId: string;
-  port?: number;
-  hostname?: string;
-  unixSocket?: string;
-  idleUnloadSec?: number;
-  resident?: boolean;
-  /** Load the weights before serving (default: on first request). */
-  preload?: boolean;
-}
-
-export async function createTranscriptionServer(opts: TranscriptionServerOptions): Promise<{ server: Server<unknown>; service: TranscriptionService }> {
-  const service = new TranscriptionService({
-    modelDir: opts.modelDir, modelId: opts.modelId, idleUnloadSec: opts.idleUnloadSec, resident: opts.resident,
-  });
-  if (opts.preload) await service.ensureLoaded();
-  const startedAt = Date.now();
-  const server = Bun.serve({
-    ...(opts.unixSocket
-      ? ({ unix: opts.unixSocket } as unknown as Record<string, never>)
-      : { port: opts.port ?? 0, ...(opts.hostname ? { hostname: opts.hostname } : {}) }),
-    idleTimeout: 0,
-    async fetch(request) {
-      const url = new URL(request.url);
-      const audio = await handleAudioRoute(url, request, { service: async () => service });
-      if (audio) return audio;
-      if (request.method === "GET" && url.pathname === "/health")
-        return Response.json({ status: "ok", transcription: { ...service.stats, sessions: service.sessionCount } });
-      if (request.method === "GET" && url.pathname === "/stats")
-        return Response.json({ model: opts.modelId, transcription: service.stats, uptime_s: (Date.now() - startedAt) / 1000 });
-      if (request.method === "GET" && url.pathname === "/v1")
-        return Response.json({
-          name: "mlx-bun", version: pkgVersion, model: opts.modelId, mode: "transcription",
-          endpoints: ["POST /v1/audio/transcriptions", "POST /v1/audio/translations", "POST /v1/audio/sessions", "POST /v1/audio/sessions/:id/audio", "POST /v1/audio/sessions/:id/finish", "DELETE /v1/audio/sessions/:id", "POST /admin/transcription/unload", "GET /v1/models", "GET /health", "GET /stats"],
-        });
-      if (request.method === "GET" && (url.pathname === "/v1/models" || url.pathname.startsWith("/v1/models/")))
-        return Response.json({
-          object: "list",
-          data: [{
-            id: opts.modelId, object: "model", created: Math.floor(startedAt / 1000), owned_by: "mlx-bun",
-            transcription: true, resident: service.resident,
-            capabilities: { transcription: true, translation: true, chat_completions: false },
-          }],
-        });
-      return Response.json({ error: { message: "not found" } }, { status: 404 });
-    },
-  });
-  return { server, service };
-}