c9ai
Version:
Universal AI assistant with vibe-based workflows, hybrid cloud+local AI, and comprehensive tool integration
189 lines (176 loc) ā¢ 6.77 kB
JavaScript
;
// Cross-platform launcher:
// 1) Find llama-server and a .gguf model
// 2) Start server on first free port (8080..8090) if not already running
// 3) Wait for /v1/models
// 4) Launch Agent API
// 5) Open llama.cpp UI in browser
const os = require("node:os");
const fs = require("node:fs");
const path = require("node:path");
const http = require("node:http");
const cp = require("node:child_process");
function which(bin) {
try {
const cmd = process.platform === "win32" ? `where ${bin}` : `command -v ${bin}`;
const out = cp.execSync(cmd, { stdio: ["ignore","pipe","ignore"] }).toString().trim();
return out.split(/\r?\n/)[0] || null;
} catch { return null; }
}
function findModelCandidates() {
const home = os.homedir();
const roots = [
path.join(home, ".c9ai", "models"),
path.join(process.cwd(), "models"),
process.cwd()
];
const list = [];
for (const dir of roots) {
try {
for (const e of fs.readdirSync(dir)) if (e.endsWith(".gguf")) list.push(path.join(dir, e));
} catch {}
}
return list;
}
async function waitHttp(url, { timeoutMs = 15000, intervalMs = 600 } = {}) {
const deadline = Date.now() + timeoutMs;
return new Promise((resolve, reject) => {
const tick = () => {
const req = http.get(url, (res) => {
const ok = res.statusCode >= 200 && res.statusCode < 400;
res.resume();
if (ok) return resolve(true);
if (Date.now() > deadline) return reject(new Error(`HTTP ${res.statusCode}`));
setTimeout(tick, intervalMs);
});
req.on("error", () => {
if (Date.now() > deadline) return reject(new Error("fetch failed"));
setTimeout(tick, intervalMs);
});
req.setTimeout(3000, () => { req.destroy(); });
};
tick();
});
}
function pickPort(start = 8080, max = 8090) {
return new Promise((resolve) => {
let port = start;
const tryPort = () => {
const srv = http.createServer();
srv.once("error", () => { port++; if (port > max) resolve(null); else tryPort(); });
srv.listen(port, "127.0.0.1", () => srv.close(() => resolve(port)));
};
tryPort();
});
}
function openBrowser(url) {
try {
if (process.platform === "darwin") {
cp.spawn("open", [url], { stdio: "ignore", detached: true });
} else if (process.platform === "win32") {
cp.spawn("cmd", ["/c", "start", "", url], { stdio: "ignore", detached: true });
} else {
const opener = which("xdg-open") || which("gio") || which("gnome-open") || which("kde-open");
if (opener) cp.spawn(opener, [url], { stdio: "ignore", detached: true });
}
} catch {}
}
(async () => {
// 1) llama-server binary
const llamaBin = which("llama-server") || which("llama-server.exe") || which("server");
if (!llamaBin) {
console.error("ā llama-server not found in PATH.");
console.error(" macOS: brew install llama.cpp");
console.error(" Linux: build from source (see repo)");
console.error(" Win: download/build llama.cpp, add folder to PATH");
process.exit(1);
}
// 2) model
const models = findModelCandidates();
if (models.length === 0) {
console.error("ā No .gguf models found in ~/.c9ai/models, ./models, or CWD.");
process.exit(1);
}
const model = process.env.LLAMACPP_MODEL || models[0];
// 3) pick port / start llama.cpp if needed
let baseUrl = process.env.LLAMACPP_BASE_URL || "http://127.0.0.1:8080";
let healthy = false;
try { healthy = await waitHttp(baseUrl + "/v1/models", { timeoutMs: 1200 }); } catch {}
let portUsed = Number(new URL(baseUrl).port || 8080);
if (!healthy) {
const port = await pickPort(8080, 8090);
if (!port) {
console.error("ā No free port in 8080..8090.");
process.exit(1);
}
portUsed = port;
baseUrl = `http://127.0.0.1:${portUsed}`;
console.log(`š Starting llama-server on ${baseUrl}`);
// Retry ladder: try fast settings first.
// IMPORTANT: We DO NOT fall back to CPU unless FORCE_CPU=true
const ctxEnv = process.env.LLAMACPP_CTX;
const nglEnv = process.env.LLAMACPP_NGL;
const forceCPU = String(process.env.FORCE_CPU || "").toLowerCase() === "true";
const tries = forceCPU
? [
{ c: ctxEnv || "2048", ngl: "0", t: "8", tb: "8", ub: "128" }, // CPU safe
{ c: "1536", ngl: "0", t: "8", tb: "8", ub: "128" }
]
: [
{ c: ctxEnv || "4096", ngl: nglEnv || "20", t: "8", tb: "8", ub: "128" }, // fast, GPU
{ c: ctxEnv || "3072", ngl: "12", t: "8", tb: "8", ub: "128" }, // less GPU
{ c: "3072", ngl: "10", t: "8", tb: "8", ub: "128" } // small step-down
];
let started = false;
for (const attempt of tries) {
console.log(`ā trying: -c ${attempt.c} -ngl ${attempt.ngl} -t ${attempt.t} -tb ${attempt.tb}`);
const args = [
"-m", model,
"-c", String(attempt.c),
"-ngl", String(attempt.ngl),
"-t", String(attempt.t),
"-tb", String(attempt.tb),
"--host", "127.0.0.1",
"--port", String(portUsed)
];
const child = cp.spawn(llamaBin, args, { stdio: "inherit" });
try {
await waitHttp(baseUrl + "/v1/models", { timeoutMs: 20000 });
console.log("ā
llama.cpp API healthy.");
started = true;
child.on("exit", (code) => {
console.error(`llama-server exited with code ${code}`);
process.exit(code || 1);
});
break;
} catch (e) {
console.warn(`ā ļø failed at -c ${attempt.c} -ngl ${attempt.ngl}: ${e.message}`);
try { child.kill("SIGINT"); } catch {}
}
}
if (!started) {
console.error("ā Could not start llama-server with GPU settings.");
if (!forceCPU) {
console.error(" (Refusing to fall back to CPU unless FORCE_CPU=true)");
}
console.error(" Try: export LLAMACPP_CTX=3072; export LLAMACPP_NGL=12; npm run start:local");
process.exit(1);
}
} else {
console.log(`ā
llama.cpp already healthy at ${baseUrl}`);
}
// start Agent API server (8787) and pass llama base URL
console.log("š Launching Agent API (port 8787) ā¦");
const agentEnv = { ...process.env, LLAMACPP_BASE_URL: baseUrl };
const api = cp.spawn(process.execPath, ["server/agent-api.js"], { stdio: "inherit", env: agentEnv });
api.on("exit", (code) => {
console.error(`Agent API exited with code ${code}`);
process.exit(code || 1);
});
// 5) open UI(s)
console.log("\nš§© Ready!");
console.log(` llama.cpp API: ${baseUrl}`);
console.log(" c9ai UI: http://127.0.0.1:8787");
openBrowser("http://127.0.0.1:8787");
})();