Open Science Assistant (OSA)

An extensible AI assistant platform for open science projects, built with LangGraph/LangChain and FastAPI.

Overview

OSA provides domain-specific AI assistants for open science tools with:

• HED Assistant: Hierarchical Event Descriptors for neuroimaging annotation
• BIDS Assistant: Brain Imaging Data Structure
• EEGLAB Assistant: EEG analysis toolbox
• NEMAR Assistant: BIDS-formatted EEG, MEG, and iEEG dataset discovery

Features:

• YAML-driven community registry - add a new assistant with just a config file
• Modular tool system for document retrieval, validation, and code execution
• Multi-source knowledge bases (GitHub, OpenALEX, Discourse forums, mailing lists)
• Embeddable chat widget for any website
• Production-ready observability via LangFuse

Installation

# From PyPI
pip install open-science-assistant

# Or with uv (recommended)
uv pip install open-science-assistant

Development Setup

# Clone and install in development mode
git clone https://github.com/OpenScience-Collective/osa.git
cd osa
uv sync --extra dev

# Install pre-commit hooks
uv run pre-commit install

Quick Start

CLI Usage

# Set up your API key
# Anthropic (what the platform itself runs on): https://console.anthropic.com/settings/keys
# OpenRouter (still supported for BYOK): https://openrouter.ai/keys
osa init

# Ask the HED assistant a question
osa ask -a hed "What is HED?"

# Start an interactive chat session
osa chat -a hed

# Show all commands
osa --help

API Server

Requires server dependencies: pip install 'open-science-assistant[server]'

# Start the API server
osa serve

# Or with uvicorn directly
uv run uvicorn src.api.main:app --reload --port 38528

Configuration

# Show current config
osa config show

# Set API keys for BYOK (Bring Your Own Key)
osa config set --anthropic-key sk-ant-...
osa config set --openrouter-key sk-or-v1-...

# Override API URL per-command
osa ask -a hed "What is HED?" --api-url https://api.osc.earth/osa-dev

Deployment

OSA can be deployed via Docker:

# Pull and run
docker pull ghcr.io/openscience-collective/osa:latest
docker run -d --name osa -p 38528:38528 \
  -e ANTHROPIC_API_KEY=your-key \
  ghcr.io/openscience-collective/osa:latest

# Check health
curl http://localhost:38528/health

See deploy/DEPLOYMENT_ARCHITECTURE.md for detailed deployment options including Apache reverse proxy and BYOK configuration.

Community Registry

OSA uses a YAML-driven registry to configure community assistants. Each community has a config.yaml that declares its documentation, system prompt, knowledge sources, and specialized tools.

# Directory structure
src/assistants/
    hed/config.yaml      # HED assistant configuration
    bids/config.yaml     # BIDS assistant (planned)

Adding a New Community

1. Create src/assistants/my-tool/config.yaml:

id: my-tool
name: My Tool
description: A research tool for neuroscience
status: available

# By default, every community runs on the shared Claude Platform key.
# Optional: only set these if the community funds its own usage instead.
# Set the named environment variable on your backend server.
# anthropic_api_key_env_var: "ANTHROPIC_API_KEY_MY_TOOL"
# openrouter_api_key_env_var: "OPENROUTER_API_KEY_MY_TOOL"

system_prompt: |
  You are a technical assistant for {name}.
  {preloaded_docs_section}
  {available_docs_section}

documentation:
  - title: Getting Started
    url: https://my-tool.org/docs
    source_url: https://raw.githubusercontent.com/org/my-tool/main/docs/intro.md
    preload: true

github:
  repos:
    - org/my-tool

2. (Optional) If you set either key env var, export it on your backend:

export ANTHROPIC_API_KEY_MY_TOOL="sk-ant-..."
# or, for a community that funds OpenRouter usage instead
export OPENROUTER_API_KEY_MY_TOOL="sk-or-v1-..."

3. Validate your configuration:

uv run osa validate src/assistants/my-tool/config.yaml

4. Start the server - the /{community-id}/ask endpoint is auto-created.

For the full guide, see the community registry documentation.

Documentation

Full documentation is available at docs.osc.earth/osa.

Development

# Run tests with coverage
uv run pytest --cov

# Format code
uv run ruff check --fix . && uv run ruff format .

License

MIT