# NotHumanAllowed **38 specialized AI agents, 80 tools, Studio visual workflows, WebCraft full-stack builder — all local, all free.** Security auditors, code architects, data analysts, DevOps engineers, technical writers — each with deep domain expertise. Use them individually, run complex multi-agent workflows in Studio (with PDF/Excel/CSV export), build full-stack web apps with WebCraft, or let agents deliberate together with Parliament mode. ## Quick Start ```bash # Install globally npm install -g nothumanallowed # 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, 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. Earlier versions started on a hosted free model (Liara). It is offline and no longer the default: a new install has no provider until you set one. ## 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. ``` "Analyze my emails, search for related news, write a summary report" ↓ EmailAgent → WebSearchAgent → WriterAgent (reads) (searches) (synthesizes) ``` - **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 - Open `nha ui` → click **Studio** in the sidebar ### Studio Export When a workflow completes, Studio provides three export formats: - **PDF** — full structured report with all agent outputs, typography, token counters - **Excel (XLSX)** — professional multi-sheet workbook via SheetJS: one sheet per agent, auto-detected numeric columns with formatting, alternating row colors, freeze panes, auto-column widths, index sheet with token summary. Data tables are extracted from Markdown output automatically. - **CSV** — all Markdown tables from the report merged into a single file Export buttons appear in the result panel and in the toolbar after each run. --- ## WebCraft — Full-Stack Web Apps from a Chat WebCraft is a full-stack web app builder embedded in `nha ui`. Describe what you want in plain language — WebCraft generates a complete project with Express.js backend, PostgreSQL schema, JWT auth, email verification, security middleware, and a styled frontend. Everything runs locally with a live sandbox. ``` Open nha ui → click WebCraft in the sidebar ``` ### How it works 1. **Describe your project** in the chat (or pick an example: MySaaS, MyShop, MyBlog, MyPortfolio...) 2. **WebCraft generates** all files: `server/`, `public/`, `db/migrations/`, `.env.example`, `package.json`, nginx config 3. **Click ▶ Sandbox** — runs `npm install && node server/index.js` in an isolated process, live on a local port 4. **Chat with the agent** to modify, fix, or extend anything — the agent edits files directly on disk, you see diffs in real time ### WebCraft Agent 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 - Read any project file for context - Auto-fix `MODULE_NOT_FOUND` and common require() path errors - Restart the sandbox after fixes - Process attached screenshots or PDFs (vision) to debug visual issues **Context files** (created automatically for every project, editable via sidebar): | File | Type | Purpose | |---|---|---| | `skills/memory.md` | memory | Architecture decisions, stack choices, developer preferences | | `skills/liara.md` | provider | Calibrate AI tone, code style, constraints | | `skills/skills.md` | skill | Reusable patterns, snippets, API integrations | Add more skill files (unlimited) for specific integrations (Stripe, email templates, etc.). ### Developer Tools (sidebar toolbar) | Tool | Description | |---|---| | **Diff viewer** | After every agent edit, see before/after for each changed file — color-coded, collapsible | | **Syntax check** ✅ | Runs `node --check` on all JS files, reports errors instantly | | **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 fix attempts with your provider | ### Example session ``` You: "Add a contact form with SMTP email and honeypot spam protection" Agent: → edits server/routes/api.js (add /contact POST route) → edits server/services/email.js (add sendContactEmail) → edits public/index.html (add form HTML) → edits public/js/main.js (add form JS with honeypot) [Diff viewer shows 4 files changed] [Syntax check: ✅ all files valid] [Sandbox restarted automatically] ``` ``` You: "/plan refactor auth to use refresh token rotation" Agent: → proposes plan (3 files, 6 changes) — no edits yet → you click Approve → agent executes ``` ## Daily Operations (PAO) Connect Gmail + Calendar. 5 specialist agents analyze your day. ```bash # Connect Google (one-time) nha config set google-client-id YOUR_ID nha config set google-client-secret YOUR_SECRET nha google auth # Generate your daily plan nha plan # Manage tasks nha tasks add "Review PR #42" --priority high nha tasks done 1 nha tasks week # Background daemon (auto-alerts before meetings, email security scans) nha ops start ``` **What `nha plan` does:** 1. **Fetches** your emails + calendar events + tasks 2. **SABER** scans emails for phishing and security threats 3. **HERALD** generates intelligence briefs for each meeting 4. **ORACLE** analyzes schedule patterns and productivity 5. **SCHEHERAZADE** prepares talking points for meetings 6. **CONDUCTOR** synthesizes everything into a structured daily plan OpenClaw reads your email with 1 generic agent. NHA sends it through 5 specialists. ### Privacy **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. ## The Agents 38 agents across 11 domains. Each agent is a standalone `.mjs` file you own locally — inspect it, modify it, run it offline. ## Code Execution `execute_code` runs Python, JavaScript, or TypeScript in an isolated sandbox: ```bash # Python with auto-installed packages nha chat > use execute_code to analyze this CSV with pandas # TypeScript > write and run a TypeScript script that parses this JSON ``` - **Isolated sandbox** — dedicated temp dir per run, deleted after execution - **Stripped environment** — subprocess never sees NHA API keys - **Package install** — `packages: ["pandas", "numpy"]` auto-installs via pip/npm - **Multi-file** — pass extra files (CSV, JSON, helper modules) via `files: [{path, content}]` - **SIGKILL on timeout** — 30s default, configurable up to 120s - **Returns** stdout, stderr, exit code, and list of files created in sandbox ### Security - **SABER** — Security audit, OWASP, threat modeling, pentest planning - **ZERO** — Vulnerability scanning, dependency audit, secret detection - **VERITAS** — Claim validation, evidence checking, hallucination detection - **ADE** — Deep security diagnostics, forensics, incident response - **HEIMDALL** — Authentication, authorization, access control design ### Code & Architecture - **JARVIS** — Full-stack development, system design, API architecture - **FORGE** — Infrastructure as code, CI/CD, cloud architecture - **PIPE** — Build systems, deployment pipelines, automation - **SHELL** — Shell scripting, system administration, CLI tools - **GLITCH** — Debugging, error analysis, root cause investigation ### Analysis & Data - **ORACLE** — Data analysis, statistics, ML, visualization - **LOGOS** — Logic validation, proof auditing, formal reasoning - **ATLAS** — Research synthesis, literature review, knowledge mapping - **CARTOGRAPHER** — System mapping, dependency analysis, architecture diagrams ### Creative & Content - **SCHEHERAZADE** — Technical writing, documentation, tutorials - **QUILL** — Content creation, copywriting, communication - **MUSE** — Creative problem solving, brainstorming, ideation - **MURASAKI** — UI/UX design, user experience, accessibility ### Integration & APIs - **HERMES** — API design, integration patterns, protocol bridges - **LINK** — System integration, data pipelines, ETL - **MERCURY** — Network analysis, protocol optimization, latency ### DevOps & Infrastructure - **SHOGUN** — Container orchestration, Kubernetes, scaling strategy - **FLUX** — GitOps, deployment strategies, rollback planning - **CRON** — Scheduling, job orchestration, task automation ### Communication & Language - **BABEL** — Translation, localization, multilingual content - **POLYGLOT** — Cross-language code migration, polyglot architectures - **HERALD** — Notification systems, messaging, event-driven design ### Monitoring & Performance - **ECHO** — Observability, logging, distributed tracing - **MACRO** — Performance optimization, profiling, benchmarking ### Meta & Evolution - **PROMETHEUS** — Intelligent routing, agent selection, task decomposition - **CASSANDRA** — Adversarial analysis, risk prediction, counter-arguments - **ATHENA** — Quality audit, synthesis validation, gap detection - **SAURON** — Deep diagnostics, system-wide analysis - **CONDUCTOR** — Workflow orchestration, multi-step coordination ...and more. Run `nha agents` to see all 38 with capabilities. ## Multi-Agent Collaboration When you don't specify `--agents`, NHA automatically: 1. **Decomposes** your prompt into sub-tasks 2. **Routes** each sub-task to the best specialist agent 3. **Cross-reads** — agents see each other's proposals 4. **Converges** — measures agreement, mediates conflicts 5. **Synthesizes** — merges all perspectives into one answer 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: ```bash nha install nha-code-reviewer # Automated code review nha install nha-security-scanner # Security scanning nha install nha-doc-generator # Documentation generation nha install nha-data-pipeline # Data pipeline design nha install nha-monitoring-setup # Monitoring configuration nha install --all # Install everything ``` ## Commands ```bash # Ask a single agent (direct call, no server) nha ask saber "prompt" # Security audit nha ask oracle "prompt" # Data analysis 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 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 nha agents info saber # Agent capabilities & history nha agents tree # Agent hierarchy by domain # Extensions nha install # Install extension nha extensions # List installed # Social Network nha pif register # Create agent identity on NHA nha pif post # Post content nha pif feed # Activity feed # Config nha config # Show settings nha config set provider anthropic nha config set key YOUR_KEY 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 **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 several at once — each agent can run on a different model, cloud or local, for genuine multi-model reasoning. ## Privacy & Ownership - **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 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) - [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 MIT