@atomic-chat/core 0.5.1

package/dist/runtime/mlx/runtime.js

@@ -0,0 +1,223 @@
+/**
+ * MLX models, served one process per model by the bundled `mlx-server` (mlx-vlm), Apple Silicon.
+ *
+ * Ported from: tauri-plugin-mlx/src/commands.rs (`load_mlx_model_impl`, unload, cleanup) and
+ * extensions/mlx-extension/src/index.ts (`performLoad`, `handleAutoIncreaseCtx`).
+ *
+ * The extension planned a load (settings, drafter, context) and the plugin ran it; here both halves
+ * are one. The core keeps the extension's model catalogue where it was — listing, importing and
+ * downloading MLX models stays in the app — and takes over what needs a process: load, unload,
+ * growing the context, restarting.
+ *
+ * Deliberate differences, each for a reason the app had no way to act on:
+ *  - loading a loaded model answers with its session, and unloading an unloaded one succeeds — the
+ *    shared semantics of every core runtime (the extension threw in both cases);
+ *  - a drafter that is switched on but not on disk is not downloaded at load time: the app still
+ *    downloads it before asking (its adapter passes `draft_model_path`), and a CLI load runs without
+ *    it, as the extension already did whenever that download failed;
+ *  - a session that dies is removed and reported with `session:died`; the plugin kept it until a
+ *    chat request found the dead port.
+ */
+import { existsSync } from 'node:fs';
+import { join } from 'node:path';
+import { resolveEagle3Draft, resolveMlxDflashDraft, resolveMtpDraft } from '../../speculative/index.js';
+import { DEFAULT_CTX_LEN, computeNextCtxLen } from '../llamacpp/ctx-ladder.js';
+import { closeLogStream, openLogStream, randomFreePort, SidecarTable, spawnAndAwaitReady, throwIfLoadCancelled, } from '../shared/index.js';
+import { buildMlxServerArgs, normalizeMlxModelPath } from './args.js';
+import { asNumber, buildMlxConfig, selectMlxDraftSettings } from './config.js';
+import { MLX_STDERR_READY_MARKERS, MLX_STDOUT_READY_MARKERS, classifyMlxStderr, describeMlxExit, mlxBinaryMissing, mlxModelMissing, mlxTimeout, } from './errors.js';
+import { readMlxMaxCtxTrain, repairLegacyShardName, resolveLocalDraftDir } from './model-files.js';
+export const MLX_SERVER_BINARY = 'mlx-server';
+/** The extension's `timeout` default, in seconds. */
+export const MLX_DEFAULT_TIMEOUT_SECS = 600;
+export class MlxRuntime {
+    options;
+    table;
+    emit;
+    constructor(options) {
+        this.options = options;
+        this.emit = options.emit ?? (() => { });
+        this.table = new SidecarTable({
+            provider: 'mlx',
+            instanceId: options.instanceId,
+            journal: options.journal,
+            emit: this.emit,
+            engine: 'MLX',
+            unloadGraceMs: 5000,
+            shutdownGraceMs: 2000,
+            describeExit: (exit, stderr) => describeMlxExit(exit, stderr),
+        });
+    }
+    binaryPath() {
+        return this.options.resourcesDir ? join(this.options.resourcesDir, MLX_SERVER_BINARY) : undefined;
+    }
+    list() {
+        return this.table.list();
+    }
+    findSession(modelId) {
+        return this.table.findSession(modelId);
+    }
+    getLoadedModels() {
+        return this.table.getLoadedModels();
+    }
+    isLoading(modelId) {
+        return this.table.isLoading(modelId);
+    }
+    /** The context a loaded session runs with. */
+    getCtxSize(modelId) {
+        return this.table.get(modelId)?.extra.ctxSize;
+    }
+    async load(modelId, opts = {}) {
+        this.table.assertRunning();
+        return this.table.load(modelId, () => this.start(modelId, opts), opts.signal);
+    }
+    async start(modelId, opts) {
+        throwIfLoadCancelled(opts.signal);
+        const settings = await this.options.readSettings();
+        const overrides = { ...(opts.overrides ?? {}) };
+        const cfg = { ...settings, ...overrides };
+        const isEmbedding = opts.isEmbedding ?? false;
+        const autoUnload = cfg['auto_unload'] === undefined || cfg['auto_unload'] === true || cfg['auto_unload'] === 'true';
+        if (autoUnload && !isEmbedding && !(opts.bypassAutoUnload ?? false)) {
+            // Loads are queued one at a time, so nothing else is starting: everything loaded goes.
+            for (const loaded of this.table.getLoadedModels()) {
+                throwIfLoadCancelled(opts.signal);
+                await this.table.unload(loaded);
+            }
+        }
+        this.table.assertRunning();
+        throwIfLoadCancelled(opts.signal);
+        const registry = this.options.registry;
+        const yml = await repairLegacyShardName(registry, modelId, await registry.read(modelId), (message) => this.emit('core:log', { level: 'warn', msg: message }));
+        const { modelPath } = registry.resolvePaths(yml);
+        const exe = opts.exePath ?? this.binaryPath();
+        if (!exe || !this.exists(exe))
+            throw mlxBinaryMissing(exe ?? join('<resources-dir>', MLX_SERVER_BINARY));
+        if (!this.exists(modelPath))
+            throw mlxModelMissing(modelPath);
+        const port = opts.port ?? (await randomFreePort(this.table.usedPorts()));
+        const maxCtxTrain = await readMlxMaxCtxTrain(modelPath);
+        throwIfLoadCancelled(opts.signal);
+        const draft = selectMlxDraftSettings(cfg);
+        const anyDrafter = Boolean(cfg.dflash_enabled) || Boolean(cfg.mtp_enabled) || Boolean(cfg.eagle3_enabled);
+        if (anyDrafter && !draft.draftPath) {
+            const restored = await (this.options.resolveDraft ?? ((kind, id) => this.localDraft(kind, id)))(draft.draftKind, modelId).catch(() => undefined);
+            if (restored) {
+                draft.draftPath = restored;
+                overrides['draft_model_path'] = restored;
+            }
+            else {
+                this.emit('core:log', {
+                    level: 'warn',
+                    msg: `${modelId} has ${draft.draftKind}_enabled=true but no ${draft.draftKind} drafter on disk; loading without drafter`,
+                });
+            }
+        }
+        const config = buildMlxConfig(cfg, draft, { ...(maxCtxTrain !== undefined ? { maxCtxTrain } : {}) });
+        const args = buildMlxServerArgs(modelPath, port, config);
+        const timeoutSecs = opts.timeoutSecs ?? asNumber(cfg['timeout']) ?? MLX_DEFAULT_TIMEOUT_SECS;
+        const env = {};
+        for (const [key, value] of Object.entries(this.options.baseEnv ?? process.env))
+            if (value !== undefined)
+                env[key] = value;
+        env['MLX_VLM_SINGLE_MODEL'] = '1';
+        throwIfLoadCancelled(opts.signal);
+        const logStream = opts.logPath ? await openLogStream(opts.logPath, 'MLX') : undefined;
+        let started;
+        try {
+            started = await (this.options.spawn ?? spawnAndAwaitReady)({ exe, args, env }, {
+                timeoutMs: timeoutSecs * 1000,
+                signal: this.table.signal,
+                ...(opts.signal ? { cancelSignal: opts.signal } : {}),
+                streamReadyMarkers: { stdout: MLX_STDOUT_READY_MARKERS, stderr: MLX_STDERR_READY_MARKERS },
+                // The plugin killed a server that did not come up at once, without a grace period.
+                timeoutGraceMs: 0,
+                timeoutError: (stderr) => mlxTimeout(timeoutSecs, stderr),
+                classifyExit: (_exit, stderr) => classifyMlxStderr(stderr),
+                onLine: (stream, line) => {
+                    logStream?.write(`[${stream}] ${line}\n`);
+                    if (opts.verbose)
+                        this.emit('core:log', { level: 'debug', msg: `[mlx/${modelId}][${stream}] ${line}` });
+                },
+            });
+        }
+        catch (error) {
+            await closeLogStream(logStream);
+            throw error;
+        }
+        return this.table.adopt({
+            info: {
+                pid: started.process.pid,
+                port,
+                model_id: modelId,
+                model_path: normalizeMlxModelPath(modelPath),
+                is_embedding: isEmbedding,
+                api_key: '',
+            },
+            process: started.process,
+            exe,
+            extra: { ctxSize: config.ctx_size, maxCtxTrain, overrides },
+            logStream,
+        }, opts.signal);
+    }
+    async localDraft(kind, modelId) {
+        const resolution = kind === 'mtp'
+            ? resolveMtpDraft(modelId)
+            : kind === 'eagle3'
+                ? resolveEagle3Draft(modelId)
+                : resolveMlxDflashDraft(modelId);
+        return resolution ? resolveLocalDraftDir(this.options.layout, resolution.repo) : undefined;
+    }
+    unload(modelId) {
+        this.table.assertRunning();
+        return this.table.unload(modelId);
+    }
+    /**
+     * Reload with the next context up the ladder, because a request did not fit — clamped to what the
+     * model was trained for, where the reload would change nothing (`at_max`).
+     */
+    async autoIncreaseCtx(modelId, reason = 'ctx-overflow') {
+        this.table.assertRunning();
+        const session = this.table.get(modelId);
+        if (!session)
+            return { ok: false, reason: 'not-loaded' };
+        const current = session.extra.ctxSize || DEFAULT_CTX_LEN;
+        const max = session.extra.maxCtxTrain;
+        const next = computeNextCtxLen(current, max);
+        if (next <= current)
+            return {
+                ok: false,
+                reason: 'at_max',
+                current_ctx_len: current,
+                ...(max !== undefined ? { max_ctx_len: max } : {}),
+            };
+        const info = await this.reload(modelId, { ...session.extra.overrides, ctx_size: next }, 'auto_increase_ctx');
+        this.emit('session:ctx-increased', { provider: 'mlx', modelId, oldCtx: current, newCtx: next, reason });
+        return { ok: true, new_ctx_len: next, session: info };
+    }
+    async recreateSession(modelId) {
+        this.table.assertRunning();
+        const session = this.table.get(modelId);
+        if (!session)
+            return { ok: false, reason: 'not-loaded' };
+        const info = await this.reload(modelId, { ...session.extra.overrides, ctx_size: session.extra.ctxSize }, 'compute_error_recovery');
+        return { ok: true, session: info };
+    }
+    /** Unload (a failure does not stop the reload) and load again without evicting other models. */
+    async reload(modelId, overrides, why) {
+        const unloaded = await this.table.unload(modelId);
+        if (!unloaded.success)
+            this.emit('core:log', {
+                level: 'warn',
+                msg: `${why}: unload of ${modelId} failed, reloading anyway: ${unloaded.error}`,
+            });
+        return this.load(modelId, { overrides, bypassAutoUnload: true });
+    }
+    shutdown() {
+        return this.table.shutdown();
+    }
+    exists(path) {
+        return (this.options.exists ?? existsSync)(path);
+    }
+}
+//# sourceMappingURL=runtime.js.map

package/dist/runtime/mlx/runtime.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"runtime.js","sourceRoot":"","sources":["../../../src/runtime/mlx/runtime.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;GAmBG;AAEH,OAAO,EAAE,UAAU,EAAE,MAAM,SAAS,CAAA;AACpC,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAA;AAKhC,OAAO,EAAE,kBAAkB,EAAE,qBAAqB,EAAE,eAAe,EAAE,MAAM,4BAA4B,CAAA;AACvG,OAAO,EAAE,eAAe,EAAE,iBAAiB,EAAE,MAAM,2BAA2B,CAAA;AAQ9E,OAAO,EACL,cAAc,EACd,aAAa,EACb,cAAc,EACd,YAAY,EACZ,kBAAkB,EAClB,oBAAoB,GACrB,MAAM,oBAAoB,CAAA;AAC3B,OAAO,EAAE,kBAAkB,EAAE,qBAAqB,EAAE,MAAM,WAAW,CAAA;AACrE,OAAO,EAAE,QAAQ,EAAE,cAAc,EAAE,sBAAsB,EAAE,MAAM,aAAa,CAAA;AAE9E,OAAO,EACL,wBAAwB,EACxB,wBAAwB,EACxB,iBAAiB,EACjB,eAAe,EACf,gBAAgB,EAChB,eAAe,EACf,UAAU,GACX,MAAM,aAAa,CAAA;AACpB,OAAO,EAAE,kBAAkB,EAAE,qBAAqB,EAAE,oBAAoB,EAAE,MAAM,kBAAkB,CAAA;AAElG,MAAM,CAAC,MAAM,iBAAiB,GAAG,YAAY,CAAA;AAC7C,qDAAqD;AACrD,MAAM,CAAC,MAAM,wBAAwB,GAAG,GAAG,CAAA;AA+B3C,MAAM,OAAO,UAAU;IAIQ;IAHZ,KAAK,CAA+B;IACpC,IAAI,CAAQ;IAE7B,YAA6B,OAA0B;QAA1B,YAAO,GAAP,OAAO,CAAmB;QACrD,IAAI,CAAC,IAAI,GAAG,OAAO,CAAC,IAAI,IAAI,CAAC,GAAG,EAAE,GAAE,CAAC,CAAC,CAAA;QACtC,IAAI,CAAC,KAAK,GAAG,IAAI,YAAY,CAAkB;YAC7C,QAAQ,EAAE,KAAK;YACf,UAAU,EAAE,OAAO,CAAC,UAAU;YAC9B,OAAO,EAAE,OAAO,CAAC,OAAO;YACxB,IAAI,EAAE,IAAI,CAAC,IAAI;YACf,MAAM,EAAE,KAAK;YACb,aAAa,EAAE,IAAI;YACnB,eAAe,EAAE,IAAI;YACrB,YAAY,EAAE,CAAC,IAAI,EAAE,MAAM,EAAE,EAAE,CAAC,eAAe,CAAC,IAAI,EAAE,MAAM,CAAC;SAC9D,CAAC,CAAA;IACJ,CAAC;IAED,UAAU;QACR,OAAO,IAAI,CAAC,OAAO,CAAC,YAAY,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,YAAY,EAAE,iBAAiB,CAAC,CAAC,CAAC,CAAC,SAAS,CAAA;IACnG,CAAC;IAED,IAAI;QACF,OAAO,IAAI,CAAC,KAAK,CAAC,IAAI,EAAE,CAAA;IAC1B,CAAC;IAED,WAAW,CAAC,OAAe;QACzB,OAAO,IAAI,CAAC,KAAK,CAAC,WAAW,CAAC,OAAO,CAAC,CAAA;IACxC,CAAC;IAED,eAAe;QACb,OAAO,IAAI,CAAC,KAAK,CAAC,eAAe,EAAE,CAAA;IACrC,CAAC;IAED,SAAS,CAAC,OAAe;QACvB,OAAO,IAAI,CAAC,KAAK,CAAC,SAAS,CAAC,OAAO,CAAC,CAAA;IACtC,CAAC;IAED,8CAA8C;IAC9C,UAAU,CAAC,OAAe;QACxB,OAAO,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,OAAO,CAAC,EAAE,KAAK,CAAC,OAAO,CAAA;IAC/C,CAAC;IAED,KAAK,CAAC,IAAI,CAAC,OAAe,EAAE,OAAyB,EAAE;QACrD,IAAI,CAAC,KAAK,CAAC,aAAa,EAAE,CAAA;QAC1B,OAAO,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,OAAO,EAAE,GAAG,EAAE,CAAC,IAAI,CAAC,KAAK,CAAC,OAAO,EAAE,IAAI,CAAC,EAAE,IAAI,CAAC,MAAM,CAAC,CAAA;IAC/E,CAAC;IAEO,KAAK,CAAC,KAAK,CAAC,OAAe,EAAE,IAAsB;QACzD,oBAAoB,CAAC,IAAI,CAAC,MAAM,CAAC,CAAA;QACjC,MAAM,QAAQ,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC,YAAY,EAAE,CAAA;QAClD,MAAM,SAAS,GAAG,EAAE,GAAG,CAAC,IAAI,CAAC,SAAS,IAAI,EAAE,CAAC,EAAE,CAAA;QAC/C,MAAM,GAAG,GAAsD,EAAE,GAAG,QAAQ,EAAE,GAAG,SAAS,EAAE,CAAA;QAC5F,MAAM,WAAW,GAAG,IAAI,CAAC,WAAW,IAAI,KAAK,CAAA;QAE7C,MAAM,UAAU,GACd,GAAG,CAAC,aAAa,CAAC,KAAK,SAAS,IAAI,GAAG,CAAC,aAAa,CAAC,KAAK,IAAI,IAAI,GAAG,CAAC,aAAa,CAAC,KAAK,MAAM,CAAA;QAClG,IAAI,UAAU,IAAI,CAAC,WAAW,IAAI,CAAC,CAAC,IAAI,CAAC,gBAAgB,IAAI,KAAK,CAAC,EAAE,CAAC;YACpE,uFAAuF;YACvF,KAAK,MAAM,MAAM,IAAI,IAAI,CAAC,KAAK,CAAC,eAAe,EAAE,EAAE,CAAC;gBAClD,oBAAoB,CAAC,IAAI,CAAC,MAAM,CAAC,CAAA;gBACjC,MAAM,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,CAAC,CAAA;YACjC,CAAC;QACH,CAAC;QACD,IAAI,CAAC,KAAK,CAAC,aAAa,EAAE,CAAA;QAC1B,oBAAoB,CAAC,IAAI,CAAC,MAAM,CAAC,CAAA;QAEjC,MAAM,QAAQ,GAAG,IAAI,CAAC,OAAO,CAAC,QAAQ,CAAA;QACtC,MAAM,GAAG,GAAG,MAAM,qBAAqB,CAAC,QAAQ,EAAE,OAAO,EAAE,MAAM,QAAQ,CAAC,IAAI,CAAC,OAAO,CAAC,EAAE,CAAC,OAAO,EAAE,EAAE,CACnG,IAAI,CAAC,IAAI,CAAC,UAAU,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,GAAG,EAAE,OAAO,EAAE,CAAC,CACvD,CAAA;QACD,MAAM,EAAE,SAAS,EAAE,GAAG,QAAQ,CAAC,YAAY,CAAC,GAAG,CAAC,CAAA;QAChD,MAAM,GAAG,GAAG,IAAI,CAAC,OAAO,IAAI,IAAI,CAAC,UAAU,EAAE,CAAA;QAC7C,IAAI,CAAC,GAAG,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC;YAAE,MAAM,gBAAgB,CAAC,GAAG,IAAI,IAAI,CAAC,iBAAiB,EAAE,iBAAiB,CAAC,CAAC,CAAA;QACxG,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC;YAAE,MAAM,eAAe,CAAC,SAAS,CAAC,CAAA;QAE7D,MAAM,IAAI,GAAG,IAAI,CAAC,IAAI,IAAI,CAAC,MAAM,cAAc,CAAC,IAAI,CAAC,KAAK,CAAC,SAAS,EAAE,CAAC,CAAC,CAAA;QACxE,MAAM,WAAW,GAAG,MAAM,kBAAkB,CAAC,SAAS,CAAC,CAAA;QACvD,oBAAoB,CAAC,IAAI,CAAC,MAAM,CAAC,CAAA;QACjC,MAAM,KAAK,GAAG,sBAAsB,CAAC,GAAG,CAAC,CAAA;QACzC,MAAM,UAAU,GAAG,OAAO,CAAC,GAAG,CAAC,cAAc,CAAC,IAAI,OAAO,CAAC,GAAG,CAAC,WAAW,CAAC,IAAI,OAAO,CAAC,GAAG,CAAC,cAAc,CAAC,CAAA;QACzG,IAAI,UAAU,IAAI,CAAC,KAAK,CAAC,SAAS,EAAE,CAAC;YACnC,MAAM,QAAQ,GAAG,MAAM,CAAC,IAAI,CAAC,OAAO,CAAC,YAAY,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,EAAE,EAAE,CAAC,IAAI,CAAC,UAAU,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC,CAAC,CAC7F,KAAK,CAAC,SAAS,EACf,OAAO,CACR,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,SAAS,CAAC,CAAA;YACxB,IAAI,QAAQ,EAAE,CAAC;gBACb,KAAK,CAAC,SAAS,GAAG,QAAQ,CAAA;gBAC1B,SAAS,CAAC,kBAAkB,CAAC,GAAG,QAAQ,CAAA;YAC1C,CAAC;iBAAM,CAAC;gBACN,IAAI,CAAC,IAAI,CAAC,UAAU,EAAE;oBACpB,KAAK,EAAE,MAAM;oBACb,GAAG,EAAE,GAAG,OAAO,QAAQ,KAAK,CAAC,SAAS,wBAAwB,KAAK,CAAC,SAAS,2CAA2C;iBACzH,CAAC,CAAA;YACJ,CAAC;QACH,CAAC;QACD,MAAM,MAAM,GAAG,cAAc,CAAC,GAAG,EAAE,KAAK,EAAE,EAAE,GAAG,CAAC,WAAW,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,WAAW,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC,CAAA;QACpG,MAAM,IAAI,GAAG,kBAAkB,CAAC,SAAS,EAAE,IAAI,EAAE,MAAM,CAAC,CAAA;QACxD,MAAM,WAAW,GAAG,IAAI,CAAC,WAAW,IAAI,QAAQ,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC,IAAI,wBAAwB,CAAA;QAC5F,MAAM,GAAG,GAA2B,EAAE,CAAA;QACtC,KAAK,MAAM,CAAC,GAAG,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC,OAAO,CAAC,OAAO,IAAI,OAAO,CAAC,GAAG,CAAC;YAC5E,IAAI,KAAK,KAAK,SAAS;gBAAE,GAAG,CAAC,GAAG,CAAC,GAAG,KAAK,CAAA;QAC3C,GAAG,CAAC,sBAAsB,CAAC,GAAG,GAAG,CAAA;QAEjC,oBAAoB,CAAC,IAAI,CAAC,MAAM,CAAC,CAAA;QACjC,MAAM,SAAS,GAAG,IAAI,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,aAAa,CAAC,IAAI,CAAC,OAAO,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,SAAS,CAAA;QACrF,IAAI,OAAO,CAAA;QACX,IAAI,CAAC;YACH,OAAO,GAAG,MAAM,CAAC,IAAI,CAAC,OAAO,CAAC,KAAK,IAAI,kBAAkB,CAAC,CACxD,EAAE,GAAG,EAAE,IAAI,EAAE,GAAG,EAAE,EAClB;gBACE,SAAS,EAAE,WAAW,GAAG,IAAI;gBAC7B,MAAM,EAAE,IAAI,CAAC,KAAK,CAAC,MAAM;gBACzB,GAAG,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,YAAY,EAAE,IAAI,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;gBACrD,kBAAkB,EAAE,EAAE,MAAM,EAAE,wBAAwB,EAAE,MAAM,EAAE,wBAAwB,EAAE;gBAC1F,mFAAmF;gBACnF,cAAc,EAAE,CAAC;gBACjB,YAAY,EAAE,CAAC,MAAM,EAAE,EAAE,CAAC,UAAU,CAAC,WAAW,EAAE,MAAM,CAAC;gBACzD,YAAY,EAAE,CAAC,KAAK,EAAE,MAAM,EAAE,EAAE,CAAC,iBAAiB,CAAC,MAAM,CAAC;gBAC1D,MAAM,EAAE,CAAC,MAAM,EAAE,IAAI,EAAE,EAAE;oBACvB,SAAS,EAAE,KAAK,CAAC,IAAI,MAAM,KAAK,IAAI,IAAI,CAAC,CAAA;oBACzC,IAAI,IAAI,CAAC,OAAO;wBACd,IAAI,CAAC,IAAI,CAAC,UAAU,EAAE,EAAE,KAAK,EAAE,OAAO,EAAE,GAAG,EAAE,QAAQ,OAAO,KAAK,MAAM,KAAK,IAAI,EAAE,EAAE,CAAC,CAAA;gBACzF,CAAC;aACF,CACF,CAAA;QACH,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,MAAM,cAAc,CAAC,SAAS,CAAC,CAAA;YAC/B,MAAM,KAAK,CAAA;QACb,CAAC;QACD,OAAO,IAAI,CAAC,KAAK,CAAC,KAAK,CACrB;YACE,IAAI,EAAE;gBACJ,GAAG,EAAE,OAAO,CAAC,OAAO,CAAC,GAAG;gBACxB,IAAI;gBACJ,QAAQ,EAAE,OAAO;gBACjB,UAAU,EAAE,qBAAqB,CAAC,SAAS,CAAC;gBAC5C,YAAY,EAAE,WAAW;gBACzB,OAAO,EAAE,EAAE;aACZ;YACD,OAAO,EAAE,OAAO,CAAC,OAAO;YACxB,GAAG;YACH,KAAK,EAAE,EAAE,OAAO,EAAE,MAAM,CAAC,QAAQ,EAAE,WAAW,EAAE,SAAS,EAAE;YAC3D,SAAS;SACV,EACD,IAAI,CAAC,MAAM,CACZ,CAAA;IACH,CAAC;IAEO,KAAK,CAAC,UAAU,CAAC,IAAkB,EAAE,OAAe;QAC1D,MAAM,UAAU,GACd,IAAI,KAAK,KAAK;YACZ,CAAC,CAAC,eAAe,CAAC,OAAO,CAAC;YAC1B,CAAC,CAAC,IAAI,KAAK,QAAQ;gBACjB,CAAC,CAAC,kBAAkB,CAAC,OAAO,CAAC;gBAC7B,CAAC,CAAC,qBAAqB,CAAC,OAAO,CAAC,CAAA;QACtC,OAAO,UAAU,CAAC,CAAC,CAAC,oBAAoB,CAAC,IAAI,CAAC,OAAO,CAAC,MAAM,EAAE,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,SAAS,CAAA;IAC5F,CAAC;IAED,MAAM,CAAC,OAAe;QACpB,IAAI,CAAC,KAAK,CAAC,aAAa,EAAE,CAAA;QAC1B,OAAO,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,OAAO,CAAC,CAAA;IACnC,CAAC;IAED;;;OAGG;IACH,KAAK,CAAC,eAAe,CAAC,OAAe,EAAE,MAAM,GAAG,cAAc;QAC5D,IAAI,CAAC,KAAK,CAAC,aAAa,EAAE,CAAA;QAC1B,MAAM,OAAO,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,OAAO,CAAC,CAAA;QACvC,IAAI,CAAC,OAAO;YAAE,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,YAAY,EAAE,CAAA;QACxD,MAAM,OAAO,GAAG,OAAO,CAAC,KAAK,CAAC,OAAO,IAAI,eAAe,CAAA;QACxD,MAAM,GAAG,GAAG,OAAO,CAAC,KAAK,CAAC,WAAW,CAAA;QACrC,MAAM,IAAI,GAAG,iBAAiB,CAAC,OAAO,EAAE,GAAG,CAAC,CAAA;QAC5C,IAAI,IAAI,IAAI,OAAO;YACjB,OAAO;gBACL,EAAE,EAAE,KAAK;gBACT,MAAM,EAAE,QAAQ;gBAChB,eAAe,EAAE,OAAO;gBACxB,GAAG,CAAC,GAAG,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,WAAW,EAAE,GAAG,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;aACnD,CAAA;QACH,MAAM,IAAI,GAAG,MAAM,IAAI,CAAC,MAAM,CAC5B,OAAO,EACP,EAAE,GAAG,OAAO,CAAC,KAAK,CAAC,SAAS,EAAE,QAAQ,EAAE,IAAI,EAAE,EAC9C,mBAAmB,CACpB,CAAA;QACD,IAAI,CAAC,IAAI,CAAC,uBAAuB,EAAE,EAAE,QAAQ,EAAE,KAAK,EAAE,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,CAAC,CAAA;QACvG,OAAO,EAAE,EAAE,EAAE,IAAI,EAAE,WAAW,EAAE,IAAI,EAAE,OAAO,EAAE,IAAI,EAAE,CAAA;IACvD,CAAC;IAED,KAAK,CAAC,eAAe,CAAC,OAAe;QACnC,IAAI,CAAC,KAAK,CAAC,aAAa,EAAE,CAAA;QAC1B,MAAM,OAAO,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,OAAO,CAAC,CAAA;QACvC,IAAI,CAAC,OAAO;YAAE,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,YAAY,EAAE,CAAA;QACxD,MAAM,IAAI,GAAG,MAAM,IAAI,CAAC,MAAM,CAC5B,OAAO,EACP,EAAE,GAAG,OAAO,CAAC,KAAK,CAAC,SAAS,EAAE,QAAQ,EAAE,OAAO,CAAC,KAAK,CAAC,OAAO,EAAE,EAC/D,wBAAwB,CACzB,CAAA;QACD,OAAO,EAAE,EAAE,EAAE,IAAI,EAAE,OAAO,EAAE,IAAI,EAAE,CAAA;IACpC,CAAC;IAED,gGAAgG;IACxF,KAAK,CAAC,MAAM,CAClB,OAAe,EACf,SAAkC,EAClC,GAAW;QAEX,MAAM,QAAQ,GAAG,MAAM,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,OAAO,CAAC,CAAA;QACjD,IAAI,CAAC,QAAQ,CAAC,OAAO;YACnB,IAAI,CAAC,IAAI,CAAC,UAAU,EAAE;gBACpB,KAAK,EAAE,MAAM;gBACb,GAAG,EAAE,GAAG,GAAG,eAAe,OAAO,8BAA8B,QAAQ,CAAC,KAAK,EAAE;aAChF,CAAC,CAAA;QACJ,OAAO,IAAI,CAAC,IAAI,CAAC,OAAO,EAAE,EAAE,SAAS,EAAE,gBAAgB,EAAE,IAAI,EAAE,CAAC,CAAA;IAClE,CAAC;IAED,QAAQ;QACN,OAAO,IAAI,CAAC,KAAK,CAAC,QAAQ,EAAE,CAAA;IAC9B,CAAC;IAEO,MAAM,CAAC,IAAY;QACzB,OAAO,CAAC,IAAI,CAAC,OAAO,CAAC,MAAM,IAAI,UAAU,CAAC,CAAC,IAAI,CAAC,CAAA;IAClD,CAAC;CACF"}

package/dist/runtime/mlx/shard-repair.d.ts

@@ -0,0 +1,25 @@
+/**
+ * Healing MLX checkpoints downloaded before the shard-naming fix.
+ *
+ * Ported from: extensions/mlx-extension/src/shardRepair.ts (verbatim logic).
+ */
+export type ShardRepairPlan = {
+    from: string;
+    to: string;
+};
+/**
+ * Decide whether a model directory holds a mis-named first shard.
+ *
+ * Downloads made before the shard-naming fix stored the repo's first
+ * `*.safetensors` sibling as a fixed `model.safetensors`, so a sharded
+ * checkpoint ends up with the shard that `model.safetensors.index.json`
+ * calls `model-00001-of-000NN.safetensors` sitting under the legacy name.
+ * Loading then fails with every shard-1 parameter reported missing.
+ *
+ * Only the unambiguous case is repaired: exactly one indexed shard is absent
+ * and the legacy file — which the index itself never references — is present.
+ * Two or more absent shards mean an interrupted download, which no rename can
+ * mend, so nothing is touched.
+ */
+export declare function planMlxShardRepair(weightMap: unknown, presentFiles: readonly string[]): ShardRepairPlan | undefined;
+export declare function repointLegacyWeightPath(path: string, shardFileName: string): string;

package/dist/runtime/mlx/shard-repair.js

@@ -0,0 +1,44 @@
+/**
+ * Healing MLX checkpoints downloaded before the shard-naming fix.
+ *
+ * Ported from: extensions/mlx-extension/src/shardRepair.ts (verbatim logic).
+ */
+const LEGACY_WEIGHT_FILE = 'model.safetensors';
+/**
+ * Decide whether a model directory holds a mis-named first shard.
+ *
+ * Downloads made before the shard-naming fix stored the repo's first
+ * `*.safetensors` sibling as a fixed `model.safetensors`, so a sharded
+ * checkpoint ends up with the shard that `model.safetensors.index.json`
+ * calls `model-00001-of-000NN.safetensors` sitting under the legacy name.
+ * Loading then fails with every shard-1 parameter reported missing.
+ *
+ * Only the unambiguous case is repaired: exactly one indexed shard is absent
+ * and the legacy file — which the index itself never references — is present.
+ * Two or more absent shards mean an interrupted download, which no rename can
+ * mend, so nothing is touched.
+ */
+export function planMlxShardRepair(weightMap, presentFiles) {
+    if (weightMap == null || typeof weightMap !== 'object')
+        return undefined;
+    const shards = new Set(Object.values(weightMap).filter((shard) => typeof shard === 'string' && shard.length > 0));
+    if (shards.size === 0 || shards.has(LEGACY_WEIGHT_FILE))
+        return undefined;
+    const present = new Set(presentFiles);
+    if (!present.has(LEGACY_WEIGHT_FILE))
+        return undefined;
+    const absent = [...shards].filter((shard) => !present.has(shard));
+    if (absent.length !== 1)
+        return undefined;
+    return { from: LEGACY_WEIGHT_FILE, to: absent[0] };
+}
+/// Repoint a `model.yml` path that still names the legacy weight file at the
+/// shard it was renamed to. Paths pointing anywhere else are left alone.
+export function repointLegacyWeightPath(path, shardFileName) {
+    const separator = path.slice(-LEGACY_WEIGHT_FILE.length - 1, -LEGACY_WEIGHT_FILE.length);
+    if (!path.endsWith(LEGACY_WEIGHT_FILE) || (separator !== '/' && separator !== '\\')) {
+        return path;
+    }
+    return `${path.slice(0, -LEGACY_WEIGHT_FILE.length)}${shardFileName}`;
+}
+//# sourceMappingURL=shard-repair.js.map

package/dist/runtime/mlx/shard-repair.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"shard-repair.js","sourceRoot":"","sources":["../../../src/runtime/mlx/shard-repair.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,MAAM,kBAAkB,GAAG,mBAAmB,CAAA;AAQ9C;;;;;;;;;;;;;GAaG;AACH,MAAM,UAAU,kBAAkB,CAChC,SAAkB,EAClB,YAA+B;IAE/B,IAAI,SAAS,IAAI,IAAI,IAAI,OAAO,SAAS,KAAK,QAAQ;QAAE,OAAO,SAAS,CAAA;IAExE,MAAM,MAAM,GAAG,IAAI,GAAG,CACpB,MAAM,CAAC,MAAM,CAAC,SAAoC,CAAC,CAAC,MAAM,CACxD,CAAC,KAAK,EAAmB,EAAE,CAAC,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,CAAC,MAAM,GAAG,CAAC,CAC1E,CACF,CAAA;IACD,IAAI,MAAM,CAAC,IAAI,KAAK,CAAC,IAAI,MAAM,CAAC,GAAG,CAAC,kBAAkB,CAAC;QAAE,OAAO,SAAS,CAAA;IAEzE,MAAM,OAAO,GAAG,IAAI,GAAG,CAAC,YAAY,CAAC,CAAA;IACrC,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,kBAAkB,CAAC;QAAE,OAAO,SAAS,CAAA;IAEtD,MAAM,MAAM,GAAG,CAAC,GAAG,MAAM,CAAC,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC,CAAA;IACjE,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,SAAS,CAAA;IAEzC,OAAO,EAAE,IAAI,EAAE,kBAAkB,EAAE,EAAE,EAAE,MAAM,CAAC,CAAC,CAAW,EAAE,CAAA;AAC9D,CAAC;AAED,6EAA6E;AAC7E,yEAAyE;AACzE,MAAM,UAAU,uBAAuB,CAAC,IAAY,EAAE,aAAqB;IACzE,MAAM,SAAS,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,kBAAkB,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC,kBAAkB,CAAC,MAAM,CAAC,CAAA;IACxF,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAC,kBAAkB,CAAC,IAAI,CAAC,SAAS,KAAK,GAAG,IAAI,SAAS,KAAK,IAAI,CAAC,EAAE,CAAC;QACpF,OAAO,IAAI,CAAA;IACb,CAAC;IACD,OAAO,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,kBAAkB,CAAC,MAAM,CAAC,GAAG,aAAa,EAAE,CAAA;AACvE,CAAC"}

package/dist/runtime/shared/env.d.ts

@@ -0,0 +1,49 @@
+/**
+ * Environment of a spawned backend process. Port of `setup_library_path`, `add_cuda_paths` and
+ * `binary_requires_cuda` in `src-tauri/utils/src/system.rs`, as pure functions over injected facts.
+ *
+ * Deliberate deviation (PLAN.md §2 decision 15): on Windows the Rust code rebuilt `PATH` twice from
+ * the *process* environment, so the CUDA directories were overwritten by the binary directory. Here
+ * both are prepended: `<exe dir>;<cuda bins...>;<PATH>`.
+ */
+export interface CudaPaths {
+    /** Directories with CUDA libraries (Linux → LD_LIBRARY_PATH). */
+    libDirs: string[];
+    /** Directories with CUDA binaries/DLLs (→ PATH). */
+    binDirs: string[];
+}
+/** The real filesystem probe used outside tests; `discoverCudaPaths` takes the seams instead of fs. */
+export declare function nodeCudaProbeEnv(platform?: NodeJS.Platform, env?: NodeJS.ProcessEnv): CudaProbeEnv;
+export interface CudaProbeEnv {
+    platform: NodeJS.Platform;
+    env: NodeJS.ProcessEnv;
+    exists: (path: string) => boolean;
+    /** Directory names inside `dir`; [] when unreadable. */
+    listDir: (dir: string) => string[];
+}
+/** Where a CUDA runtime might live on this host. Sorted, deduplicated, existing paths only. */
+export declare function discoverCudaPaths(e: CudaProbeEnv): CudaPaths;
+/** CUDA library names that make a binary depend on a CUDA runtime, per platform probe. */
+export declare function textMentionsCudaRuntime(text: string, platform: NodeJS.Platform, viaLdd?: boolean): boolean;
+export interface ProcessEnvInput {
+    platform: NodeJS.Platform;
+    baseEnv: NodeJS.ProcessEnv;
+    /** Directory of the backend executable (added to the library path; cwd on Windows). */
+    exeDir: string;
+    cuda: CudaPaths;
+    /** Caller-supplied variables (`LLAMA_API_KEY`, user `llamacpp_env`); applied first, overridable by the path setup. */
+    userEnv: Record<string, string>;
+}
+export interface ProcessEnvResult {
+    env: Record<string, string>;
+    cwd: string | undefined;
+    /** True when CUDA directories were found and injected. */
+    cudaFound: boolean;
+}
+/** Strip the Windows verbatim prefix the way the Rust code does before putting a dir on PATH. */
+export declare function stripVerbatimPrefix(dir: string): string;
+export declare function buildProcessEnv(input: ProcessEnvInput): ProcessEnvResult;
+/** `<exe dir>` from a full executable path, platform-agnostic. */
+export declare function exeDirOf(exePath: string): string;
+/** Join for the platform of the path rather than the host (used in tests and cross-platform planning). */
+export declare function joinFor(platform: NodeJS.Platform, ...parts: string[]): string;

package/dist/runtime/shared/env.js

@@ -0,0 +1,155 @@
+/**
+ * Environment of a spawned backend process. Port of `setup_library_path`, `add_cuda_paths` and
+ * `binary_requires_cuda` in `src-tauri/utils/src/system.rs`, as pure functions over injected facts.
+ *
+ * Deliberate deviation (PLAN.md §2 decision 15): on Windows the Rust code rebuilt `PATH` twice from
+ * the *process* environment, so the CUDA directories were overwritten by the binary directory. Here
+ * both are prepended: `<exe dir>;<cuda bins...>;<PATH>`.
+ */
+import { existsSync, readdirSync } from 'node:fs';
+import { join } from 'node:path';
+/** The real filesystem probe used outside tests; `discoverCudaPaths` takes the seams instead of fs. */
+export function nodeCudaProbeEnv(platform = process.platform, env = process.env) {
+    return {
+        platform,
+        env,
+        exists: (path) => existsSync(path),
+        listDir: (dir) => {
+            try {
+                return readdirSync(dir, { withFileTypes: true })
+                    .filter((e) => e.isDirectory())
+                    .map((e) => e.name);
+            }
+            catch {
+                return [];
+            }
+        },
+    };
+}
+const LINUX_COMMON_LIB_DIRS = [
+    '/usr/local/cuda/lib64',
+    '/usr/local/cuda/lib',
+    '/usr/lib/cuda/lib64',
+    '/usr/lib/cuda/lib',
+    '/opt/cuda/lib64',
+    '/opt/cuda/lib',
+    '/usr/lib/x86_64-linux-gnu',
+    '/usr/lib/x86_64-linux-gnu/nvidia',
+];
+/** Where a CUDA runtime might live on this host. Sorted, deduplicated, existing paths only. */
+export function discoverCudaPaths(e) {
+    const libs = new Set();
+    const bins = new Set();
+    if (e.platform === 'win32') {
+        const cudaPath = e.env['CUDA_PATH'];
+        if (cudaPath) {
+            const bin = `${cudaPath}\\bin`;
+            if (e.exists(bin))
+                bins.add(bin);
+        }
+        for (const [key, value] of Object.entries(e.env)) {
+            if (key.startsWith('CUDA_PATH_V') && value) {
+                const bin = `${value}\\bin`;
+                if (e.exists(bin))
+                    bins.add(bin);
+            }
+        }
+        const programFiles = e.env['ProgramFiles'] ?? 'C:\\Program Files';
+        const toolkit = `${programFiles}\\NVIDIA GPU Computing Toolkit\\CUDA`;
+        for (const name of e.listDir(toolkit)) {
+            const bin = `${toolkit}\\${name}\\bin`;
+            if (e.exists(bin))
+                bins.add(bin);
+        }
+    }
+    else if (e.platform === 'linux') {
+        const home = e.env['CUDA_HOME'] ?? e.env['CUDA_PATH'];
+        if (home) {
+            for (const sub of ['lib64', 'lib'])
+                if (e.exists(`${home}/${sub}`))
+                    libs.add(`${home}/${sub}`);
+            if (e.exists(`${home}/bin`))
+                bins.add(`${home}/bin`);
+        }
+        for (const p of LINUX_COMMON_LIB_DIRS)
+            if (e.exists(p))
+                libs.add(p);
+        for (const name of e.listDir('/usr/local')) {
+            if (!name.startsWith('cuda-'))
+                continue;
+            for (const sub of ['lib64', 'lib', 'bin']) {
+                const p = `/usr/local/${name}/${sub}`;
+                if (!e.exists(p))
+                    continue;
+                if (sub === 'bin')
+                    bins.add(p);
+                else
+                    libs.add(p);
+            }
+        }
+    }
+    return { libDirs: [...libs].sort(), binDirs: [...bins].sort() };
+}
+/** CUDA library names that make a binary depend on a CUDA runtime, per platform probe. */
+export function textMentionsCudaRuntime(text, platform, viaLdd = false) {
+    if (platform === 'win32') {
+        return ['cudart', 'cublas', 'cufft', 'curand', 'cusparse', 'cusolver', 'cudnn'].some((n) => text.includes(n));
+    }
+    if (platform === 'linux') {
+        const names = viaLdd
+            ? ['libcudart', 'libcublas', 'libcufft', 'libcurand', 'libcusparse', 'libcusolver', 'libcudnn']
+            : ['libcudart', 'libcublas', 'libcufft'];
+        return names.some((n) => text.includes(n));
+    }
+    return false;
+}
+/** Strip the Windows verbatim prefix the way the Rust code does before putting a dir on PATH. */
+export function stripVerbatimPrefix(dir) {
+    return dir.startsWith('\\\\?\\') ? dir.slice(4) : dir;
+}
+const prepend = (items, existing, sep) => {
+    const rest = (existing ?? '').replace(new RegExp(`\\${sep}+$`), '');
+    return rest === '' ? items.join(sep) : `${items.join(sep)}${sep}${rest}`;
+};
+export function buildProcessEnv(input) {
+    const env = {};
+    for (const [k, v] of Object.entries(input.baseEnv))
+        if (v !== undefined)
+            env[k] = v;
+    Object.assign(env, input.userEnv);
+    const cudaFound = input.cuda.libDirs.length > 0 || input.cuda.binDirs.length > 0;
+    let cwd;
+    switch (input.platform) {
+        case 'linux': {
+            const libs = [input.exeDir, ...input.cuda.libDirs];
+            env['LD_LIBRARY_PATH'] = prepend(libs, env['LD_LIBRARY_PATH'], ':');
+            if (input.cuda.binDirs.length)
+                env['PATH'] = prepend(input.cuda.binDirs, env['PATH'], ':');
+            break;
+        }
+        case 'win32': {
+            const dir = stripVerbatimPrefix(input.exeDir);
+            env['PATH'] = prepend([dir, ...input.cuda.binDirs], env['PATH'] ?? env['Path'], ';');
+            delete env['Path'];
+            cwd = input.exeDir;
+            break;
+        }
+        case 'darwin': {
+            env['DYLD_LIBRARY_PATH'] = prepend([input.exeDir], env['DYLD_LIBRARY_PATH'], ':');
+            break;
+        }
+        default:
+            break;
+    }
+    return { env, cwd, cudaFound };
+}
+/** `<exe dir>` from a full executable path, platform-agnostic. */
+export function exeDirOf(exePath) {
+    const i = Math.max(exePath.lastIndexOf('/'), exePath.lastIndexOf('\\'));
+    return i < 0 ? '.' : exePath.slice(0, i);
+}
+/** Join for the platform of the path rather than the host (used in tests and cross-platform planning). */
+export function joinFor(platform, ...parts) {
+    return platform === 'win32' ? parts.join('\\') : join(...parts);
+}
+//# sourceMappingURL=env.js.map

package/dist/runtime/shared/env.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"env.js","sourceRoot":"","sources":["../../../src/runtime/shared/env.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAEH,OAAO,EAAE,UAAU,EAAE,WAAW,EAAE,MAAM,SAAS,CAAA;AACjD,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAA;AAShC,uGAAuG;AACvG,MAAM,UAAU,gBAAgB,CAC9B,WAA4B,OAAO,CAAC,QAAQ,EAC5C,MAAyB,OAAO,CAAC,GAAG;IAEpC,OAAO;QACL,QAAQ;QACR,GAAG;QACH,MAAM,EAAE,CAAC,IAAI,EAAE,EAAE,CAAC,UAAU,CAAC,IAAI,CAAC;QAClC,OAAO,EAAE,CAAC,GAAG,EAAE,EAAE;YACf,IAAI,CAAC;gBACH,OAAO,WAAW,CAAC,GAAG,EAAE,EAAE,aAAa,EAAE,IAAI,EAAE,CAAC;qBAC7C,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,WAAW,EAAE,CAAC;qBAC9B,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,CAAA;YACvB,CAAC;YAAC,MAAM,CAAC;gBACP,OAAO,EAAE,CAAA;YACX,CAAC;QACH,CAAC;KACF,CAAA;AACH,CAAC;AAUD,MAAM,qBAAqB,GAAG;IAC5B,uBAAuB;IACvB,qBAAqB;IACrB,qBAAqB;IACrB,mBAAmB;IACnB,iBAAiB;IACjB,eAAe;IACf,2BAA2B;IAC3B,kCAAkC;CACnC,CAAA;AAED,+FAA+F;AAC/F,MAAM,UAAU,iBAAiB,CAAC,CAAe;IAC/C,MAAM,IAAI,GAAG,IAAI,GAAG,EAAU,CAAA;IAC9B,MAAM,IAAI,GAAG,IAAI,GAAG,EAAU,CAAA;IAC9B,IAAI,CAAC,CAAC,QAAQ,KAAK,OAAO,EAAE,CAAC;QAC3B,MAAM,QAAQ,GAAG,CAAC,CAAC,GAAG,CAAC,WAAW,CAAC,CAAA;QACnC,IAAI,QAAQ,EAAE,CAAC;YACb,MAAM,GAAG,GAAG,GAAG,QAAQ,OAAO,CAAA;YAC9B,IAAI,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC;gBAAE,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,CAAA;QAClC,CAAC;QACD,KAAK,MAAM,CAAC,GAAG,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC;YACjD,IAAI,GAAG,CAAC,UAAU,CAAC,aAAa,CAAC,IAAI,KAAK,EAAE,CAAC;gBAC3C,MAAM,GAAG,GAAG,GAAG,KAAK,OAAO,CAAA;gBAC3B,IAAI,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC;oBAAE,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,CAAA;YAClC,CAAC;QACH,CAAC;QACD,MAAM,YAAY,GAAG,CAAC,CAAC,GAAG,CAAC,cAAc,CAAC,IAAI,mBAAmB,CAAA;QACjE,MAAM,OAAO,GAAG,GAAG,YAAY,sCAAsC,CAAA;QACrE,KAAK,MAAM,IAAI,IAAI,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC,EAAE,CAAC;YACtC,MAAM,GAAG,GAAG,GAAG,OAAO,KAAK,IAAI,OAAO,CAAA;YACtC,IAAI,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC;gBAAE,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,CAAA;QAClC,CAAC;IACH,CAAC;SAAM,IAAI,CAAC,CAAC,QAAQ,KAAK,OAAO,EAAE,CAAC;QAClC,MAAM,IAAI,GAAG,CAAC,CAAC,GAAG,CAAC,WAAW,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,WAAW,CAAC,CAAA;QACrD,IAAI,IAAI,EAAE,CAAC;YACT,KAAK,MAAM,GAAG,IAAI,CAAC,OAAO,EAAE,KAAK,CAAC;gBAAE,IAAI,CAAC,CAAC,MAAM,CAAC,GAAG,IAAI,IAAI,GAAG,EAAE,CAAC;oBAAE,IAAI,CAAC,GAAG,CAAC,GAAG,IAAI,IAAI,GAAG,EAAE,CAAC,CAAA;YAC9F,IAAI,CAAC,CAAC,MAAM,CAAC,GAAG,IAAI,MAAM,CAAC;gBAAE,IAAI,CAAC,GAAG,CAAC,GAAG,IAAI,MAAM,CAAC,CAAA;QACtD,CAAC;QACD,KAAK,MAAM,CAAC,IAAI,qBAAqB;YAAE,IAAI,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC;gBAAE,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,CAAA;QACnE,KAAK,MAAM,IAAI,IAAI,CAAC,CAAC,OAAO,CAAC,YAAY,CAAC,EAAE,CAAC;YAC3C,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,OAAO,CAAC;gBAAE,SAAQ;YACvC,KAAK,MAAM,GAAG,IAAI,CAAC,OAAO,EAAE,KAAK,EAAE,KAAK,CAAC,EAAE,CAAC;gBAC1C,MAAM,CAAC,GAAG,cAAc,IAAI,IAAI,GAAG,EAAE,CAAA;gBACrC,IAAI,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC;oBAAE,SAAQ;gBAC1B,IAAI,GAAG,KAAK,KAAK;oBAAE,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,CAAA;;oBACzB,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,CAAA;YAClB,CAAC;QACH,CAAC;IACH,CAAC;IACD,OAAO,EAAE,OAAO,EAAE,CAAC,GAAG,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,OAAO,EAAE,CAAC,GAAG,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAA;AACjE,CAAC;AAED,0FAA0F;AAC1F,MAAM,UAAU,uBAAuB,CAAC,IAAY,EAAE,QAAyB,EAAE,MAAM,GAAG,KAAK;IAC7F,IAAI,QAAQ,KAAK,OAAO,EAAE,CAAC;QACzB,OAAO,CAAC,QAAQ,EAAE,QAAQ,EAAE,OAAO,EAAE,QAAQ,EAAE,UAAU,EAAE,UAAU,EAAE,OAAO,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CACzF,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC,CACjB,CAAA;IACH,CAAC;IACD,IAAI,QAAQ,KAAK,OAAO,EAAE,CAAC;QACzB,MAAM,KAAK,GAAG,MAAM;YAClB,CAAC,CAAC,CAAC,WAAW,EAAE,WAAW,EAAE,UAAU,EAAE,WAAW,EAAE,aAAa,EAAE,aAAa,EAAE,UAAU,CAAC;YAC/F,CAAC,CAAC,CAAC,WAAW,EAAE,WAAW,EAAE,UAAU,CAAC,CAAA;QAC1C,OAAO,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAA;IAC5C,CAAC;IACD,OAAO,KAAK,CAAA;AACd,CAAC;AAmBD,iGAAiG;AACjG,MAAM,UAAU,mBAAmB,CAAC,GAAW;IAC7C,OAAO,GAAG,CAAC,UAAU,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,CAAA;AACvD,CAAC;AAED,MAAM,OAAO,GAAG,CAAC,KAAe,EAAE,QAA4B,EAAE,GAAW,EAAU,EAAE;IACrF,MAAM,IAAI,GAAG,CAAC,QAAQ,IAAI,EAAE,CAAC,CAAC,OAAO,CAAC,IAAI,MAAM,CAAC,KAAK,GAAG,IAAI,CAAC,EAAE,EAAE,CAAC,CAAA;IACnE,OAAO,IAAI,KAAK,EAAE,CAAC,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,KAAK,CAAC,IAAI,CAAC,GAAG,CAAC,GAAG,GAAG,GAAG,IAAI,EAAE,CAAA;AAC1E,CAAC,CAAA;AAED,MAAM,UAAU,eAAe,CAAC,KAAsB;IACpD,MAAM,GAAG,GAA2B,EAAE,CAAA;IACtC,KAAK,MAAM,CAAC,CAAC,EAAE,CAAC,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,KAAK,CAAC,OAAO,CAAC;QAAE,IAAI,CAAC,KAAK,SAAS;YAAE,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC,CAAA;IACnF,MAAM,CAAC,MAAM,CAAC,GAAG,EAAE,KAAK,CAAC,OAAO,CAAC,CAAA;IACjC,MAAM,SAAS,GAAG,KAAK,CAAC,IAAI,CAAC,OAAO,CAAC,MAAM,GAAG,CAAC,IAAI,KAAK,CAAC,IAAI,CAAC,OAAO,CAAC,MAAM,GAAG,CAAC,CAAA;IAChF,IAAI,GAAuB,CAAA;IAC3B,QAAQ,KAAK,CAAC,QAAQ,EAAE,CAAC;QACvB,KAAK,OAAO,CAAC,CAAC,CAAC;YACb,MAAM,IAAI,GAAG,CAAC,KAAK,CAAC,MAAM,EAAE,GAAG,KAAK,CAAC,IAAI,CAAC,OAAO,CAAC,CAAA;YAClD,GAAG,CAAC,iBAAiB,CAAC,GAAG,OAAO,CAAC,IAAI,EAAE,GAAG,CAAC,iBAAiB,CAAC,EAAE,GAAG,CAAC,CAAA;YACnE,IAAI,KAAK,CAAC,IAAI,CAAC,OAAO,CAAC,MAAM;gBAAE,GAAG,CAAC,MAAM,CAAC,GAAG,OAAO,CAAC,KAAK,CAAC,IAAI,CAAC,OAAO,EAAE,GAAG,CAAC,MAAM,CAAC,EAAE,GAAG,CAAC,CAAA;YAC1F,MAAK;QACP,CAAC;QACD,KAAK,OAAO,CAAC,CAAC,CAAC;YACb,MAAM,GAAG,GAAG,mBAAmB,CAAC,KAAK,CAAC,MAAM,CAAC,CAAA;YAC7C,GAAG,CAAC,MAAM,CAAC,GAAG,OAAO,CAAC,CAAC,GAAG,EAAE,GAAG,KAAK,CAAC,IAAI,CAAC,OAAO,CAAC,EAAE,GAAG,CAAC,MAAM,CAAC,IAAI,GAAG,CAAC,MAAM,CAAC,EAAE,GAAG,CAAC,CAAA;YACpF,OAAO,GAAG,CAAC,MAAM,CAAC,CAAA;YAClB,GAAG,GAAG,KAAK,CAAC,MAAM,CAAA;YAClB,MAAK;QACP,CAAC;QACD,KAAK,QAAQ,CAAC,CAAC,CAAC;YACd,GAAG,CAAC,mBAAmB,CAAC,GAAG,OAAO,CAAC,CAAC,KAAK,CAAC,MAAM,CAAC,EAAE,GAAG,CAAC,mBAAmB,CAAC,EAAE,GAAG,CAAC,CAAA;YACjF,MAAK;QACP,CAAC;QACD;YACE,MAAK;IACT,CAAC;IACD,OAAO,EAAE,GAAG,EAAE,GAAG,EAAE,SAAS,EAAE,CAAA;AAChC,CAAC;AAED,kEAAkE;AAClE,MAAM,UAAU,QAAQ,CAAC,OAAe;IACtC,MAAM,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,OAAO,CAAC,WAAW,CAAC,GAAG,CAAC,EAAE,OAAO,CAAC,WAAW,CAAC,IAAI,CAAC,CAAC,CAAA;IACvE,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAA;AAC1C,CAAC;AAED,0GAA0G;AAC1G,MAAM,UAAU,OAAO,CAAC,QAAyB,EAAE,GAAG,KAAe;IACnE,OAAO,QAAQ,KAAK,OAAO,CAAC,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,KAAK,CAAC,CAAA;AACjE,CAAC"}

package/dist/runtime/shared/external-sessions.d.ts

@@ -0,0 +1,67 @@
+/**
+ * Sessions another process owns — the desktop app's TurboQuant, MLX and Foundation Models engines,
+ * and its llama.cpp upstream engine while that runtime is still the app's — registered with the core
+ * so the Local API Server it serves can route to them (PLAN.md §4, stage 4d).
+ *
+ * The owner publishes its whole list as a snapshot with a generation, and keeps it alive with
+ * heartbeats. A snapshot from an older generation is refused, so a restarted app cannot be
+ * overwritten by a request its previous run left in flight. A registration that stops beating
+ * expires and its sessions vanish from routing: the core never kills a process it does not own, it
+ * only stops sending traffic to it.
+ *
+ * Growing a registered session's context is the owner's job — it holds the process and its settings.
+ * The core asks with an event carrying a request id and waits for the owner's answer, with the same
+ * 60-second limit the app's own proxy gives its extension.
+ */
+import type { CoreEvents, LocalProviderId } from '../../contracts/index.js';
+export declare const EXTERNAL_SESSION_TTL_MS = 30000;
+export declare const EXTERNAL_CTX_TIMEOUT_MS = 60000;
+export interface ExternalSession {
+    provider: LocalProviderId;
+    model_id: string;
+    port: number;
+    api_key: string;
+    is_embedding: boolean;
+    pid: number | null;
+}
+export interface ExternalCtxOutcome {
+    ok: boolean;
+    new_ctx_len?: number;
+    reason?: string;
+}
+export interface ExternalSessionsOptions {
+    emit: <K extends keyof CoreEvents>(name: K, payload: CoreEvents[K]) => void;
+    now?: () => number;
+    ttlMs?: number;
+    ctxTimeoutMs?: number;
+}
+export declare class ExternalSessions {
+    private readonly options;
+    private readonly owners;
+    private readonly pendingCtx;
+    constructor(options: ExternalSessionsOptions);
+    private now;
+    private live;
+    private drop;
+    /** Replace an owner's snapshot. An older generation than the one held is a stale request: refused. */
+    publish(owner: string, generation: number, sessions: unknown): {
+        generation: number;
+        sessions: number;
+    };
+    /** Keep a registration alive. A heartbeat for a generation that is no longer held says so. */
+    heartbeat(owner: string, generation: number): {
+        alive: boolean;
+    };
+    unregister(owner: string, generation?: number): boolean;
+    /** Every live registered session, owner attached. */
+    list(): Array<ExternalSession & {
+        owner: string;
+    }>;
+    find(provider: string, matches: (modelId: string) => boolean): (ExternalSession & {
+        owner: string;
+    }) | undefined;
+    /** Ask the owner to grow a session's context, and wait for its answer or the timeout. */
+    requestCtxIncrease(owner: string, provider: string, modelId: string, trigger: string): Promise<ExternalCtxOutcome>;
+    /** The owner's answer. An unknown or already-answered request is not an error, only `false`. */
+    answerCtxIncrease(owner: string, requestId: string, outcome: unknown): boolean;
+}