u-claw/portable/lib/setup-local-model.mjs

// setup-local-model.mjs — set up a local or self-hosted model from the command line.
//
// For offline and corporate-network situations: ask a few questions, write
// openclaw.json directly, then actually test that the model answers. Deliberately
// does not depend on the dashboard, which is the thing that tends to be broken
// when someone reaches for this tool.
//
// Two kinds of endpoint:
//   1) Ollama on this machine (http://127.0.0.1:11434)
//   2) Any OpenAI-compatible endpoint — a company server, a relay (URL + token)
//
// Only the model-related fields are merged in; gateway and other settings are
// left alone, and the previous file is backed up first.
//
// Usage: node setup-local-model.mjs <CONFIG_PATH>
//
// Input: a real terminal gets interactive readline; piped input (tests, scripts)
// is read once and dequeued line by line.

import { readFileSync, writeFileSync, existsSync, copyFileSync, mkdirSync } from 'node:fs';
import { dirname } from 'node:path';
import { createInterface } from 'node:readline/promises';
import { stdin as input, stdout as output } from 'node:process';
import http from 'node:http';
import https from 'node:https';

function makePrompter() {
  if (input.isTTY) {
    const rl = createInterface({ input, output });
    return {
      ask: async (q, def) => {
        const a = (await rl.question(`${q}${def ? ` [${def}]` : ''}: `)).trim();
        return a || def || '';
      },
      close: () => rl.close(),
    };
  }
  let queued = [];
  try { queued = readFileSync(0, 'utf8').split(/\r?\n/); } catch {}
  let i = 0;
  return {
    ask: async (q, def) => {
      const raw = (queued[i++] ?? '').trim();
      const val = raw || def || '';
      output.write(`${q}${def ? ` [${def}]` : ''}: ${val}\n`);
      return val;
    },
    close: () => {},
  };
}

const CHAT_TIMEOUT_MS = 30000;
function line(s = '') { output.write(s + '\n'); }
function withScheme(raw) { return /^[a-z][a-z0-9+.-]*:\/\//i.test(raw) ? raw : `http://${raw}`; }
function hostOf(raw) {
  try { return new URL(withScheme(String(raw).trim())).hostname || null; } catch { return null; }
}

function applyNoProxyFor(baseUrl) {
  const host = hostOf(baseUrl);
  const existing = (process.env.NO_PROXY || process.env.no_proxy || '')
    .split(',')
    .map((s) => s.trim())
    .filter(Boolean);
  const merged = Array.from(new Set([...existing, 'localhost', '127.0.0.1', '::1', ...(host ? [host] : [])]));
  process.env.NO_PROXY = merged.join(',');
  process.env.no_proxy = process.env.NO_PROXY;
}

async function chatTest(baseUrl, apiKey, modelId) {
  applyNoProxyFor(baseUrl);
  const url = withScheme(baseUrl).replace(/\/+$/, '') + '/chat/completions';
  const started = Date.now();
  try {
    const res = await requestText(url, {
      method: 'POST',
      headers: { 'Content-Type': 'application/json', ...(apiKey ? { Authorization: `Bearer ${apiKey}` } : {}) },
      body: JSON.stringify({ model: modelId, messages: [{ role: 'user', content: 'Reply with exactly: connection ok' }], max_tokens: 64, stream: false }),
      timeoutMs: CHAT_TIMEOUT_MS,
    });
    const ms = Date.now() - started;
    const text = res.body;
    if (!res.ok) return { ok: false, status: res.status, body: text.slice(0, 300), ms };
    let reply = '';
    try { const j = JSON.parse(text); reply = j?.choices?.[0]?.message?.content ?? j?.choices?.[0]?.text ?? ''; }
    catch { reply = text.slice(0, 200); }
    return { ok: true, reply: String(reply).trim(), ms };
  } catch (err) {
    const ms = Date.now() - started;
    if (err?.code === 'ETIMEDOUT') return { ok: false, error: `ETIMEDOUT(>${CHAT_TIMEOUT_MS / 1000}s)`, ms };
    return { ok: false, error: `${err?.cause?.code || err?.code || err?.name || 'ERR'}: ${err?.message || ''}`, ms };
  }
}

function requestText(rawUrl, { method = 'GET', headers = {}, body, timeoutMs }) {
  return new Promise((resolve, reject) => {
    const u = new URL(rawUrl);
    const client = u.protocol === 'https:' ? https : http;
    const req = client.request(u, { method, headers }, (res) => {
      const chunks = [];
      res.on('data', (chunk) => chunks.push(chunk));
      res.on('end', () => {
        const text = Buffer.concat(chunks).toString('utf8');
        resolve({ ok: res.statusCode >= 200 && res.statusCode < 300, status: res.statusCode, body: text });
      });
    });
    req.setTimeout(timeoutMs, () => {
      const err = new Error('request timed out');
      err.code = 'ETIMEDOUT';
      req.destroy(err);
    });
    req.on('error', reject);
    if (body) req.write(body);
    req.end();
  });
}

async function main() {
  const configPath = process.argv[2] || process.env.OPENCLAW_CONFIG_PATH;
  if (!configPath) { line('Usage: node setup-local-model.mjs <CONFIG_PATH>'); process.exitCode = 2; return; }

  line('========================================');
  line('  U-Claw — set up a local or self-hosted model');
  line('========================================');
  line('');

  const rl = makePrompter();
  const ask = rl.ask;

  try {
    line('Which kind of model?');
    line('  1) Ollama running on this machine (http://127.0.0.1:11434)');
    line('  2) A self-hosted or company endpoint (any OpenAI-compatible URL + token)');
    const kind = await ask('Enter 1 or 2', '1');

    let providerKey, baseUrl, apiKey, modelId;
    if (kind === '2') {
      providerKey = 'newapi';
      line('');
      line('The endpoint usually looks like http://192.168.1.50:3000/v1 — most need the /v1 on the end.');
      baseUrl = await ask('Endpoint URL', 'http://192.168.1.50:3000/v1');
      apiKey = await ask('token / API Key', '');
      modelId = await ask('Model ID — your admin will have given you this, e.g. deepseek-v3', '');
    } else {
      providerKey = 'ollama';
      line('');
      baseUrl = await ask('Ollama URL — the default is right unless you changed it', 'http://127.0.0.1:11434/v1');
      // Ollama's OpenAI-compatible endpoint lives at /v1; add it if missing
      if (!/\/v1\/?$/.test(baseUrl)) baseUrl = baseUrl.replace(/\/+$/, '') + '/v1';
      apiKey = 'ollama'; // Ollama ignores auth; any placeholder works
      modelId = await ask('Model name — run `ollama list` to see yours, e.g. qwen2.5', 'qwen2.5');
    }

    if (!baseUrl || !modelId) { line(''); line('No endpoint or model ID given — nothing was changed.'); process.exitCode = 2; return; }

    // Merge into the existing config so gateway and other settings survive; back up first
    let config = {};
    if (existsSync(configPath)) {
      try { config = JSON.parse(readFileSync(configPath, 'utf8')); }
      catch { config = {}; }
      try { copyFileSync(configPath, configPath + '.bak'); } catch {}
    } else {
      try { mkdirSync(dirname(configPath), { recursive: true }); } catch {}
    }

    config.gateway ||= { mode: 'local', auth: { token: 'uclaw' } };
    config.models ||= {};
    config.models.mode = 'merge';
    config.models.providers ||= {};
    config.models.providers[providerKey] = {
      baseUrl,
      apiKey,
      api: 'openai-completions',
      models: [{ id: modelId, name: modelId }],
    };
    config.agents ||= {};
    config.agents.defaults ||= {};
    config.agents.defaults.model ||= {};
    config.agents.defaults.model.primary = `${providerKey}/${modelId}`;

    writeFileSync(configPath, JSON.stringify(config, null, 2), 'utf8');
    line('');
    line(`✓ Saved to ${configPath}`);
    line(`  provider=${providerKey}  baseUrl=${baseUrl}  model=${modelId}`);
    if (existsSync(configPath + '.bak')) line('  Your previous settings were kept as openclaw.json.bak');
    line('');

    // Prove it actually works rather than just claiming it was saved
    line('Testing it — sending one message to the model…');
    const r = await chatTest(baseUrl, apiKey, modelId);
    line('');
    if (r.ok) {
      line(`✓ It works. The model replied in ${r.ms}ms: ${r.reply.slice(0, 120) || '(empty reply, but the request succeeded)'}`);
      line('');
      line('You are set. Start U-Claw normally and it will use this model.');
    } else if (r.status) {
      line(`✗ The server answered with HTTP ${r.status} in ${r.ms}ms: ${r.body}`);
      line('');
      if (r.status === 401 || r.status === 403) line('  The network is fine but the token was rejected. Ollama ignores auth entirely; for a company endpoint, check the token with whoever issued it.');
      else if (r.status === 404) line('  The network is fine but the path or model ID is wrong. Check whether the endpoint needs /v1 on the end, and that the model name matches exactly.');
      else line('  The network is fine; the server itself errored. Send the text above to whoever runs it.');
    } else {
      const host = (() => { try { return new URL(withScheme(baseUrl)).hostname; } catch { return baseUrl; } })();
      line(`✗ Could not reach it: ${r.error} (${r.ms}ms)`);
      line('');
      if (providerKey === 'ollama') {
        line('  Ollama is not answering. It is usually not running, or the model was never pulled. On this machine, run:');
        line('    ollama serve                  starts it, skip if it already runs in the background');
        line(`    ollama pull ${modelId}        downloads the model — do this while you still have internet`);
        line('  Then run this tool again.');
      } else {
        line('  Either the address is wrong, the network cannot reach it, a firewall is blocking it, or the service is down.');
        line('  Test it from this machine with:');
        line(`    ping ${host}`);
        line('  If that also fails, ask whoever runs the server to confirm the address, port and firewall rules.');
      }
    }
  } finally {
    rl.close();
  }
}

main();