Skip to content

AION-by-ASA-Computer/AION_Agent

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

230 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

AION Agent

Self-hosted AI agent platform with MCP tool integration, multi-level memory (STM/LTM), YAML profiles, skills, and Plan Mode for structured multi-step work.

CI OpenSSF Scorecard License Python FastAPI Haystack Next.js React TypeScript Docker Redis SQLite MCP Caddy

Commercial site: https://aion-asa.com · Docs: docs/

AION Agent is a self-hosted, enterprise-grade platform for building and running truly sovereign AI. Design custom agent profiles, extend capabilities through modular skills and native MCP tool integration, and share memory seamlessly across teams — all while keeping fully autonomous agents isolated at the organizational level. Built for security, control, and scale, AION Agent puts your AI infrastructure entirely in your hands.

Screenshots

Chat UI

AION Chat UI: streaming chat, plan dock, attachments

Admin UI

AION Admin UI: profiles, users, memory, agent DB

Quick start (Docker)

The recommended way to run AION is the full Docker stack (Caddy + backend + chat-ui + admin-ui + docs + Redis).

Local (HTTP, no domain)

git clone https://github.com/AION-by-ASA-Computer/AION_Agent.git
cd AION_Agent

cp .env.example .env
# Required: set your LLM endpoint (see DEPLOY DOCKER section in .env.example)
#   AION_API_URL=http://host.docker.internal:11434/v1   # Ollama on the host
#   AION_MODEL=llama3.2
#   DOMAIN=:80

./scripts/setup-aion-env.sh --docker
docker compose up -d --build
URL Service
http://localhost/ Chat UI
http://localhost/admin Admin UI
http://localhost/docs/ Documentation
http://localhost/api/health Backend health

Production (HTTPS via Let's Encrypt)

git clone https://github.com/AION-by-ASA-Computer/AION_Agent.git
cd AION_Agent

cp .env.example .env
# Set DOMAIN=your.example.com, LETS_ENCRYPT_EMAIL, AION_PUBLIC_API_URL, secrets
./scripts/setup-aion-env.sh --docker
docker compose up -d --build

Caddy provisions TLS automatically. One DNS A record pointing at the server is enough.

Details: docs/deployment/docker.md

Pre-built images (GHCR)

Published on each GitHub Release. Pin a version in production:

export AION_VERSION=1.0.0
docker compose -f docker-compose.yml -f docker-compose.ghcr.yml pull
docker compose -f docker-compose.yml -f docker-compose.ghcr.yml up -d --no-build

For always-current dev/staging pulls, omit AION_VERSION (defaults to latest via docker-compose.ghcr.yml).

See docs/opensource/releases.md.

Development compose (hot reload)

API + chat-ui + Redis only — run admin-ui and docs via pnpm dev when needed:

docker compose -f docker-compose.dev.yml up

Local development (without Docker)

Prerequisites: Python 3.13+, uv, pnpm 9+, OpenAI-compatible LLM.

cp .env.example .env
# Set AION_API_URL, AION_MODEL, AION_LLM_API_KEY

./scripts/setup-aion-env.sh
uv venv && uv pip install -r requirements.txt

# Terminal 1 — API (port 8001, single worker required)
uvicorn src.api.main:app --reload --reload-exclude data/sessions

# Terminal 2 — Chat UI (port 8003)
cd chat-ui && pnpm install && pnpm dev

Open http://localhost:8003. Optional admin UI: cd admin-ui && pnpm dev --webpack (port 3870).

# Curated CI test suite (no live LLM)
./scripts/run_ci_tests.sh

# Lint
uv run ruff check --config ruff.toml src/
uv run ruff format --check --config ruff.toml src/

Constraints: backend must run with one worker; use pnpm in JS packages; import src.aion_env before reading os.environ in scripts.

Assistant onboarding: AGENTS.md, CLAUDE.md.

Setup

After successfully booting the AION Agent codebase (either using Docker or through Local Development), you MUST perform the initial configuration using the interactive First Setup Wizard in the Admin Panel:

  1. Access the Admin Panel: Open your browser and navigate to the /admin path (e.g., http://localhost/admin or http://localhost:3870 in dev).
  2. Log In: Authenticate using the default credentials:
    • Username: admin
    • Password: admin
  3. Configure the System: If the platform is not yet initialized, you will be redirected automatically to /first-setup. Follow the steps on screen:
    • 1. Default LLM Provider: Select your target language model provider (OpenAI, Anthropic, Google Gemini, Ollama, or vLLM), then specify the model name, API key, optional API base URL, max chat tokens, and thinking token budget.
    • 2. Autonomous Memory (Embeddings): Set up the embeddings model (such as text-embedding-3-small or Google's models/text-embedding-004) and API credentials used to compute vector embeddings for Long-Term Memory (LTM).
    • 3. OCR Document Processing (Optional): Toggle vision-based OCR capabilities and configure service parameters (e.g., GLM-OCR model and base URL) to enable parsing and extracting text from uploaded images and PDFs.
    • 4. Web Integration & Search: Configure web search providers (Tavily, Brave Search, or a self-hosted SearXNG instance) to allow the agent to fetch search results. Define the default search provider and the search fallback order.
    • 5. Sandbox Security & Exec Policy: Secure your local environment by selecting an execution and path security template. You can choose between the Development Policy (loads fs_policy.dev.yaml), the Production Policy (loads fs_policy.example.yaml), or write a custom YAML policy blueprint.
    • 6. Review & Initialize: Confirm all configurations. Clicking Initialize AION Kernel saves these settings directly to the .env file (handled by settings_api.py), writes the security policy to config/fs_policy.yaml, and automatically restarts the AION kernel/backend to apply changes.

Default credentials

When admin auth is enabled (default): admin / admin — change password on first login.

Chat auth is optional (AION_CHAT_PASSWORD_AUTH=0 opens chat without login; useful for local dev).

Generate secrets for production:

openssl rand -hex 32   # AION_CHAT_AUTH_SECRET (if chat password auth enabled)

Features

Area Description
Chat UI (chat-ui/) Primary Next.js client: SSE streaming, attachments, plan dock
Admin UI (admin-ui/) Profiles, users, memory, agent DB management
MCP tools Dynamic stdio/SSE tools; registry in config_std/mcp_registry.yaml
Memory STM window, LTM extraction (MemPalace), SQLite + FTS history
Profiles & skills YAML profiles; Markdown skills with frontmatter
Plan Mode Tool-first plans, human approval, background execution
Docker Production stack with Caddy path routing; dev compose with hot reload

Project layout

src/              FastAPI backend, agent pipeline, memory, MCP manager
chat-ui/          Primary Next.js chat client
admin-ui/         Admin dashboard (Next.js)
website/          Docusaurus docs site
config_std/       Committed config templates (synced to config/ at setup)
mcp_servers_std/  Committed MCP server sources
docs/             Documentation source of truth
data/             Runtime data (gitignored; eval fixtures whitelisted)

See docs/architecture/source-tree.md.

Contributing

See CONTRIBUTING.md. Open PRs against main on this repository.

Documentation

Rendered site: build from website/ or read sources in docs/.

Topic Path
Architecture docs/architecture/
Configuration / env docs/configuration/
API & runtime docs/api-and-runtime/
MCP docs/mcp/
Security docs/security/

Telemetry

OpenTelemetry and Opik hooks exist but are off by default (AION_OTEL_ENABLED=0). No phone-home analytics in the default configuration.

Security

Report vulnerabilities privately — SECURITY.md.

Disclaimer

AION Agent is under active development. APIs, configuration, and behavior may change between releases without prior notice.

The software is provided “as is”, without warranty of any kind. AION / ASA Computer assumes no liability for damages, data loss, security incidents, or operational issues arising from the use of this project in any environment (including production).

You are responsible for evaluating fitness for your use case, securing your deployment, and complying with applicable laws and third-party license terms.

License

Apache License 2.0. Third-party notices: NOTICE.

Star History

Star History Chart

About

Enterprise-ready platform for sovereign AI. Build custom agent profiles, manage skills and MCP integrations, share memory across users, and run fully isolated autonomous agents at scale

Topics

Resources

Code of conduct

Contributing

Security policy

Stars

Watchers

Forks

Releases

Packages

Used by

Contributors

Languages