Skip to content

See docs/DISCLAIMER_SNIPPET.md

Project Overview

AGI-Alpha-Agent-v0 explores a meta-agentic framework where agents spawn, evaluate and refine other agents. The codebase ships an offline-friendly demo called α‑AGI Insight that runs either locally or via the OpenAI Agents API when keys are provided.

The Insight demo implements a best‑first search over agent rewrite chains. It can forecast disruptive sectors and iteratively improve plans by rewriting itself. When no cloud credentials exist, it falls back to a local Meta-Agentic Tree Search with small sample datasets.

Key capabilities include:

  • Modular orchestrator that selects between local and remote runtimes
  • Tools to run demos entirely offline using a wheelhouse
  • Example agents and a minimal browser client

Minimal Setup

  1. Verify the environment and install Python packages: bash python check_env.py --auto-install
  2. Copy the sample environment: bash cp alpha_factory_v1/.env.sample .env
  3. Launch the default stack: bash ./quickstart.sh
  4. Run the Insight demo: bash alpha-agi-insight-v1 --episodes 5 Add API keys in .env to enable cloud features; otherwise the demo stays offline.

For a deeper dive, read quickstart.md and the other documents in this folder. The Architecture Overview page summarises how the orchestrator, agents and memory components interact.