@fraziym/axoniz 0.1.0

package/dist/core/backend/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../../src/core/backend/index.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AACH,OAAO,EAAE,KAAK,EAAE,MAAM,aAAa,CAAC;AACpC,OAAO,EAAE,cAAc,EAAE,cAAc,EAAqB,MAAM,cAAc,CAAC;AACjF,OAAO,EAAE,eAAe,EAAE,MAAM,eAAe,CAAC;AAChD,OAAO,EAAE,qBAAqB,EAAE,MAAM,mBAAmB,CAAC;AAC1D,OAAO,EAAE,uBAAuB,EAAE,MAAM,aAAa,CAAC;AACtD,OAAO,EAAE,aAAa,EAAE,MAAM,aAAa,CAAC;AAC5C,OAAO,EAAE,eAAe,EAAE,MAAM,eAAe,CAAC;AAChD,OAAO,EACL,OAAO,EACP,kBAAkB,EAClB,YAAY,EACZ,gBAAgB,GAOjB,MAAM,YAAY,CAAC;AAEpB,OAAO,EACL,OAAO,EACP,kBAAkB,EAClB,YAAY,EACZ,gBAAgB,EAChB,eAAe,EACf,qBAAqB,EACrB,uBAAuB,EACvB,aAAa,EACb,eAAe,GAChB,CAAC;AAUF;;;GAGG;AACH,MAAM,UAAU,UAAU,CAAC,GAAiB,EAAE,KAAK,GAAc,EAAE;IACjE,MAAM,MAAM,GAAG,GAAG,CAAC,GAAG,KAAK,SAAS,IAAI,GAAG,CAAC,GAAG,KAAK,IAAI,CAAC;IACzD,MAAM,IAAI,GAAG,MAAM;QACjB,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,GAAG,CAAC,eAAe,IAAI,UAAU,CAAC;QAC/C,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,QAAQ,IAAI,GAAG,CAAC,OAAO,IAAI,UAAU,CAAC;aAC9C,WAAW,EAAE;aACb,IAAI,EAAE,CAAC;IACd,MAAM,CAAC,GAAG,cAAc,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC;IAEpC,KAAK,CAAC,8BAA8B,IAAI,EAAE,CAAC,CAAC;IAE5C,MAAM,GAAG,GAAG,CAAC,GAAW,EAAE,QAAgB,EAAU,EAAE;QACpD,MAAM,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC;QACjB,MAAM,CAAC,GAAG,MAAM,CAAC,CAAC,CAAC,CAAC;QACpB,OAAO,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC;IAC3C,CAAC,CAAC;IACF,MAAM,GAAG,GAAG,CAAC,GAAW,EAAE,QAAgB,EAAU,EAAE;QACpD,MAAM,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC;QACjB,OAAO,CAAC,KAAK,SAAS,IAAI,CAAC,KAAK,IAAI,IAAI,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC;IAC1E,CAAC,CAAC;IAEF,IAAI,IAAI,KAAK,QAAQ,IAAI,IAAI,KAAK,mBAAmB,EAAE,CAAC;QACtD,OAAO,IAAI,uBAAuB,CAAC;YACjC,OAAO,EAAE,GAAG,CAAC,UAAU,EAAE,2BAA2B,CAAC;YACrD,MAAM,EAAE,GAAG,CAAC,SAAS,EAAE,EAAE,CAAC;YAC1B,SAAS,EAAE,GAAG,CAAC,YAAY,EAAE,QAAQ,CAAC;YACtC,WAAW,EAAE,GAAG,CAAC,aAAa,EAAE,GAAG,CAAC;YACpC,SAAS,EAAE,GAAG,CAAC,YAAY,EAAE,IAAI,CAAC;YAClC,KAAK;SACN,CAAC,CAAC;IACL,CAAC;IAED,IAAI,IAAI,KAAK,QAAQ,EAAE,CAAC;QACtB,OAAO,IAAI,aAAa,CAAC;YACvB,OAAO,EAAE,GAAG,CAAC,UAAU,EAAE,wBAAwB,CAAC;YAClD,SAAS,EAAE,GAAG,CAAC,YAAY,EAAE,QAAQ,CAAC;YACtC,WAAW,EAAE,GAAG,CAAC,aAAa,EAAE,GAAG,CAAC;YACpC,SAAS,EAAE,GAAG,CAAC,YAAY,EAAE,IAAI,CAAC;YAClC,KAAK;SACN,CAAC,CAAC;IACL,CAAC;IAED,IAAI,IAAI,KAAK,UAAU,EAAE,CAAC;QACxB,OAAO,IAAI,eAAe,CAAC;YACzB,OAAO,EAAE,GAAG,CAAC,UAAU,EAAE,uBAAuB,CAAC;YACjD,SAAS,EAAE,GAAG,CAAC,YAAY,EAAE,EAAE,CAAC;YAChC,MAAM,EAAE,GAAG,CAAC,SAAS,EAAE,WAAW,CAAC;YACnC,WAAW,EAAE,GAAG,CAAC,aAAa,EAAE,GAAG,CAAC;YACpC,SAAS,EAAE,GAAG,CAAC,YAAY,EAAE,IAAI,CAAC;YAClC,QAAQ,EAAE,CAAC,CAAC,SAAS,KAAK,SAAS,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,SAAS,CAAC;YACjE,SAAS,EAAE,OAAO,CAAC,CAAC,CAAC,UAAU,CAAC;YAChC,MAAM,EAAE,GAAG,CAAC,SAAS,EAAE,EAAE,CAAC;YAC1B,KAAK;SACN,CAAC,CAAC;IACL,CAAC;IAED,IAAI,IAAI,KAAK,aAAa,IAAI,IAAI,KAAK,iBAAiB,EAAE,CAAC;QACzD,OAAO,IAAI,qBAAqB,CAAC;YAC/B,OAAO,EAAE,GAAG,CAAC,UAAU,EAAE,uBAAuB,CAAC;YACjD,SAAS,EAAE,GAAG,CAAC,YAAY,EAAE,EAAE,CAAC;YAChC,WAAW,EAAE,GAAG,CAAC,aAAa,EAAE,GAAG,CAAC;YACpC,SAAS,EAAE,GAAG,CAAC,YAAY,EAAE,IAAI,CAAC;YAClC,IAAI,EAAE,GAAG,CAAC,OAAO,EAAE,KAAK,CAAC;YACzB,QAAQ,EAAE,GAAG,CAAC,WAAW,EAAE,EAAE,CAAC;YAC9B,SAAS,EAAE,OAAO,CAAC,CAAC,CAAC,UAAU,CAAC;YAChC,UAAU,EAAE,GAAG,CAAC,cAAc,EAAE,CAAC,CAAC,CAAC;YACnC,KAAK;SACN,CAAC,CAAC;IACL,CAAC;IAED,iDAAiD;IACjD,OAAO,IAAI,eAAe,CAAC;QACzB,SAAS,EAAE,GAAG,CAAC,YAAY,EAAE,EAAE,CAAC;QAChC,WAAW,EAAE,GAAG,CAAC,aAAa,EAAE,GAAG,CAAC;QACpC,SAAS,EAAE,GAAG,CAAC,YAAY,EAAE,IAAI,CAAC;QAClC,IAAI,EAAE,GAAG,CAAC,OAAO,EAAE,KAAK,CAAC;QACzB,UAAU,EAAE,GAAG,CAAC,cAAc,EAAE,CAAC,CAAC,CAAC;QACnC,KAAK;KACN,CAAC,CAAC;AACL,CAAC;AASD,+CAA+C;AAC/C,MAAM,UAAU,sBAAsB;IACpC,OAAO;QACL;YACE,EAAE,EAAE,UAAU;YACd,IAAI,EAAE,wBAAwB;YAC9B,QAAQ,EAAE,gBAAgB;YAC1B,YAAY,EAAE,iBAAiB;SAChC;QACD;YACE,EAAE,EAAE,iBAAiB;YACrB,IAAI,EAAE,iBAAiB;YACvB,QAAQ,EAAE,qBAAqB;YAC/B,YAAY,EAAE,iBAAiB;SAChC;QACD;YACE,EAAE,EAAE,UAAU;YACd,IAAI,EAAE,iBAAiB;YACvB,QAAQ,EAAE,mBAAmB;YAC7B,YAAY,EAAE,oBAAoB;SACnC;QACD;YACE,EAAE,EAAE,QAAQ;YACZ,IAAI,EAAE,qBAAqB;YAC3B,QAAQ,EAAE,oBAAoB;YAC9B,YAAY,EAAE,mBAAmB;SAClC;QACD;YACE,EAAE,EAAE,QAAQ;YACZ,IAAI,EAAE,QAAQ;YACd,QAAQ,EAAE,gBAAgB;YAC1B,YAAY,EAAE,mBAAmB;SAClC;KACF,CAAC;AACJ,CAAC;AAED,OAAO,EAAE,cAAc,EAAE,CAAC"}

package/dist/core/backend/llama_server.d.ts

@@ -0,0 +1,39 @@
+import { Backend, ConnectionError, type BackendHealth, type ChatMessage, type CompletionResult } from "./types.js";
+export interface LlamaCppServerOptions {
+    baseUrl?: string;
+    modelPath?: string;
+    temperature?: number;
+    maxTokens?: number;
+    nCtx?: number;
+    llamaBin?: string;
+    autoStart?: boolean;
+    nGpuLayers?: number;
+    tools?: unknown[];
+}
+export declare class LlamaCppServerBackend extends Backend {
+    readonly baseUrl: string;
+    readonly modelPath: string;
+    readonly temperature: number;
+    readonly maxTokens: number;
+    readonly nCtx: number;
+    readonly llamaBin: string;
+    readonly autoStart: boolean;
+    readonly nGpuLayers: number;
+    private proc;
+    private starting;
+    constructor(opts?: LlamaCppServerOptions);
+    get name(): string;
+    private headers;
+    private port;
+    /** Resolve the llama-server executable path. */
+    private resolveBinary;
+    /** Ensure a llama-server is reachable, starting one when configured to. */
+    ensureServer(): Promise<void>;
+    load(): Promise<string>;
+    complete(messages: ChatMessage[], maxTokens?: number): Promise<CompletionResult>;
+    streamText(messages: ChatMessage[], signal?: AbortSignal, maxTokens?: number): AsyncGenerator<string>;
+    healthCheck(): Promise<BackendHealth>;
+    unload(): void;
+}
+export { ConnectionError };
+//# sourceMappingURL=llama_server.d.ts.map

package/dist/core/backend/llama_server.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"llama_server.d.ts","sourceRoot":"","sources":["../../../src/core/backend/llama_server.ts"],"names":[],"mappings":"AAYA,OAAO,EACL,OAAO,EACP,eAAe,EAOf,KAAK,aAAa,EAClB,KAAK,WAAW,EAChB,KAAK,gBAAgB,EACtB,MAAM,YAAY,CAAC;AAGpB,MAAM,WAAW,qBAAqB;IACpC,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,KAAK,CAAC,EAAE,OAAO,EAAE,CAAC;CACnB;AAED,qBAAa,qBAAsB,SAAQ,OAAO;IAChD,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;IACzB,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;IAC7B,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;IAC1B,QAAQ,CAAC,SAAS,EAAE,OAAO,CAAC;IAC5B,QAAQ,CAAC,UAAU,EAAE,MAAM,CAAC;IAE5B,OAAO,CAAC,IAAI,CAA6B;IACzC,OAAO,CAAC,QAAQ,CAAS;IAEzB,YAAY,IAAI,GAAE,qBAA0B,EAW3C;IAED,IAAI,IAAI,IAAI,MAAM,CAEjB;IAED,OAAO,CAAC,OAAO;IAIf,OAAO,CAAC,IAAI;IAUZ,gDAAgD;IAChD,OAAO,CAAC,aAAa;IAMrB,2EAA2E;IACrE,YAAY,IAAI,OAAO,CAAC,IAAI,CAAC,CA+ClC;IAEc,IAAI,IAAI,OAAO,CAAC,MAAM,CAAC,CAIrC;IAEK,QAAQ,CAAC,QAAQ,EAAE,WAAW,EAAE,EAAE,SAAS,CAAC,EAAE,MAAM,GAAG,OAAO,CAAC,gBAAgB,CAAC,CAoBrF;IAEM,UAAU,CACf,QAAQ,EAAE,WAAW,EAAE,EACvB,MAAM,CAAC,EAAE,WAAW,EACpB,SAAS,CAAC,EAAE,MAAM,GACjB,cAAc,CAAC,MAAM,CAAC,CAsBxB;IAEK,WAAW,IAAI,OAAO,CAAC,aAAa,CAAC,CAY1C;IAEQ,MAAM,IAAI,IAAI,CAStB;CACF;AAMD,OAAO,EAAE,eAAe,EAAE,CAAC"}

package/dist/core/backend/llama_server.js

@@ -0,0 +1,182 @@
+/**
+ * LlamaCppServer backend — connects to a running `llama-server` (or auto-starts one).
+ * Port of `LlamaCppServerBackend` from axoniz/core/backend.py.
+ *
+ * NOTE: node-llama-cpp does NOT ship a `serve` command, so auto-start spawns the
+ * real llama.cpp `llama-server` binary (config `llama_bin`, or `llama-server` on PATH).
+ */
+import fs from "node:fs";
+import os from "node:os";
+import { spawn } from "node:child_process";
+import { whichSync } from "../which.js";
+import { Backend, ConnectionError, errMsg, httpPost, httpStreamLines, HttpStatusError, parseSseDelta, textResponse, } from "./types.js";
+import { debug, warn } from "../debug.js";
+export class LlamaCppServerBackend extends Backend {
+    baseUrl;
+    modelPath;
+    temperature;
+    maxTokens;
+    nCtx;
+    llamaBin;
+    autoStart;
+    nGpuLayers;
+    proc = null;
+    starting = false;
+    constructor(opts = {}) {
+        super();
+        this.baseUrl = (opts.baseUrl ?? "http://localhost:8080").replace(/\/+$/, "");
+        this.modelPath = opts.modelPath ?? "";
+        this.temperature = opts.temperature ?? 0.2;
+        this.maxTokens = opts.maxTokens ?? 4096;
+        this.nCtx = opts.nCtx ?? 8192;
+        this.llamaBin = opts.llamaBin ?? "";
+        this.autoStart = opts.autoStart ?? false;
+        this.nGpuLayers = opts.nGpuLayers ?? -1;
+        this.tools = opts.tools ?? [];
+    }
+    get name() {
+        return "llamacpp_server";
+    }
+    headers() {
+        return { "Content-Type": "application/json" };
+    }
+    port() {
+        try {
+            const u = new URL(this.baseUrl);
+            return u.port || "8080";
+        }
+        catch {
+            const m = this.baseUrl.split(":").pop() ?? "8080";
+            return m.split("/")[0] || "8080";
+        }
+    }
+    /** Resolve the llama-server executable path. */
+    resolveBinary() {
+        if (this.llamaBin && fs.existsSync(this.llamaBin))
+            return this.llamaBin;
+        const name = this.llamaBin || "llama-server";
+        return whichSync(name);
+    }
+    /** Ensure a llama-server is reachable, starting one when configured to. */
+    async ensureServer() {
+        const health = await this.healthCheck();
+        if (health.status === "ok")
+            return;
+        if (this.starting)
+            return;
+        if (!this.autoStart || !this.modelPath)
+            return;
+        const binPath = this.resolveBinary();
+        if (!binPath) {
+            warn(`[LlamaCppServer] binary not found: ${this.llamaBin || "llama-server"}`);
+            return;
+        }
+        this.starting = true;
+        try {
+            const args = [
+                "-m",
+                this.modelPath,
+                "--port",
+                this.port(),
+                "--ctx-size",
+                String(this.nCtx),
+                "-ngl",
+                String(this.nGpuLayers),
+                "--threads",
+                String(Math.max(1, os.cpus().length - 1)),
+            ];
+            this.proc = spawn(binPath, args, {
+                stdio: "ignore",
+                detached: false,
+                windowsHide: true,
+            });
+            this.proc.on("error", (e) => {
+                warn(`[LlamaCppServer] spawn failed: ${e.message}`);
+            });
+            for (let i = 0; i < 120; i++) {
+                await sleep(500);
+                if ((await this.healthCheck()).status === "ok") {
+                    debug(`[LlamaCppServer] started at ${this.baseUrl}`);
+                    return;
+                }
+                if (this.proc.exitCode !== null)
+                    break;
+            }
+            warn("[LlamaCppServer] server did not start in time");
+        }
+        finally {
+            this.starting = false;
+        }
+    }
+    async load() {
+        await this.ensureServer();
+        const h = await this.healthCheck();
+        return h.status === "ok" ? "ok" : `[ERROR] LlamaCppServer not reachable at ${this.baseUrl}`;
+    }
+    async complete(messages, maxTokens) {
+        const payload = {
+            model: "local",
+            messages,
+            temperature: this.temperature,
+            max_tokens: maxTokens ?? this.maxTokens,
+            stream: false,
+        };
+        try {
+            const resp = (await httpPost(`${this.baseUrl}/v1/chat/completions`, payload, this.headers()));
+            return textResponse(resp.choices?.[0]?.message?.content ?? "");
+        }
+        catch (e) {
+            return textResponse(`[ERROR] LlamaCppServer: ${errMsg(e)}`);
+        }
+    }
+    async *streamText(messages, signal, maxTokens) {
+        const payload = {
+            model: "local",
+            messages,
+            temperature: this.temperature,
+            max_tokens: maxTokens ?? this.maxTokens,
+            stream: true,
+        };
+        try {
+            for await (const line of httpStreamLines(`${this.baseUrl}/v1/chat/completions`, payload, this.headers(), undefined, signal)) {
+                const token = parseSseDelta(line);
+                if (token)
+                    yield token;
+            }
+        }
+        catch (e) {
+            yield `[ERROR] LlamaCppServer stream: ${errMsg(e)}`;
+        }
+    }
+    async healthCheck() {
+        try {
+            const res = await fetch(`${this.baseUrl}/health`, { signal: AbortSignal.timeout(3000) });
+            if (!res.ok)
+                throw new HttpStatusError(res.status, await res.text().catch(() => ""));
+            return { status: "ok", backend: "llamacpp_server", url: this.baseUrl };
+        }
+        catch (e) {
+            return {
+                status: "error",
+                error: e instanceof HttpStatusError ? e.message : errMsg(e),
+                backend: "llamacpp_server",
+            };
+        }
+    }
+    unload() {
+        if (this.proc && this.proc.exitCode === null) {
+            try {
+                this.proc.kill();
+            }
+            catch {
+                /* ignore */
+            }
+        }
+        this.proc = null;
+    }
+}
+function sleep(ms) {
+    return new Promise((r) => setTimeout(r, ms));
+}
+export { ConnectionError };
+//# sourceMappingURL=llama_server.js.map

package/dist/core/backend/llama_server.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"llama_server.js","sourceRoot":"","sources":["../../../src/core/backend/llama_server.ts"],"names":[],"mappings":"AAAA;;;;;;GAMG;AACH,OAAO,EAAE,MAAM,SAAS,CAAC;AACzB,OAAO,EAAE,MAAM,SAAS,CAAC;AAEzB,OAAO,EAAE,KAAK,EAAqB,MAAM,oBAAoB,CAAC;AAC9D,OAAO,EAAE,SAAS,EAAE,MAAM,aAAa,CAAC;AACxC,OAAO,EACL,OAAO,EACP,eAAe,EACf,MAAM,EACN,QAAQ,EACR,eAAe,EACf,eAAe,EACf,aAAa,EACb,YAAY,GAIb,MAAM,YAAY,CAAC;AACpB,OAAO,EAAE,KAAK,EAAE,IAAI,EAAE,MAAM,aAAa,CAAC;AAc1C,MAAM,OAAO,qBAAsB,SAAQ,OAAO;IACvC,OAAO,CAAS;IAChB,SAAS,CAAS;IAClB,WAAW,CAAS;IACpB,SAAS,CAAS;IAClB,IAAI,CAAS;IACb,QAAQ,CAAS;IACjB,SAAS,CAAU;IACnB,UAAU,CAAS;IAEpB,IAAI,GAAwB,IAAI,CAAC;IACjC,QAAQ,GAAG,KAAK,CAAC;IAEzB,YAAY,IAAI,GAA0B,EAAE;QAC1C,KAAK,EAAE,CAAC;QACR,IAAI,CAAC,OAAO,GAAG,CAAC,IAAI,CAAC,OAAO,IAAI,uBAAuB,CAAC,CAAC,OAAO,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;QAC7E,IAAI,CAAC,SAAS,GAAG,IAAI,CAAC,SAAS,IAAI,EAAE,CAAC;QACtC,IAAI,CAAC,WAAW,GAAG,IAAI,CAAC,WAAW,IAAI,GAAG,CAAC;QAC3C,IAAI,CAAC,SAAS,GAAG,IAAI,CAAC,SAAS,IAAI,IAAI,CAAC;QACxC,IAAI,CAAC,IAAI,GAAG,IAAI,CAAC,IAAI,IAAI,IAAI,CAAC;QAC9B,IAAI,CAAC,QAAQ,GAAG,IAAI,CAAC,QAAQ,IAAI,EAAE,CAAC;QACpC,IAAI,CAAC,SAAS,GAAG,IAAI,CAAC,SAAS,IAAI,KAAK,CAAC;QACzC,IAAI,CAAC,UAAU,GAAG,IAAI,CAAC,UAAU,IAAI,CAAC,CAAC,CAAC;QACxC,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,KAAK,IAAI,EAAE,CAAC;IAChC,CAAC;IAED,IAAI,IAAI;QACN,OAAO,iBAAiB,CAAC;IAC3B,CAAC;IAEO,OAAO;QACb,OAAO,EAAE,cAAc,EAAE,kBAAkB,EAAE,CAAC;IAChD,CAAC;IAEO,IAAI;QACV,IAAI,CAAC;YACH,MAAM,CAAC,GAAG,IAAI,GAAG,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;YAChC,OAAO,CAAC,CAAC,IAAI,IAAI,MAAM,CAAC;QAC1B,CAAC;QAAC,MAAM,CAAC;YACP,MAAM,CAAC,GAAG,IAAI,CAAC,OAAO,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,IAAI,MAAM,CAAC;YAClD,OAAO,CAAC,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,MAAM,CAAC;QACnC,CAAC;IACH,CAAC;IAED,gDAAgD;IACxC,aAAa;QACnB,IAAI,IAAI,CAAC,QAAQ,IAAI,EAAE,CAAC,UAAU,CAAC,IAAI,CAAC,QAAQ,CAAC;YAAE,OAAO,IAAI,CAAC,QAAQ,CAAC;QACxE,MAAM,IAAI,GAAG,IAAI,CAAC,QAAQ,IAAI,cAAc,CAAC;QAC7C,OAAO,SAAS,CAAC,IAAI,CAAC,CAAC;IACzB,CAAC;IAED,2EAA2E;IAC3E,KAAK,CAAC,YAAY;QAChB,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,WAAW,EAAE,CAAC;QACxC,IAAI,MAAM,CAAC,MAAM,KAAK,IAAI;YAAE,OAAO;QACnC,IAAI,IAAI,CAAC,QAAQ;YAAE,OAAO;QAC1B,IAAI,CAAC,IAAI,CAAC,SAAS,IAAI,CAAC,IAAI,CAAC,SAAS;YAAE,OAAO;QAE/C,MAAM,OAAO,GAAG,IAAI,CAAC,aAAa,EAAE,CAAC;QACrC,IAAI,CAAC,OAAO,EAAE,CAAC;YACb,IAAI,CAAC,sCAAsC,IAAI,CAAC,QAAQ,IAAI,cAAc,EAAE,CAAC,CAAC;YAC9E,OAAO;QACT,CAAC;QAED,IAAI,CAAC,QAAQ,GAAG,IAAI,CAAC;QACrB,IAAI,CAAC;YACH,MAAM,IAAI,GAAG;gBACX,IAAI;gBACJ,IAAI,CAAC,SAAS;gBACd,QAAQ;gBACR,IAAI,CAAC,IAAI,EAAE;gBACX,YAAY;gBACZ,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC;gBACjB,MAAM;gBACN,MAAM,CAAC,IAAI,CAAC,UAAU,CAAC;gBACvB,WAAW;gBACX,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI,EAAE,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;aAC1C,CAAC;YACF,IAAI,CAAC,IAAI,GAAG,KAAK,CAAC,OAAO,EAAE,IAAI,EAAE;gBAC/B,KAAK,EAAE,QAAQ;gBACf,QAAQ,EAAE,KAAK;gBACf,WAAW,EAAE,IAAI;aAClB,CAAC,CAAC;YACH,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,OAAO,EAAE,CAAC,CAAC,EAAE,EAAE;gBAC1B,IAAI,CAAC,kCAAkC,CAAC,CAAC,OAAO,EAAE,CAAC,CAAC;YACtD,CAAC,CAAC,CAAC;YAEH,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC;gBAC7B,MAAM,KAAK,CAAC,GAAG,CAAC,CAAC;gBACjB,IAAI,CAAC,MAAM,IAAI,CAAC,WAAW,EAAE,CAAC,CAAC,MAAM,KAAK,IAAI,EAAE,CAAC;oBAC/C,KAAK,CAAC,+BAA+B,IAAI,CAAC,OAAO,EAAE,CAAC,CAAC;oBACrD,OAAO;gBACT,CAAC;gBACD,IAAI,IAAI,CAAC,IAAI,CAAC,QAAQ,KAAK,IAAI;oBAAE,MAAM;YACzC,CAAC;YACD,IAAI,CAAC,+CAA+C,CAAC,CAAC;QACxD,CAAC;gBAAS,CAAC;YACT,IAAI,CAAC,QAAQ,GAAG,KAAK,CAAC;QACxB,CAAC;IACH,CAAC;IAEQ,KAAK,CAAC,IAAI;QACjB,MAAM,IAAI,CAAC,YAAY,EAAE,CAAC;QAC1B,MAAM,CAAC,GAAG,MAAM,IAAI,CAAC,WAAW,EAAE,CAAC;QACnC,OAAO,CAAC,CAAC,MAAM,KAAK,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,2CAA2C,IAAI,CAAC,OAAO,EAAE,CAAC;IAC9F,CAAC;IAED,KAAK,CAAC,QAAQ,CAAC,QAAuB,EAAE,SAAkB;QACxD,MAAM,OAAO,GAAG;YACd,KAAK,EAAE,OAAO;YACd,QAAQ;YACR,WAAW,EAAE,IAAI,CAAC,WAAW;YAC7B,UAAU,EAAE,SAAS,IAAI,IAAI,CAAC,SAAS;YACvC,MAAM,EAAE,KAAK;SACd,CAAC;QACF,IAAI,CAAC;YACH,MAAM,IAAI,GAAG,CAAC,MAAM,QAAQ,CAC1B,GAAG,IAAI,CAAC,OAAO,sBAAsB,EACrC,OAAO,EACP,IAAI,CAAC,OAAO,EAAE,CACf,CAEA,CAAC;YACF,OAAO,YAAY,CAAC,IAAI,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,OAAO,IAAI,EAAE,CAAC,CAAC;QACjE,CAAC;QAAC,OAAO,CAAC,EAAE,CAAC;YACX,OAAO,YAAY,CAAC,2BAA2B,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;QAC9D,CAAC;IACH,CAAC;IAED,KAAK,CAAC,CAAC,UAAU,CACf,QAAuB,EACvB,MAAoB,EACpB,SAAkB;QAElB,MAAM,OAAO,GAAG;YACd,KAAK,EAAE,OAAO;YACd,QAAQ;YACR,WAAW,EAAE,IAAI,CAAC,WAAW;YAC7B,UAAU,EAAE,SAAS,IAAI,IAAI,CAAC,SAAS;YACvC,MAAM,EAAE,IAAI;SACb,CAAC;QACF,IAAI,CAAC;YACH,IAAI,KAAK,EAAE,MAAM,IAAI,IAAI,eAAe,CACtC,GAAG,IAAI,CAAC,OAAO,sBAAsB,EACrC,OAAO,EACP,IAAI,CAAC,OAAO,EAAE,EACd,SAAS,EACT,MAAM,CACP,EAAE,CAAC;gBACF,MAAM,KAAK,GAAG,aAAa,CAAC,IAAI,CAAC,CAAC;gBAClC,IAAI,KAAK;oBAAE,MAAM,KAAK,CAAC;YACzB,CAAC;QACH,CAAC;QAAC,OAAO,CAAC,EAAE,CAAC;YACX,MAAM,kCAAkC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC;QACtD,CAAC;IACH,CAAC;IAED,KAAK,CAAC,WAAW;QACf,IAAI,CAAC;YACH,MAAM,GAAG,GAAG,MAAM,KAAK,CAAC,GAAG,IAAI,CAAC,OAAO,SAAS,EAAE,EAAE,MAAM,EAAE,WAAW,CAAC,OAAO,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;YACzF,IAAI,CAAC,GAAG,CAAC,EAAE;gBAAE,MAAM,IAAI,eAAe,CAAC,GAAG,CAAC,MAAM,EAAE,MAAM,GAAG,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,EAAE,CAAC,CAAC,CAAC;YACrF,OAAO,EAAE,MAAM,EAAE,IAAI,EAAE,OAAO,EAAE,iBAAiB,EAAE,GAAG,EAAE,IAAI,CAAC,OAAO,EAAE,CAAC;QACzE,CAAC;QAAC,OAAO,CAAC,EAAE,CAAC;YACX,OAAO;gBACL,MAAM,EAAE,OAAO;gBACf,KAAK,EAAE,CAAC,YAAY,eAAe,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC;gBAC3D,OAAO,EAAE,iBAAiB;aAC3B,CAAC;QACJ,CAAC;IACH,CAAC;IAEQ,MAAM;QACb,IAAI,IAAI,CAAC,IAAI,IAAI,IAAI,CAAC,IAAI,CAAC,QAAQ,KAAK,IAAI,EAAE,CAAC;YAC7C,IAAI,CAAC;gBACH,IAAI,CAAC,IAAI,CAAC,IAAI,EAAE,CAAC;YACnB,CAAC;YAAC,MAAM,CAAC;gBACP,YAAY;YACd,CAAC;QACH,CAAC;QACD,IAAI,CAAC,IAAI,GAAG,IAAI,CAAC;IACnB,CAAC;CACF;AAED,SAAS,KAAK,CAAC,EAAU;IACvB,OAAO,IAAI,OAAO,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,UAAU,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC;AAC/C,CAAC;AAED,OAAO,EAAE,eAAe,EAAE,CAAC"}

package/dist/core/backend/llamacpp.d.ts

@@ -0,0 +1,40 @@
+import { Backend, type BackendHealth, type ChatMessage, type CompletionResult } from "./types.js";
+export interface LlamaCppOptions {
+    modelPath: string;
+    temperature?: number;
+    maxTokens?: number;
+    nCtx?: number;
+    nGpuLayers?: number;
+    tools?: unknown[];
+}
+export declare class LlamaCppBackend extends Backend {
+    readonly modelPath: string;
+    readonly temperature: number;
+    readonly maxTokens: number;
+    readonly nCtx: number;
+    readonly nGpuLayers: number;
+    private llama;
+    private model;
+    private context;
+    private session;
+    private loadError;
+    constructor(opts: LlamaCppOptions);
+    get name(): string;
+    load(): Promise<string>;
+    isLoaded(): boolean;
+    unload(): void;
+    private disposeAll;
+    /** Split conversation history into a system prompt and chat turns. */
+    private splitMessages;
+    /**
+     * node-llama-cpp chat sessions hold the KV cache across prompts, so replaying
+     * history each turn would double-apply it. We therefore prime a fresh session
+     * with the system prompt and send only the latest user turn, mirroring how
+     * llama-cpp-python reuses its cache while still receiving full history.
+     */
+    private buildPrompt;
+    complete(messages: ChatMessage[], maxTokens?: number): Promise<CompletionResult>;
+    streamText(messages: ChatMessage[], signal?: AbortSignal, maxTokens?: number): AsyncGenerator<string>;
+    healthCheck(): Promise<BackendHealth>;
+}
+//# sourceMappingURL=llamacpp.d.ts.map

package/dist/core/backend/llamacpp.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"llamacpp.d.ts","sourceRoot":"","sources":["../../../src/core/backend/llamacpp.ts"],"names":[],"mappings":"AAaA,OAAO,EACL,OAAO,EAGP,KAAK,aAAa,EAClB,KAAK,WAAW,EAChB,KAAK,gBAAgB,EACtB,MAAM,YAAY,CAAC;AA+BpB,MAAM,WAAW,eAAe;IAC9B,SAAS,EAAE,MAAM,CAAC;IAClB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,KAAK,CAAC,EAAE,OAAO,EAAE,CAAC;CACnB;AAED,qBAAa,eAAgB,SAAQ,OAAO;IAC1C,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;IAC7B,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,UAAU,EAAE,MAAM,CAAC;IAE5B,OAAO,CAAC,KAAK,CAA8B;IAC3C,OAAO,CAAC,KAAK,CAA2B;IACxC,OAAO,CAAC,OAAO,CAA6B;IAC5C,OAAO,CAAC,OAAO,CAAiC;IAChD,OAAO,CAAC,SAAS,CAAuB;IAExC,YAAY,IAAI,EAAE,eAAe,EAQhC;IAED,IAAI,IAAI,IAAI,MAAM,CAEjB;IAEc,IAAI,IAAI,OAAO,CAAC,MAAM,CAAC,CAqCrC;IAEQ,QAAQ,IAAI,OAAO,CAE3B;IAEQ,MAAM,IAAI,IAAI,CAEtB;YAEa,UAAU;IA2BxB,sEAAsE;IACtE,OAAO,CAAC,aAAa;IAUrB;;;;;OAKG;IACH,OAAO,CAAC,WAAW;IAeb,QAAQ,CAAC,QAAQ,EAAE,WAAW,EAAE,EAAE,SAAS,CAAC,EAAE,MAAM,GAAG,OAAO,CAAC,gBAAgB,CAAC,CAgBrF;IAEM,UAAU,CACf,QAAQ,EAAE,WAAW,EAAE,EACvB,MAAM,CAAC,EAAE,WAAW,EACpB,SAAS,CAAC,EAAE,MAAM,GACjB,cAAc,CAAC,MAAM,CAAC,CA6BxB;IAEK,WAAW,IAAI,OAAO,CAAC,aAAa,CAAC,CAS1C;CACF"}

package/dist/core/backend/llamacpp.js

@@ -0,0 +1,200 @@
+/**
+ * LlamaCpp in-process GGUF backend via node-llama-cpp.
+ * Port of `LlamaCppBackend` from axoniz/core/backend.py
+ * (llama-cpp-python -> node-llama-cpp).
+ *
+ * Prompt/KV-cache reuse maps onto a persistent LlamaContext + sequence, which is
+ * created once and reused across calls (the JS equivalent of the Python
+ * `Llama(...)` instance holding the KV cache).
+ */
+import os from "node:os";
+import path from "node:path";
+import fs from "node:fs";
+import { AsyncQueue } from "./async_queue.js";
+import { Backend, errMsg, textResponse, } from "./types.js";
+export class LlamaCppBackend extends Backend {
+    modelPath;
+    temperature;
+    maxTokens;
+    nCtx;
+    nGpuLayers;
+    llama = null;
+    model = null;
+    context = null;
+    session = null;
+    loadError = null;
+    constructor(opts) {
+        super();
+        this.modelPath = opts.modelPath;
+        this.temperature = opts.temperature ?? 0.2;
+        this.maxTokens = opts.maxTokens ?? 4096;
+        this.nCtx = opts.nCtx ?? 32768;
+        this.nGpuLayers = opts.nGpuLayers ?? -1;
+        this.tools = opts.tools ?? [];
+    }
+    get name() {
+        return "llamacpp";
+    }
+    async load() {
+        if (this.session)
+            return "ok";
+        if (!this.modelPath || !fs.existsSync(this.modelPath)) {
+            this.loadError = `[ERROR] GGUF not found: ${this.modelPath}`;
+            return this.loadError;
+        }
+        try {
+            // Dynamic import keeps node-llama-cpp optional: providers that do not need
+            // in-process inference must keep working when it is unavailable.
+            const mod = (await import("node-llama-cpp"));
+            this.llama = await mod.getLlama();
+            this.model = await this.llama.loadModel({
+                modelPath: this.modelPath,
+                gpuLayers: this.nGpuLayers,
+            });
+            this.context = await this.model.createContext({
+                contextSize: this.nCtx,
+                flashAttention: true,
+                batchSize: 512,
+                threads: Math.max(1, os.cpus().length - 1),
+            });
+            this.session = new mod.LlamaChatSession({ contextSequence: this.context.getSequence() });
+            this.loadError = null;
+            return "ok";
+        }
+        catch (e) {
+            const msg = errMsg(e);
+            // Distinguish a missing optional dependency from a genuine load failure.
+            this.loadError = msg.includes("Cannot find package") || msg.includes("ERR_MODULE_NOT_FOUND")
+                ? "[ERROR] node-llama-cpp not installed. Run: pnpm add node-llama-cpp"
+                : `[ERROR] LlamaCpp load failed: ${msg}`;
+            return this.loadError;
+        }
+    }
+    isLoaded() {
+        return this.session !== null;
+    }
+    unload() {
+        void this.disposeAll();
+    }
+    async disposeAll() {
+        try {
+            await this.session?.constructor;
+        }
+        catch {
+            /* ignore */
+        }
+        this.session = null;
+        try {
+            await this.context?.dispose?.();
+        }
+        catch {
+            /* ignore */
+        }
+        this.context = null;
+        try {
+            await this.model?.dispose?.();
+        }
+        catch {
+            /* ignore */
+        }
+        this.model = null;
+        try {
+            await this.llama?.dispose?.();
+        }
+        catch {
+            /* ignore */
+        }
+        this.llama = null;
+    }
+    /** Split conversation history into a system prompt and chat turns. */
+    splitMessages(messages) {
+        const systems = [];
+        const turns = [];
+        for (const m of messages) {
+            if (m.role === "system")
+                systems.push(m.content);
+            else
+                turns.push(m);
+        }
+        return { system: systems.join("\n\n"), turns };
+    }
+    /**
+     * node-llama-cpp chat sessions hold the KV cache across prompts, so replaying
+     * history each turn would double-apply it. We therefore prime a fresh session
+     * with the system prompt and send only the latest user turn, mirroring how
+     * llama-cpp-python reuses its cache while still receiving full history.
+     */
+    buildPrompt(messages) {
+        const { system, turns } = this.splitMessages(messages);
+        const lastUser = [...turns].reverse().find((t) => t.role === "user");
+        const priorContext = turns
+            .slice(0, -1)
+            .map((t) => `${t.role}: ${t.content}`)
+            .join("\n");
+        const parts = [];
+        if (system)
+            parts.push(system);
+        if (priorContext)
+            parts.push(`### CONVERSATION SO FAR\n${priorContext}`);
+        parts.push(lastUser?.content ?? "");
+        return parts.join("\n\n");
+    }
+    async complete(messages, maxTokens) {
+        if (!this.session) {
+            const res = await this.load();
+            if (res !== "ok")
+                return textResponse(res);
+        }
+        try {
+            const limit = maxTokens ?? this.maxTokens;
+            const prompt = this.buildPrompt(messages);
+            const content = await this.session.prompt(prompt, {
+                maxTokens: limit,
+                temperature: this.temperature,
+            });
+            return textResponse(content);
+        }
+        catch (e) {
+            return textResponse(`[ERROR] LlamaCpp: ${errMsg(e)}`);
+        }
+    }
+    async *streamText(messages, signal, maxTokens) {
+        if (!this.session) {
+            const res = await this.load();
+            if (res !== "ok") {
+                yield res;
+                return;
+            }
+        }
+        const queue = new AsyncQueue();
+        const prompt = this.buildPrompt(messages);
+        void (async () => {
+            try {
+                await this.session.prompt(prompt, {
+                    maxTokens: maxTokens ?? this.maxTokens,
+                    temperature: this.temperature,
+                    signal,
+                    onTextChunk: (chunk) => queue.push(chunk),
+                });
+                queue.close();
+            }
+            catch (e) {
+                queue.fail(new Error(`[ERROR] LlamaCpp stream: ${errMsg(e)}`));
+            }
+        })();
+        for await (const token of queue) {
+            yield token;
+        }
+    }
+    async healthCheck() {
+        const loaded = this.isLoaded();
+        return {
+            status: loaded ? "ok" : "not_loaded",
+            backend: "llamacpp",
+            model: this.modelPath ? path.basename(this.modelPath) : "none",
+            loaded,
+            ...(this.loadError ? { error: this.loadError } : {}),
+        };
+    }
+}
+//# sourceMappingURL=llamacpp.js.map

package/dist/core/backend/llamacpp.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"llamacpp.js","sourceRoot":"","sources":["../../../src/core/backend/llamacpp.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AACH,OAAO,EAAE,MAAM,SAAS,CAAC;AACzB,OAAO,IAAI,MAAM,WAAW,CAAC;AAC7B,OAAO,EAAE,MAAM,SAAS,CAAC;AACzB,OAAO,EAAE,UAAU,EAAE,MAAM,kBAAkB,CAAC;AAC9C,OAAO,EACL,OAAO,EACP,MAAM,EACN,YAAY,GAIb,MAAM,YAAY,CAAC;AAwCpB,MAAM,OAAO,eAAgB,SAAQ,OAAO;IACjC,SAAS,CAAS;IAClB,WAAW,CAAS;IACpB,SAAS,CAAS;IAClB,IAAI,CAAS;IACb,UAAU,CAAS;IAEpB,KAAK,GAAyB,IAAI,CAAC;IACnC,KAAK,GAAsB,IAAI,CAAC;IAChC,OAAO,GAAwB,IAAI,CAAC;IACpC,OAAO,GAA4B,IAAI,CAAC;IACxC,SAAS,GAAkB,IAAI,CAAC;IAExC,YAAY,IAAqB;QAC/B,KAAK,EAAE,CAAC;QACR,IAAI,CAAC,SAAS,GAAG,IAAI,CAAC,SAAS,CAAC;QAChC,IAAI,CAAC,WAAW,GAAG,IAAI,CAAC,WAAW,IAAI,GAAG,CAAC;QAC3C,IAAI,CAAC,SAAS,GAAG,IAAI,CAAC,SAAS,IAAI,IAAI,CAAC;QACxC,IAAI,CAAC,IAAI,GAAG,IAAI,CAAC,IAAI,IAAI,KAAK,CAAC;QAC/B,IAAI,CAAC,UAAU,GAAG,IAAI,CAAC,UAAU,IAAI,CAAC,CAAC,CAAC;QACxC,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,KAAK,IAAI,EAAE,CAAC;IAChC,CAAC;IAED,IAAI,IAAI;QACN,OAAO,UAAU,CAAC;IACpB,CAAC;IAEQ,KAAK,CAAC,IAAI;QACjB,IAAI,IAAI,CAAC,OAAO;YAAE,OAAO,IAAI,CAAC;QAC9B,IAAI,CAAC,IAAI,CAAC,SAAS,IAAI,CAAC,EAAE,CAAC,UAAU,CAAC,IAAI,CAAC,SAAS,CAAC,EAAE,CAAC;YACtD,IAAI,CAAC,SAAS,GAAG,2BAA2B,IAAI,CAAC,SAAS,EAAE,CAAC;YAC7D,OAAO,IAAI,CAAC,SAAS,CAAC;QACxB,CAAC;QAED,IAAI,CAAC;YACH,2EAA2E;YAC3E,iEAAiE;YACjE,MAAM,GAAG,GAAG,CAAC,MAAM,MAAM,CAAC,gBAAgB,CAAC,CAG1C,CAAC;YAEF,IAAI,CAAC,KAAK,GAAG,MAAM,GAAG,CAAC,QAAQ,EAAE,CAAC;YAClC,IAAI,CAAC,KAAK,GAAG,MAAM,IAAI,CAAC,KAAK,CAAC,SAAS,CAAC;gBACtC,SAAS,EAAE,IAAI,CAAC,SAAS;gBACzB,SAAS,EAAE,IAAI,CAAC,UAAU;aAC3B,CAAC,CAAC;YACH,IAAI,CAAC,OAAO,GAAG,MAAM,IAAI,CAAC,KAAK,CAAC,aAAa,CAAC;gBAC5C,WAAW,EAAE,IAAI,CAAC,IAAI;gBACtB,cAAc,EAAE,IAAI;gBACpB,SAAS,EAAE,GAAG;gBACd,OAAO,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI,EAAE,CAAC,MAAM,GAAG,CAAC,CAAC;aAC3C,CAAC,CAAC;YACH,IAAI,CAAC,OAAO,GAAG,IAAI,GAAG,CAAC,gBAAgB,CAAC,EAAE,eAAe,EAAE,IAAI,CAAC,OAAO,CAAC,WAAW,EAAE,EAAE,CAAC,CAAC;YACzF,IAAI,CAAC,SAAS,GAAG,IAAI,CAAC;YACtB,OAAO,IAAI,CAAC;QACd,CAAC;QAAC,OAAO,CAAC,EAAE,CAAC;YACX,MAAM,GAAG,GAAG,MAAM,CAAC,CAAC,CAAC,CAAC;YACtB,yEAAyE;YACzE,IAAI,CAAC,SAAS,GAAG,GAAG,CAAC,QAAQ,CAAC,qBAAqB,CAAC,IAAI,GAAG,CAAC,QAAQ,CAAC,sBAAsB,CAAC;gBAC1F,CAAC,CAAC,oEAAoE;gBACtE,CAAC,CAAC,iCAAiC,GAAG,EAAE,CAAC;YAC3C,OAAO,IAAI,CAAC,SAAS,CAAC;QACxB,CAAC;IACH,CAAC;IAEQ,QAAQ;QACf,OAAO,IAAI,CAAC,OAAO,KAAK,IAAI,CAAC;IAC/B,CAAC;IAEQ,MAAM;QACb,KAAK,IAAI,CAAC,UAAU,EAAE,CAAC;IACzB,CAAC;IAEO,KAAK,CAAC,UAAU;QACtB,IAAI,CAAC;YACH,MAAM,IAAI,CAAC,OAAO,EAAE,WAAW,CAAC;QAClC,CAAC;QAAC,MAAM,CAAC;YACP,YAAY;QACd,CAAC;QACD,IAAI,CAAC,OAAO,GAAG,IAAI,CAAC;QACpB,IAAI,CAAC;YACH,MAAM,IAAI,CAAC,OAAO,EAAE,OAAO,EAAE,EAAE,CAAC;QAClC,CAAC;QAAC,MAAM,CAAC;YACP,YAAY;QACd,CAAC;QACD,IAAI,CAAC,OAAO,GAAG,IAAI,CAAC;QACpB,IAAI,CAAC;YACH,MAAM,IAAI,CAAC,KAAK,EAAE,OAAO,EAAE,EAAE,CAAC;QAChC,CAAC;QAAC,MAAM,CAAC;YACP,YAAY;QACd,CAAC;QACD,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC;QAClB,IAAI,CAAC;YACH,MAAM,IAAI,CAAC,KAAK,EAAE,OAAO,EAAE,EAAE,CAAC;QAChC,CAAC;QAAC,MAAM,CAAC;YACP,YAAY;QACd,CAAC;QACD,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC;IACpB,CAAC;IAED,sEAAsE;IAC9D,aAAa,CAAC,QAAuB;QAC3C,MAAM,OAAO,GAAa,EAAE,CAAC;QAC7B,MAAM,KAAK,GAAkB,EAAE,CAAC;QAChC,KAAK,MAAM,CAAC,IAAI,QAAQ,EAAE,CAAC;YACzB,IAAI,CAAC,CAAC,IAAI,KAAK,QAAQ;gBAAE,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC;;gBAC5C,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QACrB,CAAC;QACD,OAAO,EAAE,MAAM,EAAE,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,EAAE,KAAK,EAAE,CAAC;IACjD,CAAC;IAED;;;;;OAKG;IACK,WAAW,CAAC,QAAuB;QACzC,MAAM,EAAE,MAAM,EAAE,KAAK,EAAE,GAAG,IAAI,CAAC,aAAa,CAAC,QAAQ,CAAC,CAAC;QACvD,MAAM,QAAQ,GAAG,CAAC,GAAG,KAAK,CAAC,CAAC,OAAO,EAAE,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,MAAM,CAAC,CAAC;QACrE,MAAM,YAAY,GAAG,KAAK;aACvB,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;aACZ,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,GAAG,CAAC,CAAC,IAAI,KAAK,CAAC,CAAC,OAAO,EAAE,CAAC;aACrC,IAAI,CAAC,IAAI,CAAC,CAAC;QAEd,MAAM,KAAK,GAAa,EAAE,CAAC;QAC3B,IAAI,MAAM;YAAE,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QAC/B,IAAI,YAAY;YAAE,KAAK,CAAC,IAAI,CAAC,4BAA4B,YAAY,EAAE,CAAC,CAAC;QACzE,KAAK,CAAC,IAAI,CAAC,QAAQ,EAAE,OAAO,IAAI,EAAE,CAAC,CAAC;QACpC,OAAO,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;IAC5B,CAAC;IAED,KAAK,CAAC,QAAQ,CAAC,QAAuB,EAAE,SAAkB;QACxD,IAAI,CAAC,IAAI,CAAC,OAAO,EAAE,CAAC;YAClB,MAAM,GAAG,GAAG,MAAM,IAAI,CAAC,IAAI,EAAE,CAAC;YAC9B,IAAI,GAAG,KAAK,IAAI;gBAAE,OAAO,YAAY,CAAC,GAAG,CAAC,CAAC;QAC7C,CAAC;QACD,IAAI,CAAC;YACH,MAAM,KAAK,GAAG,SAAS,IAAI,IAAI,CAAC,SAAS,CAAC;YAC1C,MAAM,MAAM,GAAG,IAAI,CAAC,WAAW,CAAC,QAAQ,CAAC,CAAC;YAC1C,MAAM,OAAO,GAAG,MAAM,IAAI,CAAC,OAAQ,CAAC,MAAM,CAAC,MAAM,EAAE;gBACjD,SAAS,EAAE,KAAK;gBAChB,WAAW,EAAE,IAAI,CAAC,WAAW;aAC9B,CAAC,CAAC;YACH,OAAO,YAAY,CAAC,OAAO,CAAC,CAAC;QAC/B,CAAC;QAAC,OAAO,CAAC,EAAE,CAAC;YACX,OAAO,YAAY,CAAC,qBAAqB,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;QACxD,CAAC;IACH,CAAC;IAED,KAAK,CAAC,CAAC,UAAU,CACf,QAAuB,EACvB,MAAoB,EACpB,SAAkB;QAElB,IAAI,CAAC,IAAI,CAAC,OAAO,EAAE,CAAC;YAClB,MAAM,GAAG,GAAG,MAAM,IAAI,CAAC,IAAI,EAAE,CAAC;YAC9B,IAAI,GAAG,KAAK,IAAI,EAAE,CAAC;gBACjB,MAAM,GAAG,CAAC;gBACV,OAAO;YACT,CAAC;QACH,CAAC;QAED,MAAM,KAAK,GAAG,IAAI,UAAU,EAAU,CAAC;QACvC,MAAM,MAAM,GAAG,IAAI,CAAC,WAAW,CAAC,QAAQ,CAAC,CAAC;QAE1C,KAAK,CAAC,KAAK,IAAI,EAAE;YACf,IAAI,CAAC;gBACH,MAAM,IAAI,CAAC,OAAQ,CAAC,MAAM,CAAC,MAAM,EAAE;oBACjC,SAAS,EAAE,SAAS,IAAI,IAAI,CAAC,SAAS;oBACtC,WAAW,EAAE,IAAI,CAAC,WAAW;oBAC7B,MAAM;oBACN,WAAW,EAAE,CAAC,KAAa,EAAE,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC;iBAClD,CAAC,CAAC;gBACH,KAAK,CAAC,KAAK,EAAE,CAAC;YAChB,CAAC;YAAC,OAAO,CAAC,EAAE,CAAC;gBACX,KAAK,CAAC,IAAI,CAAC,IAAI,KAAK,CAAC,4BAA4B,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;YACjE,CAAC;QACH,CAAC,CAAC,EAAE,CAAC;QAEL,IAAI,KAAK,EAAE,MAAM,KAAK,IAAI,KAAK,EAAE,CAAC;YAChC,MAAM,KAAK,CAAC;QACd,CAAC;IACH,CAAC;IAED,KAAK,CAAC,WAAW;QACf,MAAM,MAAM,GAAG,IAAI,CAAC,QAAQ,EAAE,CAAC;QAC/B,OAAO;YACL,MAAM,EAAE,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,YAAY;YACpC,OAAO,EAAE,UAAU;YACnB,KAAK,EAAE,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,MAAM;YAC9D,MAAM;YACN,GAAG,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,KAAK,EAAE,IAAI,CAAC,SAAS,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;SACrD,CAAC;IACJ,CAAC;CACF"}

package/dist/core/backend/lmstudio.d.ts

@@ -0,0 +1,53 @@
+import { Backend, type BackendHealth, type ChatMessage, type CompletionResult } from "./types.js";
+interface LMModel {
+    id?: string;
+    model_id?: string;
+    state?: string;
+    is_loaded?: boolean;
+    [key: string]: unknown;
+}
+export interface LMStudioOptions {
+    baseUrl?: string;
+    modelName?: string;
+    apiKey?: string;
+    temperature?: number;
+    maxTokens?: number;
+    autoLoad?: boolean;
+    autoStart?: boolean;
+    lmsBin?: string;
+    tools?: unknown[];
+}
+export declare class LMStudioBackend extends Backend {
+    readonly baseUrl: string;
+    modelName: string;
+    readonly apiKey: string;
+    readonly temperature: number;
+    readonly maxTokens: number;
+    readonly autoLoad: boolean;
+    readonly autoStart: boolean;
+    readonly lmsBin: string;
+    private loadedModel;
+    private proc;
+    constructor(opts?: LMStudioOptions);
+    get name(): string;
+    private headers;
+    /** Start the LM Studio server via the `lms` CLI when it is not running. */
+    ensureServer(): Promise<void>;
+    private tryStartServer;
+    private nativeRequest;
+    /** List available models from LM Studio (native, then OpenAI-compatible). */
+    discoverModels(): Promise<LMModel[]>;
+    /** Return the currently loaded model, or null. */
+    getLoadedModel(): Promise<LMModel | null>;
+    load(): Promise<string>;
+    isLoaded(): boolean;
+    unload(): void;
+    private resolveModel;
+    private nativeComplete;
+    private openaiComplete;
+    complete(messages: ChatMessage[], maxTokens?: number): Promise<CompletionResult>;
+    streamText(messages: ChatMessage[], signal?: AbortSignal, maxTokens?: number): AsyncGenerator<string>;
+    healthCheck(): Promise<BackendHealth>;
+}
+export {};
+//# sourceMappingURL=lmstudio.d.ts.map

package/dist/core/backend/lmstudio.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"lmstudio.d.ts","sourceRoot":"","sources":["../../../src/core/backend/lmstudio.ts"],"names":[],"mappings":"AAQA,OAAO,EACL,OAAO,EAOP,KAAK,aAAa,EAClB,KAAK,WAAW,EAChB,KAAK,gBAAgB,EACtB,MAAM,YAAY,CAAC;AAIpB,UAAU,OAAO;IACf,EAAE,CAAC,EAAE,MAAM,CAAC;IACZ,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,CAAC,GAAG,EAAE,MAAM,GAAG,OAAO,CAAC;CACxB;AAED,MAAM,WAAW,eAAe;IAC9B,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,QAAQ,CAAC,EAAE,OAAO,CAAC;IACnB,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,KAAK,CAAC,EAAE,OAAO,EAAE,CAAC;CACnB;AAED,qBAAa,eAAgB,SAAQ,OAAO;IAC1C,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;IACzB,SAAS,EAAE,MAAM,CAAC;IAClB,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC;IACxB,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;IAC7B,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;IAC3B,QAAQ,CAAC,QAAQ,EAAE,OAAO,CAAC;IAC3B,QAAQ,CAAC,SAAS,EAAE,OAAO,CAAC;IAC5B,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAC;IAExB,OAAO,CAAC,WAAW,CAAwB;IAC3C,OAAO,CAAC,IAAI,CAAyC;IAErD,YAAY,IAAI,GAAE,eAAoB,EAWrC;IAED,IAAI,IAAI,IAAI,MAAM,CAEjB;IAED,OAAO,CAAC,OAAO;IAOf,2EAA2E;IACrE,YAAY,IAAI,OAAO,CAAC,IAAI,CAAC,CAKlC;YAEa,cAAc;YAyBd,aAAa;IAkB3B,6EAA6E;IACvE,cAAc,IAAI,OAAO,CAAC,OAAO,EAAE,CAAC,CAiBzC;IAED,kDAAkD;IAC5C,cAAc,IAAI,OAAO,CAAC,OAAO,GAAG,IAAI,CAAC,CAG9C;IAEc,IAAI,IAAI,OAAO,CAAC,MAAM,CAAC,CAkCrC;IAEQ,QAAQ,IAAI,OAAO,CAE3B;IAEQ,MAAM,IAAI,IAAI,CAUtB;YAEa,YAAY;YAUZ,cAAc;YAmBd,cAAc;IAkBtB,QAAQ,CAAC,QAAQ,EAAE,WAAW,EAAE,EAAE,SAAS,CAAC,EAAE,MAAM,GAAG,OAAO,CAAC,gBAAgB,CAAC,CAWrF;IAEM,UAAU,CACf,QAAQ,EAAE,WAAW,EAAE,EACvB,MAAM,CAAC,EAAE,WAAW,EACpB,SAAS,CAAC,EAAE,MAAM,GACjB,cAAc,CAAC,MAAM,CAAC,CAuBxB;IAEK,WAAW,IAAI,OAAO,CAAC,aAAa,CAAC,CA6B1C;CACF"}