ouro-ai 0.5.3

An open-source AI agent — run it as a Coding agent CLI or deploy it as a bot just like JARVIS.

Ouro is derived from Ouroboros—the ancient symbol of a serpent consuming its own tail to form a perfect circle. It represents the ultimate cycle: a closed loop of self-consumption, constant renewal, and infinite iteration.

At Ouro AI Lab, this is our blueprint. We are building the next generation of AI agents capable of autonomous evolution—systems that learn from their own outputs, refine their own logic, and achieve a state of infinite self-improvement.

Two Modes, One Agent

Ouro ships with a unified agent core and two deployment modes:

Architecture

Ouro is organized into three layers with strict downward-only imports:

Each layer has its own README — start with the umbrella overview, then drill into core, capabilities, or interfaces.

Features

🤖 Agent Swarm — Multi-Agent Swarm with Persistent Tasks

The flagship feature. Enable with ENABLE_AGENT_SWARM=true in ~/.ouro/config.

• Persistent Task Store — SQLite-backed tasks with dependency graphs (task_create, task_claim, task_update, task_list, task_get, task_delete)
• Atomic Task Claiming — Agents race to claim available tasks; one agent, one in-progress task
• Auto-Swarm — Complex tasks are automatically decomposed and executed by multiple agents in parallel
• Replaces legacy TodoTool and MultiTaskTool when enabled

🔄 Self-Verification — Ralph Loop

The agent verifies its own answer against the original task and re-enters the loop if incomplete. Enable with --verify or RALPH_LOOP_MAX_ITERATIONS=3.

🧠 Memory System

LLM-driven compression, file-based long-term memory, FTS5 conversation recall, and YAML session persistence resumable across restarts.

💬 Dual Deployment

Same agent core, two modes:

• CLI — Interactive REPL with rich TUI, slash commands, session resume
• Bot — Persistent IM assistant for Lark, Slack, WeChat with proactive cron scheduling

🔐 OAuth Login

--login / /login to authenticate with ChatGPT Codex subscription models.

📊 Benchmarks

First-class Harbor integration for agent evaluation (see Evaluation).

Quick Start

Prerequisites: Python 3.12+ and one of uv (recommended) or pipx.

# Recommended: installs ouro in an isolated environment and exposes global
# `ouro-cli` and `ouro-bot` commands
uv tool install ouro-ai

# Alternative
pipx install ouro-ai

# Upgrading later
uv tool upgrade ouro-ai      # or: pipx upgrade ouro-ai

Plain pip install ouro-ai also works but is not recommended — it mixes ouro's dependencies into your active Python environment. Use uv tool / pipx so the ouro-cli / ouro-bot binaries are on $PATH without needing to activate a venv.

On first run, ~/.ouro/models.yaml is created. Add your API key:

models:
  openai/gpt-4o:
    api_key: sk-...
default: openai/gpt-4o
current: openai/gpt-4o

Then run:

# Interactive mode
ouro-cli

# Single task
ouro-cli --task "Calculate 123 * 456"

# Bot mode
ouro-bot

See LiteLLM Providers for the full provider list.

Evaluation

Ouro can be evaluated on agent benchmarks using Harbor. A convenience script harbor-run.sh is provided at the repo root:

1. Edit harbor-run.sh to set your model, dataset, and ouro version.
2. Run:

export OURO_API_KEY=<your-api-key>
./harbor-run.sh                    # run with defaults in the script
./harbor-run.sh -l 5               # limit to 5 tasks
./harbor-run.sh --n-concurrent 4   # 4 parallel workers

Extra flags are forwarded to harbor run, so any Harbor CLI option works. See ouro_harbor/README.md for full details.

Documentation

• CLI Guide -- interactive mode, task mode, commands, shortcuts
• Bot Guide -- IM bot setup, commands, proactive mechanisms, personality
• Configuration -- model setup, runtime settings, custom endpoints
• Examples -- usage patterns and programmatic API
• Sandbox Providers -- BoxLite/smolvm sandbox tools and configuration
• Memory Management -- compression, persistence, token tracking
• Task V2 -- persistent task store with dependency graphs (Phase 1)
• Extending -- adding tools, agents, LLM providers
• Packaging -- building, publishing, Docker

Contributing

Contributions are welcome! Please open an issue or submit a pull request.

For development setup (install from source):

git clone https://github.com/ouro-ai-labs/ouro.git
cd ouro
./scripts/bootstrap.sh         # creates .venv and installs editable + dev deps
source .venv/bin/activate
./scripts/dev.sh check         # precommit + typecheck + tests

End-users should prefer uv tool install ouro-ai (see Quick Start); the source checkout is only needed when contributing.

License

MIT License