Quickstart

Get the assistant running in your terminal in 5 minutes

Prerequisites

• Python 3.12+
• PortAudio (for audio support)
  • macOS: brew install portaudio
  • Ubuntu/Debian: sudo apt-get install portaudio19-dev python3-dev
• A Unify account — sign up free, grab your API key from the dashboard
• An LLM API key — OpenAI or Anthropic

Install

Unity depends on two sibling repos. Clone all three as siblings:

git clone https://github.com/unifyai/unity.git
git clone https://github.com/unifyai/unify.git
git clone https://github.com/unifyai/unillm.git

cd unity
pip install uv && uv sync

Configure

cp .env.example .env

Open .env and fill in:

UNIFY_KEY=your-unify-key
OPENAI_API_KEY=sk-...        # and/or ANTHROPIC_API_KEY

Everything else has sensible defaults. The sandbox connects to Unify’s hosted backend (api.unify.ai) for persistence — no local database needed.

Run

source .venv/bin/activate
python -m sandboxes.conversation_manager.sandbox --project_name Sandbox --overwrite

Select Mode 1 (SandboxSimulatedActor) at the configuration prompt. This runs the full ConversationManager brain with simulated backends.

Interact

Once the REPL starts:

> msg Hey, can you help me organize my upcoming week?
> sms I need to reschedule my meeting with Sarah to Thursday
> email Project Update | Here are the Q3 numbers you asked for...

Commands: msg (Unify message), sms (SMS), email (email), call (phone call), meet (video meeting). Type help for the full list.

Going deeper

Mode 1 simulates everything to show the ConversationManager’s orchestration. For the real CodeAct architecture (where the Actor writes and executes Python plans against the manager APIs), select Mode 2 at the configuration prompt.

See the full sandbox docs at sandboxes/conversation_manager/README.md.

Next steps

• Steerable Handles — the core architectural pattern
• CodeAct — how the Actor writes programs, not tool calls
• Architecture overview — the distributed manager system