AZMX AI
The sovereign agent platform.
Install
# macOS — Homebrew (live now via our tap; in homebrew-cask review)
brew install --cask AzmxAI/azmx/azmx
# macOS / Linux — one-line installer (live)
curl -fsSL https://azmx.ai/install | sh
# Windows — winget (manifest submitted to microsoft/winget-pkgs; live after review)
winget install AzmxAI.AZMX
Or grab a signed installer for your platform from the latest release:
| Platform | File |
| --- | --- |
| macOS (Apple Silicon) | AZMX.AI_<version>_aarch64.dmg |
| macOS (Intel) | AZMX.AI_<version>_x64.dmg |
| Linux | AZMX.AI_<version>_amd64.AppImage, .deb, or .rpm |
| Windows | AZMX.AI_<version>_x64_en-US.msi or *_x64-setup.exe |
Detailed walkthrough per platform: SETUP.md.
Why AZMX
- Your code stays on your machine. AZMX never proxies AI requests through our servers. Your prompts go directly from your device to whichever provider you chose, or to a local model. We don't sit in the middle — and we can't.
- Free local AI without an API key. One-click Ollama setup. Curated coding models (Qwen2.5-Coder, Granite Code, Llama, all Apache 2.0). Your code never leaves the device.
- Or BYOK any major provider. OpenAI, Anthropic, Google, Groq, xAI, Cerebras, DeepSeek, NVIDIA NIM, Azure OpenAI, any OpenAI-compatible endpoint. Pick whichever, switch any time.
- Per-call agent approval. The agent asks before it writes a file or runs a shell command. Configurable from Permissive to Paranoid (typed confirmation for destructive ops).
- Hash-chained, tamper-evident audit log. Every tool call recorded locally. Export to your SIEM on the paid tiers.
- Real terminal, real editor, real file explorer. xterm.js terminal, CodeMirror editor with vim mode + inline AI autocomplete, file tree with git badges. Not a marketing page that calls itself "the future of terminals."
- MCP-native. 17-server curated catalog (GitHub, GitLab, Kubernetes, Postgres, SQLite, Redis, Slack, Drive, …). Add your own in one JSON file.
- ~10 MB installer. Native. No Electron. Cold-start under a second. No telemetry, no account, no email-required-to-use.
The application source is proprietary. This repository publishes release artifacts and user-facing docs — installers, the auto-updater manifest, and the community contribution surfaces (skills, agents, MCP servers, snippets, translations). Source contributions land upstream and ship to you via the auto-updater.
First-run in under 3 minutes
- Open AZMX AI.
- The first-run tour appears. On step "Free local AI", click Set up local AI.
- AZMX walks you through installing Ollama (~150 MB).
- Pick Qwen2.5-Coder 7B (default — ~4.7 GB, Apache 2.0). The pull streams in-app.
- When the bar turns green, the AI panel is live. Ask anything.
Prefer BYOK? Skip step 2's button and use the BYOK step instead. Both paths land at a working composer.
🛠 Extend AZMX
Your contribution ships to every AZMX user worldwide on next release. No CLA. You keep copyright. You grant a permissive license. The bundled distribution gets richer every month because of community work.
Five surfaces, every one of them a single Markdown (or JSON) file:
| Build a … | What it is | Where to start |
|---|---|---|
| 🤖 Skill | A discipline the agent loads on demand to act like a domain expert (e.g. "Postgres query review"). One Markdown file. | skills/ → SKILL_AUTHORING.md |
| 🧑🚀 Sub-agent | A specialist the main agent delegates to, with bounded tools + predictable output (e.g. test-writer, migration-planner). | agents/ → AGENT_AUTHORING.md |
| 🔌 MCP connector | A new tool the agent can call — bridge to your service, your data, your internal CLI. | mcp-servers/ → MCP_AUTHORING.md |
| ✨ Snippet | A pre-baked prompt template accessible via #name in the composer. | snippets/ → SNIPPET_AUTHORING.md |
| 🌍 Translation | The onboarding card in your language — 5 seeded, ~10 priority next. | TRANSLATIONS.md |
Plus: add a recipe to the cookbook, fix a doc, or showcase someone else's work.
Every contributor gets credit in release notes. Featured contributors get a complimentary Pro license; repeat featured contributors get Teams. Read the principles →
📦 Official npm packages
Two MIT-licensed packages ship under the @azmxailabs npm scope. Use them to make your AI assistant aware of AZMX, or to build your own approval-gated agent with the same primitives AZMX uses internally.
| Package | What it does | Install |
|---|---|---|
| @azmxailabs/mcp 
| Official Model Context Protocol server. Drop it into Claude Desktop, Claude Code, Cursor, Windsurf, Continue, OpenAI Codex CLI, Cline, or any other MCP client — and the assistant gains grounded, authoritative knowledge about AZMX (pricing, BYOK providers, security posture, comparisons, install steps, latest release). | npx -y @azmxailabs/mcp |
| @azmxailabs/agent-sdk 
| TypeScript SDK that ships the four primitives behind AZMX as standalone, dependency-free modules — approval gate, deny-list, hash-chained audit log, BYOK provider router. Build your own agent (CI script, CLI, desktop app, server) with the same security posture AZMX has, none of the UI. | npm install @azmxailabs/agent-sdk |
Step-by-step walkthrough for either package: docs/DEVELOPER-GUIDE.md
Full reference docs on the website:
- MCP server (per-client setup recipes, tool/resource/prompt inventory, examples): azmx.ai/docs#azmxai-mcp
- Agent SDK (concepts, full API, recipes, production checklist): azmx.ai/docs#agent-sdk
Source: packages/ — pure TypeScript, Node ≥ 18, ESM, MIT.
Documentation
Get started
| | | | --- | --- | | SETUP.md | Install per platform · first-run setup | | MANUAL.md | Full feature reference — every panel, every shortcut | | FAQ.md | Common questions on privacy, licensing, performance, models | | CHANGELOG.md | Release notes (per-version detail on the releases page) |
Reference
| | | | --- | --- | | docs/TIERS.md | What's in Free / Pro / Teams / Enterprise | | docs/MODELS.md | Supported AI providers · BYOK · local models | | docs/CONNECTORS.md | The bundled MCP catalog · custom + workspace MCP | | docs/AGENTS.md | What the agent can + can't do · approval gates · audit log | | docs/KEYBINDINGS.md | Every shortcut, by category | | docs/RECIPES.md | Workflow cookbook — copy + adapt | | docs/GLOSSARY.md | Terms you'll see across the docs | | docs/ROADMAP.md | What we're working on |
Policies + posture
| | | | --- | --- | | SECURITY.md | Vulnerability reporting · what's protected | | PRIVACY.md | What we collect (nothing, by default) · what lives where | | LICENSE.md | EULA pointer · third-party notices · trademarks | | CODE_OF_CONDUCT.md | How we behave on this project | | CONTRIBUTING.md | How to engage with this repo | | docs/COMPLIANCE.md | SBOM · SOC 2 · DPA · FIPS · PIV/CAC | | docs/TELEMETRY.md | What gets sent (nothing, by default) | | docs/AIRGAPPED.md | Run AZMX with zero outbound |
Operate
| | | | --- | --- | | docs/TROUBLESHOOTING.md | Problem → fix, by surface | | docs/DATA_PORTABILITY.md | Export · backup · move