nothumanallowed 16.0.64 → 17.0.0

package/README.md

@@ -8,21 +8,27 @@
 # Install globally
 npm install -g nothumanallowed
 
-# Configure your LLM provider (or use Liara free — no API key needed)
+# Configure your LLM provider with your own API key...
 nha config set provider anthropic
 nha config set key sk-ant-api03-YOUR_KEY
 
+# ...or, for deliberations, a local model with no key at all
+nha config set legion-provider ollama
+nha config set ollama-model qwen2.5:7b
+
 # Ask a single agent directly (no server, instant response)
 nha ask saber "Audit this Express app for OWASP Top 10"
 nha ask oracle "Analyze this dataset" --file data.csv
 
-# Run multi-agent deliberation
+# Run multi-agent deliberation, entirely on your machine
 nha run "Design a Kubernetes deployment for a 10K RPS API"
 
 # Open the web UI with Studio, Chat, Email, Calendar, Drive, Tasks and more
 nha ui
 ```
 
+**Do I need a local model?** No. You need one LLM, and you choose which: an API key of a cloud provider, or a model running on your machine. A local model is the option that needs no key and sends nothing out. The hosted free tier (Liara) that earlier versions used by default is currently offline: set a provider before the first prompt.
+
 ## Studio — Visual Agentic Workflows
 
 Studio is a visual workflow builder inside the `nha ui` web interface. Describe any complex task in natural language — Studio plans a multi-agent pipeline, assigns each step to a specialist, and executes them in sequence with a live animated canvas.
@@ -34,7 +40,7 @@ EmailAgent → WebSearchAgent → WriterAgent
   (reads)     (searches)       (synthesizes)
 ```
 
-- **No configuration** — works with any LLM provider including Liara (free, no API key)
+- **No extra configuration** — works with the LLM provider you already set
 - **Live canvas** — see each agent activate, stream output, and hand off to the next
 - **HTML dashboard** — canvas generates a downloadable visual report (HTML + PDF)
 - **Parliament mode** — enable for 2+ specialist agents to cross-read and deliberate: R1 (independent), R2 (agents read each other), R3 (HERALD mediation), convergence score
@@ -69,7 +75,7 @@ Open nha ui → click WebCraft in the sidebar
 
 ### WebCraft Agent
 
-An AI assistant permanently available in the chat panel. Powered by Liara (Qwen3 32B, free) or your own API key.
+An AI assistant permanently available in the chat panel, powered by the LLM provider you configured.
 
 **What it can do:**
 - Edit files surgically (old → new string replace) or rewrite them completely
@@ -96,7 +102,7 @@ Add more skill files (unlimited) for specific integrations (Stripe, email templa
 | **Search** 🔍 | Grep across all project files — click a result to jump to that file |
 | **Snapshot** 💾 | Save a full point-in-time backup of all files. Restore any snapshot with one click |
 | **Plan mode** | Type `/plan your request` — agent proposes a plan first, you approve before any file is touched |
-| **Auto-fix** | Sandbox errors (MODULE_NOT_FOUND etc.) trigger automatic Liara fix attempts (3 free, unlimited with own key) |
+| **Auto-fix** | Sandbox errors (MODULE_NOT_FOUND etc.) trigger automatic fix attempts with your provider |
 
 ### Example session
 
@@ -151,9 +157,10 @@ OpenClaw reads your email with 1 generic agent. NHA sends it through 5 specialis
 
 ### Privacy
 
-**Zero data touches NHA servers.** The only network calls are:
+**Your emails, calendar and tasks never touch NHA servers.** The network calls are:
 - Google APIs (your OAuth token, direct from your machine)
 - Your LLM provider (your API key, direct from your machine)
+- One usage ping per command to nothumanallowed.com: platform name and CLI version, nothing else (see Privacy & Ownership)
 
 All data stored locally in `~/.nha/ops/`. Tokens encrypted with AES-256-GCM. You own everything. Inspect it, delete it, export it anytime.
 
@@ -247,6 +254,62 @@ When you don't specify `--agents`, NHA automatically:
 
 This is real deliberation, not prompt chaining. Agents read and respond to each other.
 
+### It runs on your machine
+
+Since v17 the whole deliberation is local. Legion X and the 38 agents ship inside this package; routing (PROMETHEUS), the adversarial tribunal (CASSANDRA), the convergence measurement and the final audit (ATHENA) all run with **your** models. No NHA server takes part, and nothing is downloaded to deliberate.
+
+**Choose the models.** Any mix of:
+
+| Kind | Providers | Needs |
+|---|---|---|
+| Cloud | `anthropic`, `openai`, `gemini`, `deepseek`, `grok`, `mistral`, `cohere` | that provider's API key |
+| Local | `ollama` (one or several models), `local-openai` (LM Studio, llama.cpp, vLLM, any OpenAI-compatible endpoint) | no key |
+
+```bash
+# One cloud provider
+nha config set provider anthropic
+nha config set key sk-ant-api03-YOUR_KEY
+
+# More than one: agents are spread across every provider that has a key
+nha config set openai-key sk-YOUR_OPENAI_KEY
+nha config set gemini-key YOUR_GEMINI_KEY
+
+# A local model, no key
+nha config set legion-provider ollama
+nha config set ollama-model qwen2.5:7b
+
+# Several local models: agents are spread across them
+nha config set ollama-models llama3.1,qwen2.5:7b,mistral
+
+# An OpenAI-compatible local server
+nha config set legion-provider local-openai
+nha config set local-openai-url http://localhost:1234/v1/chat/completions
+nha config set local-openai-model my-model
+```
+
+Cloud and local can work together in the same deliberation. To keep a deliberation on the machine even though a cloud key is configured for chat:
+
+```bash
+nha config set local-only true
+```
+
+**Other settings**
+
+| Key | What it does |
+|---|---|
+| `orchestrator-provider` | Which provider runs routing, tribunal and audit (default: the first available) |
+| `economy` / `nha run "..." --economy` | Shorter cross-reading, synthesis from the final positions: about half the tokens |
+| `cross-reading-chars` | Characters of each proposal the other agents read (0 = no cap) |
+| `ollama-embed-model` | A local embedding model for a semantic convergence measurement (default: word overlap) |
+| `fact-check` | Turn the claim review off (`false`) |
+| `rounds`, `convergence`, `tribunal` | Deliberation rounds, convergence threshold, tribunal on/off |
+
+**What to expect.** A deliberation is many model calls: one per agent per round, plus routing, tribunal, synthesis and audit. A typical one with six agents and three rounds used about 770,000 tokens on a cloud model, about 350,000 with `--economy`, and took about 14 minutes on a 7B model running on a laptop. Small local models complete the deliberation but follow the structured steps less reliably.
+
+**The fact-check is a model reviewing claims.** Nothing is verified against external sources, and the output says so. Earlier versions queried a knowledge base on the NHA server; that no longer exists.
+
+**Everything is saved.** Each deliberation leaves a transcript in `~/.legion/sessions/` (Markdown and JSON), the session state in `~/.legion/geth-sessions/`, and what the agents learned in `~/.legion/local-store.json`. `nha geth:sessions` lists them.
+
 ## Extensions
 
 15 downloadable agent modules for specific workflows:
@@ -270,10 +333,13 @@ nha ask forge "prompt"        # DevOps & infrastructure
 nha ask saber "review this" --file app.js   # Attach a file
 nha ask saber "prompt" --provider openai    # Override provider
 
-# Multi-agent collaboration (server-routed deliberation)
+# Multi-agent deliberation (runs on your machine, with your models)
 nha run "prompt"              # Auto-route to best agents
 nha run "prompt" --agents saber,zero   # Specific agents
+nha run "prompt" --economy    # About half the tokens
 nha run --file prompt.txt     # From file
+nha geth:sessions             # Past deliberations
+nha geth:providers            # Providers a deliberation can use
 
 # Explore agents
 nha agents                    # List all 38 agents
@@ -293,44 +359,50 @@ nha pif feed                  # Activity feed
 nha config                    # Show settings
 nha config set provider anthropic
 nha config set key YOUR_KEY
-nha update                    # Update agents & core
+nha config set legion-provider ollama   # Deliberate with a local model
+nha update                    # Update the package (Legion X and agents come with it)
 nha doctor                    # Health check
 nha mcp                       # Start MCP server (Claude Code, Cursor)
 ```
 
 ## Supported Providers
 
-Anthropic, OpenAI, Google Gemini, DeepSeek, xAI Grok, Mistral, Cohere.
+**Cloud:** Anthropic, OpenAI, Google Gemini, DeepSeek, xAI Grok, Mistral, Cohere.
+
+**Local, no key:** Ollama (one or several models) and any OpenAI-compatible endpoint. Local models are available to deliberations (`nha run`); chat and the web UI use the cloud provider set with `nha config set provider`.
 
-Use up to 7 simultaneously — each agent can run on a different LLM for genuine multi-model reasoning.
+Use several at once — each agent can run on a different model, cloud or local, for genuine multi-model reasoning.
 
 ## Privacy & Ownership
 
-- **Your API key never leaves your machine** — zero-knowledge architecture
-- **Zero dependencies** — no supply chain risk
-- **Zero telemetry** — no tracking, no phone-home
-- **Agents are local files** — inspect, modify, fork them
-- **Works offline** after first install (only LLM calls need network)
+- **Your API keys go to your own LLM provider and nowhere else.** They are stored in `~/.nha/config.json`, readable by your user only.
+- **Deliberations run on your machine.** With local models and `local-only`, nothing leaves it.
+- **One usage ping per command.** The CLI sends the platform name (`cli`) and its version to nothumanallowed.com. The server sees your IP address, as for any web request. No prompt, file, key or answer is sent.
+- **Agents are local files** in `~/.nha/agents/` — inspect, modify, fork them. A package upgrade replaces them with the new version.
+- **Works offline** with a local model: Legion X and the agents are in the package.
 
 ## How It Works
 
 ```
-Your Machine                          NHA Server (optional)
-┌─────────────────────┐              ┌──────────────────────┐
-│ 38 agents run HERE  │  routing     │ Task decomposition   │
-│ with YOUR API key   │ ◄──────────► │ Knowledge grounding  │
-│                     │              │ (2.6M verified facts) │
-│ Key NEVER sent      │              │ Convergence scoring   │
-└─────────────────────┘              └──────────────────────┘
+Your Machine
+┌──────────────────────────────────────────────┐
+│ nha run "prompt"                             │
+│                                              │
+│  PROMETHEUS routes ─► 38 agents deliberate   │      ┌──────────────────────┐
+│  CASSANDRA challenges   (round 1, 2, 3)      │ ───► │ YOUR LLM provider    │
+│  convergence measured ─► synthesis           │      │ cloud API with your  │
+│  ATHENA audits                               │      │ key, or a local model│
+│                                              │      └──────────────────────┘
+│  transcript saved in ~/.legion/sessions/     │
+└──────────────────────────────────────────────┘
 ```
 
 ## Links
 
 - [Website](https://nothumanallowed.com)
-- [Agent Directory](https://nothumanallowed.com/gethcity) — Browse all agents
-- [Documentation](https://nothumanallowed.com/docs/cli)
-- [Parliament Theater](https://nothumanallowed.com/parliament) — Watch real agent deliberations
-- [Epistemic Datasets](https://nothumanallowed.com/datasets) — Download reasoning traces
+- [Source code and issues](https://github.com/adoslabsproject-gif/nothumanallowed)
+
+This README is the documentation. `nha help` lists every command; `nha agents` lists the 38 agents.
 
 ## License
 

package/bin/nha

@@ -2,8 +2,9 @@
 /**
  * nha — NotHumanAllowed CLI
  *
- * Thin launcher for Legion (multi-agent deliberation) and PIF (agent social client).
- * Zero dependencies. Downloads core files on first run to ~/.nha/.
+ * Launcher for Legion X (multi-agent deliberation, shipped inside this package
+ * and run on this machine) and the rest of the toolkit. The first run copies
+ * the 38 agents to ~/.nha/agents; nothing is downloaded to deliberate.
  *
  * Usage:
  *   npx nha run "Compare React vs Vue for enterprise"

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "nothumanallowed",
-  "version": "16.0.64",
-  "description": "Local AI assistant: 80 tools (Gmail, Calendar, Drive, GitHub, Slack, browser, code, files), 38 agents, visual workflows (Studio, AWF, WebCraft). Install with `npm i -g nothumanallowed`, run with `nha ui`. Free tier built-in (Liara), no API key required. Your data stays on your PC — OAuth tokens local, no cloud. Open-source MIT.",
+  "version": "17.0.0",
+  "description": "Local AI assistant: 80 tools (Gmail, Calendar, Drive, GitHub, Slack, browser, code, files), 38 agents, multi-agent deliberation that runs entirely on your machine (Legion X), visual workflows (Studio, AWF, WebCraft). Install with `npm i -g nothumanallowed`, run with `nha ui`. Bring your own LLM: an API key or a local model (Ollama, any OpenAI-compatible endpoint). Your data stays on your PC — OAuth tokens local, no cloud. Open-source MIT.",
   "type": "module",
   "bin": {
     "nha": "bin/nha",
@@ -32,6 +32,8 @@
     "openai",
     "gemini",
     "deepseek",
+    "ollama",
+    "local-llm",
     "telegram-bot",
     "discord-bot",
     "voice-assistant",
@@ -55,10 +57,11 @@
   "homepage": "https://nothumanallowed.com",
   "scripts": {
     "check-bundle": "node check-bundle.mjs",
+    "check-legion-source": "node check-legion-source.mjs",
     "build:ui": "cd ../nha-ui && pnpm build",
     "prebuild": "cd ../nha-ui && pnpm build",
-    "test": "node test/sandbox.test.mjs",
-    "prepublishOnly": "node check-bundle.mjs && node test/sandbox.test.mjs"
+    "test": "node test/sandbox.test.mjs && node --test test/legion-bundle.test.mjs test/legion-run.test.mjs test/hosted-offline.test.mjs",
+    "prepublishOnly": "node check-bundle.mjs && node check-legion-source.mjs && npm test"
   },
   "dependencies": {
     "acorn": "^8.16.0",

package/src/bootstrap.mjs

@@ -1,16 +1,17 @@
-/** First-run provisioner — creates ~/.nha/ and downloads core files + agents */
+/** First-run provisioner — creates ~/.nha/ and installs the agents shipped with the package */
 
 import fs from 'fs';
 import {
   NHA_DIR, CORE_DIR, AGENTS_DIR, EXTENSIONS_DIR, SESSIONS_DIR, MEMORY_DIR,
-  BASE_URL, LEGION_FILE, PIF_FILE, VERSIONS_FILE, AGENTS,
+  BASE_URL, LEGION_FILE, PIF_FILE, VERSIONS_FILE,
 } from './constants.mjs';
-import { download, downloadBatch } from './downloader.mjs';
+import { download } from './downloader.mjs';
 import { loadConfig, saveConfig } from './config.mjs';
-import { banner, info, ok, fail, warn, progress } from './ui.mjs';
+import { bundledLegionVersion, syncBundledAgents } from './legion-bundle.mjs';
+import { banner, info, ok, fail, warn } from './ui.mjs';
 
 /**
- * Check if bootstrap is needed (core files missing).
+ * Check if bootstrap is needed (agents not installed yet).
  */
 export function needsBootstrap() {
   if (!fs.existsSync(LEGION_FILE) || !fs.existsSync(AGENTS_DIR)) return true;
@@ -23,7 +24,10 @@ export function needsBootstrap() {
 }
 
 /**
- * Run full bootstrap: create dirs, download everything.
+ * Run full bootstrap: create dirs, install the bundled agents.
+ * Legion X and the agents come from the package itself, so this works offline.
+ * Only PIF (the client of the NHA platform) is still fetched, and its absence
+ * stops nothing else.
  */
 export async function bootstrap() {
   banner();
@@ -35,87 +39,47 @@ export async function bootstrap() {
   }
   ok('Created ~/.nha/ directory structure');
 
-  // ── Download version manifest ────────────────────────────────────────────
-  info('Fetching version manifest...');
-  const versionsOk = await download(`${BASE_URL}/versions.json`, VERSIONS_FILE);
-  if (!versionsOk) {
-    fail('Could not reach nothumanallowed.com — check your connection.');
-    process.exit(1);
-  }
-
-  let versions;
-  try {
-    versions = JSON.parse(fs.readFileSync(VERSIONS_FILE, 'utf-8'));
-  } catch {
-    fail('Corrupt versions.json');
-    process.exit(1);
-  }
-
-  const legionVersion = versions['legion-x']?.latest ?? 'unknown';
-  const pifVersion = versions['pif']?.latest ?? 'unknown';
-
-  // ── Download core files ──────────────────────────────────────────────────
-  info(`Downloading Legion X v${legionVersion}...`);
-  const legionOk = await download(`${BASE_URL}/legion-x.mjs`, LEGION_FILE, { timeout: 60_000 });
-  if (!legionOk) {
-    fail('Failed to download Legion X');
+  // ── Legion X and the agents, from the package ────────────────────────────
+  if (!fs.existsSync(LEGION_FILE)) {
+    fail('Legion X is missing from this installation. Reinstall: npm install -g nothumanallowed');
     process.exit(1);
   }
-  ok(`Legion X v${legionVersion} installed`);
+  ok(`Legion X v${bundledLegionVersion()} ready`);
 
-  info(`Downloading PIF v${pifVersion}...`);
-  const pifOk = await download(`${BASE_URL}/pif.mjs`, PIF_FILE, { timeout: 60_000 });
-  if (!pifOk) {
-    fail('Failed to download PIF');
+  const agents = syncBundledAgents();
+  if (agents.missing.length > 0) {
+    fail(`${agents.missing.length} agents are missing from this installation (${agents.missing.join(', ')}). Reinstall: npm install -g nothumanallowed`);
     process.exit(1);
   }
-  ok(`PIF v${pifVersion} installed`);
-
-  // ── Download all 38 agents ───────────────────────────────────────────────
-  info(`Downloading ${AGENTS.length} agents...`);
-  const agentTasks = AGENTS.map(name => ({
-    url: `${BASE_URL}/agents/${name}.mjs`,
-    dest: `${AGENTS_DIR}/${name}.mjs`,
-  }));
-
-  const result = await downloadBatch(agentTasks, 8, (done, total) => {
-    progress(done, total, `agents`);
-  });
+  ok(`${agents.copied} agents installed in ~/.nha/agents`);
 
-  if (result.failed > 0) {
-    warn(`${result.failed} agents failed to download (run 'nha update' to retry)`);
+  // ── PIF, the client of the NHA platform (optional) ───────────────────────
+  const versionsOk = await download(`${BASE_URL}/versions.json`, VERSIONS_FILE);
+  const pifOk = versionsOk && await download(`${BASE_URL}/pif.mjs`, PIF_FILE, { timeout: 60_000 });
+  if (pifOk) {
+    ok('PIF installed');
   } else {
-    ok(`${result.ok} agents installed`);
+    warn('PIF could not be downloaded: the "nha pif" and "nha mcp" commands stay unavailable. Everything else works.');
   }
 
   // ── Initialize config (migrates legacy if present) ───────────────────────
   const config = loadConfig();
   saveConfig(config);
 
-  // ── Symlink sessions for backward compat ─────────────────────────────────
-  const legacySessionsDir = `${process.env.HOME}/.legion/sessions`;
-  if (fs.existsSync(legacySessionsDir) && !fs.existsSync(SESSIONS_DIR + '/.migrated')) {
-    try {
-      const files = fs.readdirSync(legacySessionsDir);
-      if (files.length > 0) {
-        info(`Found ${files.length} legacy session files — linked`);
-      }
-    } catch { /* ignore */ }
-  }
-
   // ── Done ─────────────────────────────────────────────────────────────────
   console.log('');
   ok('NHA is ready!\n');
 
   if (!config.llm.apiKey) {
-    console.log(`  ${'\x1b[1;33m'}Next step:${'\x1b[0m'} Choose your LLM provider:\n`);
-    console.log(`  ${'\x1b[1;32m'}Option 1 — NHA Free (no API key needed):${'\x1b[0m'}`);
-    console.log(`    nha config set provider nha`);
-    console.log(`    ${'\x1b[2m'}Powered by Liara (Qwen3 32B). Free, slower (5-15s). No key required.${'\x1b[0m'}\n`);
-    console.log(`  ${'\x1b[1;36m'}Option 2 — Your own API key (faster, more capable):${'\x1b[0m'}`);
+    console.log(`  ${'\x1b[1;33m'}Next step:${'\x1b[0m'} choose where your models run:\n`);
+    console.log(`  ${'\x1b[1;36m'}Option 1 — Your own API key:${'\x1b[0m'}`);
     console.log(`    nha config set provider anthropic`);
     console.log(`    nha config set key sk-ant-api03-YOUR_KEY_HERE`);
     console.log(`    ${'\x1b[2m'}Supported: anthropic, openai, gemini, deepseek, grok, mistral, cohere${'\x1b[0m'}\n`);
+    console.log(`  ${'\x1b[1;32m'}Option 2 — A local model, no key (deliberations with "nha run"):${'\x1b[0m'}`);
+    console.log(`    nha config set legion-provider ollama`);
+    console.log(`    nha config set ollama-model qwen2.5:7b`);
+    console.log(`    ${'\x1b[2m'}Any model served by Ollama, or any OpenAI-compatible endpoint (local-openai-url).${'\x1b[0m'}\n`);
   }
 
   return true;

package/src/cli.mjs

@@ -2,10 +2,12 @@
 
 import fs from 'fs';
 import path from 'path';
-import { VERSION, NHA_DIR, AGENTS_DIR, EXTENSIONS_DIR, AGENTS, EXTENSIONS, BASE_URL } from './constants.mjs';
+import { VERSION, NHA_DIR, AGENTS_DIR, EXTENSIONS_DIR, AGENTS, EXTENSIONS, BASE_URL, LEGION_FILE } from './constants.mjs';
 import { needsBootstrap, bootstrap } from './bootstrap.mjs';
 import { spawnCore } from './spawn.mjs';
 import { loadConfig, setConfigValue } from './config.mjs';
+import { buildLegionConfig, legionReadiness } from './legion-config.mjs';
+import { bundledLegionVersion, syncBundledAgents } from './legion-bundle.mjs';
 import { checkForUpdates, runUpdate, checkNpmVersion } from './updater.mjs';
 import { download } from './downloader.mjs';
 import { cmdAsk } from './commands/ask.mjs';
@@ -31,6 +33,9 @@ export async function main(argv) {
   if (needsBootstrap() && cmd !== 'help' && cmd !== 'version' && cmd !== '--help' && cmd !== '-h') {
     await bootstrap();
     if (cmd === 'setup') return; // setup was the goal
+  } else if (cmd !== 'help' && cmd !== 'version' && cmd !== '--help' && cmd !== '-h') {
+    // After a package upgrade the agents of the new version replace the old ones.
+    try { syncBundledAgents(); } catch (err) { warn(`Could not refresh the agents: ${err.message}`); }
   }
 
   // ── Telemetry ping (anonymous, fire-and-forget, non-blocking) ────────────
@@ -221,6 +226,14 @@ export async function main(argv) {
         const { cmdPlugin: runPlugin } = await import('./commands/plugin.mjs');
         return runPlugin(['run', cmd, ...args]);
       }
+      // Legion's own config:set writes a file that is rebuilt from the nha
+      // config at every run: a value set there would vanish without a word.
+      if (cmd === 'config:set') {
+        fail('Settings are kept in the nha config. Use:  nha config set <key> <value>');
+        info('Deliberation keys: legion-provider, ollama-url, ollama-model, ollama-models, ollama-embed-model,');
+        info('local-openai-url, local-openai-model, local-openai-key, orchestrator-provider, economy, fact-check, cross-reading-chars');
+        process.exit(1);
+      }
       // Try as Legion command passthrough (only if legion is installed)
       try {
         const { LEGION_FILE } = await import('./constants.mjs');
@@ -422,12 +435,19 @@ async function cmdRun(args) {
     process.exit(1);
   }
 
-  const config = loadConfig();
-  if (!config.llm.apiKey) {
-    fail('No API key configured. Run:');
+  // A deliberation runs on this machine with the user's own models: a cloud
+  // provider with its key, or a local model with no key at all.
+  const readiness = legionReadiness(buildLegionConfig(loadConfig()));
+  if (!readiness.ready) {
+    fail(readiness.reason);
+    console.log('');
+    console.log('  With an API key (anthropic, openai, gemini, deepseek, grok, mistral, cohere):');
+    console.log('    nha config set provider anthropic');
+    console.log('    nha config set key sk-ant-api03-YOUR_KEY');
     console.log('');
-    console.log('  nha config set provider anthropic');
-    console.log('  nha config set key sk-ant-api03-YOUR_KEY');
+    console.log('  With a local model, no key (Ollama or any OpenAI-compatible endpoint):');
+    console.log('    nha config set legion-provider ollama');
+    console.log('    nha config set ollama-model qwen2.5:7b');
     console.log('');
     process.exit(1);
   }
@@ -533,6 +553,8 @@ function cmdConfig(args) {
       console.log('');
       info('Keys: provider, key, openai-key, gemini-key, deepseek-key, grok-key (X.AI), groq-key (voice/Whisper), mistral-key, cohere-key, model, timeout');
       info('      verbose, immersive, deliberation, rounds, convergence, tribunal, knowledge');
+      info('      legion-provider, ollama-url, ollama-model, ollama-models, ollama-embed-model (local models for "nha run")');
+      info('      local-openai-url, local-openai-model, local-openai-key, orchestrator-provider, economy, fact-check, cross-reading-chars');
       info('      google-client-id, google-client-secret');
       info('      microsoft-client-id, microsoft-client-secret, microsoft-tenant');
       info('      telegram-bot-token, discord-bot-token, responder-auto-route');
@@ -541,7 +563,9 @@ function cmdConfig(args) {
     }
     const success = setConfigValue(key, value);
     if (success) {
-      ok(`${key} = ${value.startsWith('sk-') ? value.slice(0, 12) + '...' : value}`);
+      // A secret is never echoed back, whatever it starts with.
+      const secret = /(key|token|secret)$/i.test(key) || value.startsWith('sk-');
+      ok(`${key} = ${secret ? '***' : value}`);
     } else {
       fail(`Unknown config key: ${key}`);
     }
@@ -577,6 +601,14 @@ function cmdConfig(args) {
   console.log(`    Rounds:       ${W}${config.deliberation.rounds}${NC}`);
   console.log(`    Convergence:  ${W}${config.deliberation.convergence}${NC}`);
   console.log(`    Tribunal:     ${config.deliberation.tribunalEnabled ? G + 'yes' : D + 'no'}${NC}`);
+  const legionConfig = buildLegionConfig(config);
+  const readiness = legionReadiness(legionConfig);
+  console.log(`    Provider:     ${W}${legionConfig.provider || '(any configured)'}${NC}`);
+  console.log(`    Can run with: ${readiness.ready ? G + readiness.providers.join(', ') : R + readiness.reason}${NC}`);
+  if (legionConfig.ollamaModels || legionConfig.ollamaModel) {
+    console.log(`    Local models: ${W}${legionConfig.ollamaModels || legionConfig.ollamaModel}${NC}  ${D}(${legionConfig.ollamaUrl})${NC}`);
+  }
+  if (legionConfig.economy) console.log(`    Economy:      ${G}yes${NC}`);
 
   console.log(`\n  ${C}Agent Identity${NC}`);
   if (config.agent.name) {
@@ -745,10 +777,10 @@ async function cmdDoctor() {
   console.log(`  Node.js:          ${nodeV >= 22 ? G + process.version : Y + process.version + ' (need 22+)'}${NC}`);
 
   // Check core files
-  const legionOk = fs.existsSync(path.join(NHA_DIR, 'core', 'legion-x.mjs'));
+  const legionOk = fs.existsSync(LEGION_FILE);
   const pifOk = fs.existsSync(path.join(NHA_DIR, 'core', 'pif.mjs'));
-  console.log(`  Legion X:         ${legionOk ? G + 'installed' : R + 'missing'}${NC}`);
-  console.log(`  PIF:              ${pifOk ? G + 'installed' : R + 'missing'}${NC}`);
+  console.log(`  Legion X:         ${legionOk ? G + 'v' + bundledLegionVersion() + ' (bundled)' : R + 'missing'}${NC}`);
+  console.log(`  PIF:              ${pifOk ? G + 'installed' : D + 'not installed (optional)'}${NC}`);
 
   // Check agents
   const agentCount = AGENTS.filter(a => fs.existsSync(path.join(AGENTS_DIR, `${a}.mjs`))).length;
@@ -769,6 +801,10 @@ async function cmdDoctor() {
   // Check API key
   console.log(`  API Key:          ${config.llm.apiKey ? G + 'configured (' + config.llm.provider + ')' : R + 'NOT SET'}${NC}`);
 
+  // Deliberations run locally: which of the user's providers they can use
+  const readiness = legionReadiness(buildLegionConfig(config));
+  console.log(`  Deliberations:    ${readiness.ready ? G + 'ready (' + readiness.providers.join(', ') + ')' : R + readiness.reason}${NC}`);
+
   // Check connectivity
   try {
     const res = await fetch('https://nothumanallowed.com/api/v1/health', {
@@ -787,24 +823,27 @@ async function cmdDoctor() {
   if (!config.llm.apiKey) {
     warn('Configure an API key: nha config set provider anthropic && nha config set key YOUR_KEY');
   }
-  if (!legionOk || !pifOk) {
-    warn('Missing core files. Run "nha update" to re-download.');
+  if (!readiness.ready) {
+    warn('For "nha run" with a local model and no key: nha config set legion-provider ollama && nha config set ollama-model <model>');
+  }
+  if (!legionOk) {
+    warn('Legion X is missing from this installation. Reinstall: npm install -g nothumanallowed');
   }
   if (agentCount < AGENTS.length) {
-    warn(`${AGENTS.length - agentCount} agents missing. Run "nha update" to re-download.`);
+    warn(`${AGENTS.length - agentCount} agents missing. Run "nha update" to restore them from the package.`);
   }
 }
 
 // ── nha version ────────────────────────────────────────────────────────────
 function cmdVersion() {
   console.log(`nha v${VERSION}`);
+  console.log(`Legion X v${bundledLegionVersion()}`);
 
-  // Show core versions if available
+  // PIF is downloaded separately: show its version if it is there
   const versionsFile = path.join(NHA_DIR, 'core', 'versions.json');
   if (fs.existsSync(versionsFile)) {
     try {
       const v = JSON.parse(fs.readFileSync(versionsFile, 'utf-8'));
-      if (v['legion-x']?.latest) console.log(`Legion X v${v['legion-x'].latest}`);
       if (v['pif']?.latest) console.log(`PIF v${v['pif'].latest}`);
     } catch {}
   }
@@ -823,8 +862,10 @@ function cmdHelp() {
   console.log(`    agents info <name>    Show agent capabilities & domain`);
   console.log(`    scan <path>           Security scan a project with SABER + ZERO`);
   console.log(`    scan . ${D}--output report.md${NC}   Save report to file`);
-  console.log(`    run "prompt"          Multi-agent collaboration (server-routed)`);
-  console.log(`    run "prompt" ${D}--agents saber,zero${NC}   Collaborate with specific agents\n`);
+  console.log(`    run "prompt"          Multi-agent deliberation, on this machine with your own models`);
+  console.log(`    run "prompt" ${D}--agents saber,zero${NC}   Deliberate with specific agents`);
+  console.log(`    run "prompt" ${D}--economy${NC}   About half the tokens (shorter cross-reading)`);
+  console.log(`    ${D}Local model, no key: config set legion-provider ollama + config set ollama-model <model>${NC}\n`);
 
   console.log(`  ${C}Daily Operations${NC}  ${D}(Gmail + Calendar + Tasks)${NC}`);
   console.log(`    ui                    Open local web dashboard (http://127.0.0.1:3847)`);
@@ -902,7 +943,7 @@ function cmdHelp() {
   console.log(`  ${C}Configuration${NC}`);
   console.log(`    config                Show current config`);
   console.log(`    config set <k> <v>    Set a config value`);
-  console.log(`    update                Update agents, core files & npm package`);
+  console.log(`    update                Update the npm package (Legion X and agents come with it)`);
   console.log(`    upgrade               Update npm package only (alias: self-update)`);
   console.log(`    doctor                Health check`);
   console.log(`    mcp                   Start MCP server (Claude Code, Cursor)\n`);
@@ -914,6 +955,7 @@ function cmdHelp() {
 
   console.log(`  ${D}38 agents: security, code, data, devops, creative, integration, and more.${NC}`);
   console.log(`  ${D}Use them solo or let them collaborate via multi-round deliberation.${NC}`);
-  console.log(`  ${D}Your API key never leaves your machine. Zero dependencies. Zero telemetry.${NC}`);
-  console.log(`  ${D}Docs: https://nothumanallowed.com/docs/cli — v${VERSION}${NC}\n`);
+  console.log(`  ${D}Your API keys go to your own LLM provider and nowhere else.${NC}`);
+  console.log(`  ${D}Each command sends one usage ping (platform and version) to nothumanallowed.com.${NC}`);
+  console.log(`  ${D}Docs: https://github.com/adoslabsproject-gif/nothumanallowed#readme — v${VERSION}${NC}\n`);
 }

package/src/commands/autostart.mjs

@@ -146,7 +146,7 @@ function launchdStatus() {
 function generateSystemdUnit() {
   return `[Unit]
 Description=NHA PAO Background Daemon
-Documentation=https://nothumanallowed.com/docs/cli
+Documentation=https://github.com/adoslabsproject-gif/nothumanallowed#readme
 After=network-online.target
 Wants=network-online.target
 

package/src/commands/chat.mjs

@@ -572,7 +572,7 @@ export async function cmdChat(args) {
 
   const activeProvider = config.llm.provider || 'anthropic';
   if (activeProvider !== 'nha' && !config.llm.apiKey) {
-    fail('No API key configured. Run: nha config set key YOUR_KEY\n  Or use the free tier: nha config set provider nha');
+    fail('No API key configured. Run: nha config set key YOUR_KEY');
     process.exit(1);
   }
 

package/src/commands/ops.mjs

@@ -105,7 +105,7 @@ export async function cmdOps(args) {
       let inactiveHintTel = 'configured but inactive (try: nha ops stop && nha ops start)';
       if (reason.startsWith('missing_key:')) {
         const p = reason.slice('missing_key:'.length);
-        inactiveHintTel = `configured but LLM key missing for provider "${p}" — fix with:  nha config set provider nha  (free Liara)  OR  nha config set ${p}-key YOUR_KEY`;
+        inactiveHintTel = `configured but LLM key missing for provider "${p}" — fix with:  nha config set key YOUR_KEY`;
       }
       console.log(`  Telegram:         ${responder.telegram ? G + 'active' + NC : telegramConfigured ? Y + inactiveHintTel + NC : D + 'not configured' + NC}`);
       console.log(`  Discord:          ${responder.discord ? G + 'active' + NC : discordConfigured ? Y + inactiveHintTel + NC : D + 'not configured' + NC}`);