https://github.com/brandonkorous/axon
Self-hosted AI command center. Orchestrate AI advisors with persistent memory, voice, and a real-time boardroom.
https://github.com/brandonkorous/axon
ai ai-agents business-tools claude command-center developer-tools docker llm multi-agent ollama open-source personal-assistant self-hosted voice-ai
Last synced: 2 months ago
JSON representation
Self-hosted AI command center. Orchestrate AI advisors with persistent memory, voice, and a real-time boardroom.
- Host: GitHub
- URL: https://github.com/brandonkorous/axon
- Owner: brandonkorous
- License: agpl-3.0
- Created: 2026-03-23T22:41:13.000Z (3 months ago)
- Default Branch: main
- Last Pushed: 2026-04-01T00:58:40.000Z (3 months ago)
- Last Synced: 2026-04-03T02:46:09.578Z (3 months ago)
- Topics: ai, ai-agents, business-tools, claude, command-center, developer-tools, docker, llm, multi-agent, ollama, open-source, personal-assistant, self-hosted, voice-ai
- Language: Python
- Homepage:
- Size: 57.8 MB
- Stars: 1
- Watchers: 1
- Forks: 0
- Open Issues: 1
-
Metadata Files:
- Readme: README.md
- Changelog: CHANGELOG.md
- Contributing: CONTRIBUTING.md
- Funding: .github/FUNDING.yml
- License: LICENSE
- Code of conduct: CODE_OF_CONDUCT.md
- Codeowners: .github/CODEOWNERS
- Security: SECURITY.md
Awesome Lists containing this project
README
Your self-hosted AI command center.
Orchestrate AI advisors with persistent memory, voice interfaces, and a real-time boardroom — entirely on your infrastructure.
---
---
## Why Axon?
Most AI tools give you a chatbot. Axon gives you a **boardroom**.
Run multiple AI advisors — CEO, CTO, COO, and any custom persona you define — that maintain persistent memory across sessions, debate each other in real-time, and proactively surface insights you didn't think to ask for. Everything runs on your machine. Your data never leaves your infrastructure.
---
## Features
### Intelligence
| | Feature | Description |
| ------------- | ------------------------- | --------------------------------------------------------------------------------- |
| **Agents** | Specialist Advisors | A team of domain experts with distinct personas, vaults, voices, and delegation rules — not one generic chatbot |
| **Boardroom** | Real-Time Huddles | Group discussions with multiple AI specialists. Six modes: standard, vote, devil's advocate, pressure test, quick take, decision |
| **Brain** | Structured Reasoning | Graph-based reasoning engine with multi-strategy evaluation, confidence scoring, contradiction detection, and full decision traceability |
| **Delegate** | Task Delegation | Advisors delegate work to each other autonomously (sync or async) across research, audit, implementation, and investigation tasks |
| **Recruit** | Agent Recruitment | Request new specialist agents on the fly with user approval workflows |
### Memory
| | Feature | Description |
| ------------- | ------------------------- | --------------------------------------------------------------------------------- |
| **Vault** | Neural Memory Trees | Obsidian-compatible markdown vaults with YAML frontmatter, wikilinks, full-text search, and graph-based relationship tracking |
| **Consolidate** | Memory Consolidation | LLM-driven vault maintenance — duplicate merging, stale archiving, contradiction detection, orphan adoption, and confidence scoring |
| **Recall** | Intelligent Recall | Context-aware retrieval that surfaces relevant vault entries during conversations. Outcome linking lets advisors learn from experience |
| **Watch** | Vault Sync | File watcher for external changes — edit vaults in Obsidian and they sync automatically |
### Interaction
| | Feature | Description |
| ------------- | ------------------------- | --------------------------------------------------------------------------------- |
| **Voice** | Voice-First Interface | Whisper STT, Piper/ElevenLabs/Azure TTS, per-advisor voice catalog, adjustable speed, continuous voice mode |
| **Connect** | Platform Integrations | Slack (Socket Mode), Microsoft Teams (Bot Framework), Zoom (meetings + transcription), Discord (server deployment) |
| **Dashboard** | Command Center | Unified view: active agents, kanban task board, issues, approvals, vault health, and per-agent cost tracking |
| **Commands** | Slash Commands | Direct memory operations: `/sleep`, `/remember`, `/recall`, `/forget`, `/tasks`, `/status` |
### Autonomy
| | Feature | Description |
| ------------- | ------------------------- | --------------------------------------------------------------------------------- |
| **Plugins** | Shell Access & Sandbox | Grant agents host filesystem access (shell) or containerized execution (sandbox) via configurable plugins with executable allowlists |
| **Sandbox** | Isolated Sandboxes | 8 sandbox image types (base, browser, code, data, ML, documents, media, full) with dependency chains and resource limits |
| **Schedule** | Proactive Scheduling | Background heartbeat that picks up pending tasks and reviews completed work at configurable intervals |
| **Research** | Deep Research | Two-tier LLM strategy (local compression + reasoning analysis), web scraping, YouTube transcript extraction, multi-source synthesis |
| **Media** | Media Processing | YouTube transcript extraction and analysis with two-tier compression for cost-effective processing |
### Infrastructure
| | Feature | Description |
| ------------- | ------------------------- | --------------------------------------------------------------------------------- |
| **Models** | Org-Level Model Management | Register models per org, assign roles (navigator, reasoning, memory, agent), with per-agent overrides. Supports Anthropic, OpenAI, DeepSeek, Google, Groq, xAI, or fully local via Ollama |
| **K8s** | Kubernetes Support | Run sandboxes as Docker containers (local) or Kubernetes pods (production) — provider abstraction with pre-built images from ghcr.io |
| **Orgs** | Multi-Organization | Isolated vaults, agents, and settings per organization with pre-built templates (Startup, Student, Job Hunt, Family, Creator) |
| **Shield** | Full Audit Trail | Append-only, immutable audit logs filterable by date, agent, action, or tool — complete transparency |
| **Extend** | Plugins & Skills | Plugin architecture with registry and 10 built-in skills (brainstorming, code review, debugging, decision analysis, etc.) |
| **Secure** | Encryption & Isolation | AES encryption for stored credentials, container sandboxing for workers, network isolation via Docker or Kubernetes NetworkPolicies |
| **Work** | Task & Issue Management | Tasks (P0-P3 priority) with parent-child relationships, activity threads, simplified lifecycle (pending/in progress/blocked/done/accepted), and auto-generated achievements |
---
## System Requirements
| | Cloud LLMs (API) | Local LLMs (Ollama) |
|---|---|---|
| **Docker** | Required | Required |
| **RAM** | 4 GB minimum | 16 GB minimum, 32 GB+ recommended |
| **Disk** | 2 GB | 10–50 GB (depends on models) |
| **GPU** | Not needed | Optional — CUDA GPU with 8 GB+ VRAM strongly recommended |
| **API Key** | Anthropic or OpenAI | Not needed |
### Local LLM Model Sizing Guide
| Your Hardware | Recommended Models | Approx. Download |
|---|---|---|
| 8 GB RAM, no GPU | `llama3:8b`, `phi4-mini:3.8b` | 3–5 GB |
| 16 GB RAM or 6 GB+ VRAM | `qwen2.5:7b`, `mistral:7b` | 4–5 GB |
| 32 GB RAM or 8 GB+ VRAM | `qwen2.5:14b`, `mistral-small:22b` | 9–13 GB |
| 48 GB+ RAM or 12 GB+ VRAM | `qwen2.5:32b`, `llama3.1:70b` | 20–40 GB |
> **Not sure what your system can handle?** Run `axon doctor` after installing — it detects your hardware and recommends models automatically.
---
## Quick Start
### Option A: Axon CLI (recommended)
The CLI detects your system, walks you through LLM provider setup, and recommends models based on your hardware.
**macOS / Linux:**
```bash
# 1. Install the CLI
curl -sS https://get.useaxon.dev | sh
# 2. Create a workspace (interactive — detects hardware, picks models)
axon init my-workspace
# 3. Launch
cd my-workspace && axon start
```
**Windows (PowerShell):**
```powershell
# 1. Install the CLI (requires Docker Desktop)
irm https://get.useaxon.dev | iex
# 2. Create a workspace (restart your terminal first if prompted)
axon init my-workspace
# 3. Launch
cd my-workspace; axon start
```
> **Windows via WSL?** If you prefer WSL, use the macOS/Linux instructions above inside your WSL terminal.
### Option B: Manual setup
```bash
# 1. Clone the repository
git clone https://github.com/brandonkorous/axon.git
cd axon
# 2. Configure your environment
cp .env.example .env # Windows: copy .env.example .env
# Edit .env — add your API keys (Anthropic, OpenAI, or use Ollama for fully local)
# 3. Launch
docker compose up
```
Open **[http://localhost:3000](http://localhost:3000)** and meet your advisors.
> **Want fully local LLMs?** See [Local LLM Support](#local-llm-support) below.
---
## Architecture
Axon runs three services via Docker Compose:
```
┌─────────────────────────────────────────────┐
│ Frontend │
│ React 19 · Vite · TailwindCSS │
│ DaisyUI · Framer Motion │
│ :3000 │
└──────────────────┬──────────────────────────┘
│ REST / WebSocket
┌──────────────────▼──────────────────────────┐
│ Backend │
│ FastAPI · SQLAlchemy · LiteLLM │
│ SQLite (default) or Postgres │
│ :8000 │
└──────┬───────────────┬──────────────┬───────┘
│ │ │
┌──────▼──────┐ ┌──────▼──────┐ ┌────▼────────────┐
│ LLM Providers│ │ Ollama │ │ Sandboxes │
│Claude, OpenAI│ │ Local LLMs │ │ Docker or K8s │
└─────────────┘ │ :11434 │ │ ghcr.io registry │
└─────────────┘ └─────────────────┘
```
- **Frontend** — React SPA with real-time agent activity, boardroom view, and vault management
- **Backend** — FastAPI server handling agent orchestration, memory persistence, and multi-provider LLM routing via LiteLLM
- **Ollama** (optional) — Run models like `llama3`, `qwen2.5`, and others entirely on your hardware
- **Sandboxes** — Isolated execution environments (Docker containers or Kubernetes pods) with pre-built images for code, browser, data science, ML, and more
---
Configuration
### Environment Variables
Copy `.env.example` to `.env` and configure:
| Variable | Description | Required |
| ------------------- | ------------------------------------------------ | ------------------- |
| `ANTHROPIC_API_KEY` | Anthropic API key for Claude models | If using Claude |
| `OPENAI_API_KEY` | OpenAI API key | If using OpenAI |
| `DEFAULT_MODEL` | Fallback LLM model (prefer org-level model management) | No |
| `OLLAMA_BASE_URL` | Ollama endpoint (default: `http://ollama:11434`) | If using local LLMs |
| `DATABASE_URL` | Database connection string (default: SQLite) | No |
| `VAULT_PATH` | Path to the memory vault directory | No |
For a full list of options, see [`.env.example`](.env.example).
Adding Custom Agents
### Create a New Advisor
Define a new advisor by adding a YAML file to the personas directory:
```yaml
# personas/cfo-advisor.yaml
name: CFO Advisor
role: Chief Financial Officer
description: Financial strategy, fundraising, unit economics, and fiscal discipline.
model: claude-sonnet-4-20250514
voice: onyx
system_prompt: |
You are a seasoned CFO advising a startup. You focus on burn rate,
runway, unit economics, and fundraising strategy. Be direct and
data-driven. Flag financial risks early.
```
Restart the backend and your new advisor appears in the dashboard. No code changes required.
Local LLM Support
### Run Fully Local with Ollama
No API keys needed. Run everything on your machine:
```bash
docker compose --profile local-llm up
```
This starts Ollama alongside the frontend and backend. Three models are pulled on first start:
| Model | Size | Default Role |
|-------|------|-------------|
| `qwen2.5:7b` | ~4.5 GB | Navigator (tool routing, intent classification) |
| `llama3:8b` | ~4.7 GB | Memory (vault recall, consolidation) |
| `qwen2.5:14b` | ~9 GB | Reasoning (agent conversations) |
Customize via environment variables:
```env
OLLAMA_NAVIGATOR_MODEL=qwen2.5:7b
OLLAMA_MEMORY_MODEL=llama3:8b
OLLAMA_MODEL=qwen2.5:14b
```
To pull additional models or re-run after config changes:
```bash
# Force re-run model pulls
docker compose --profile local-llm up ollama-init --force-recreate
# Pull a specific model manually
docker exec axon-ollama-1 ollama pull qwen2.5:7b
# List available models
docker exec axon-ollama-1 ollama list
```
On first load, Axon prompts you to **register models and assign roles** at the org level. Use "Discover Ollama Models" to auto-detect your local models.
Supported local models include `qwen2.5`, `llama3`, `mistral`, and any model available in the [Ollama library](https://ollama.com/library).
---
## Roadmap
- [x] Multi-agent orchestration with persistent memory
- [x] Voice-first interface with per-persona voices
- [x] Real-time boardroom / Huddle sessions (6 modes)
- [x] Docker Compose deployment
- [x] Multi-LLM support (Claude, OpenAI, Ollama)
- [x] Achievement system and audit logging
- [x] Slack, Teams, Zoom, and Discord integrations
- [x] Plugin system with registry and built-in web research
- [x] Agent-to-agent delegation chains (sync and async)
- [x] Scheduled agent behaviors (task pickup, done review)
- [x] Structured reasoning engine with decision graphs
- [x] Memory consolidation and intelligent recall
- [x] Shell access and sandbox plugins for agent execution
- [x] Deep research with web scraping and YouTube transcripts
- [x] Skills system (brainstorming, code review, debugging, decision analysis, etc.)
- [x] Organization templates (Startup, Student, Job Hunt, Family, Creator)
- [x] Simplified task lifecycle with parent-child tasks and auto-achievements
- [x] Google Calendar and Linear integrations
- [x] Org-level model management with role assignments and curated catalog
- [x] Kubernetes sandbox support with provider abstraction (Docker + K8s)
- [x] Pre-built sandbox images via GitHub Container Registry
- [ ] Tool router — navigator model selects and instructs agent tools
- [ ] RAG over uploaded documents and codebases
- [ ] Mobile companion app
- [ ] Multi-user collaboration with role-based access
- [ ] One-click cloud deploy templates (Railway, Fly.io)
- [ ] Webhook triggers for external event-driven advice
---
## Contributing
We welcome contributions of all kinds. See [**CONTRIBUTING.md**](CONTRIBUTING.md) for guidelines on getting started, code style, and the PR process.
Before opening a large PR, please open an issue or discussion first so we can align on approach.
---
## Community
- **Website** — [useaxon.dev](https://useaxon.dev) for docs, blog, and getting started guides
- **Discord** — [Join the server](https://discord.gg/axon) for support, feature discussions, and showcases
- **GitHub Discussions** — [Ask questions and share ideas](https://github.com/brandonkorous/axon/discussions)
- **Twitter / X** — Follow [@axon_ai](https://twitter.com/axon_ai) for updates
---
## License
Axon is licensed under the [**GNU Affero General Public License v3.0 (AGPL-3.0)**](LICENSE).
You are free to use, modify, and self-host Axon. If you distribute a modified version or run it as a network service, you must make your source code available under the same license.
---
Built with ❤️ by the Axon community