Report Abuse

Basic Information

This repository is an open-source, conceptual research prototype and meta-agentic platform (Alpha‑Factory / α‑AGI) designed to create, orchestrate and evaluate populations of intelligent agents and higher-order “meta‑agents.” It bundles an orchestrator that auto-discovers agents, a MuZero‑style world‑model and planner, a Memory & Knowledge Fabric (pgvector + Neo4j) and multiple domain agents for finance, biotech, manufacturing, policy, energy, supply‑chain, drug design and more. The project provides runnable demos, a browser insight UI, CLI tools and an API server with a POST /simulate endpoint. It supports cloud and air‑gapped deployments, falls back to local distilled models when APIs are absent, and integrates optional runtimes such as OpenAI Agents, Google ADK and MCP. The README focuses on reproducible demo workflows, Docker/Helm deployments, offline wheelhouse installs, and governance, observability and safety tooling.

Links

Categorization

App Details

Features
The codebase exposes a modular orchestrator with REST and gRPC facades, agent hot‚Äëloading, Kafka telemetry and signed proof‚Äëof‚Äëalpha messages. A World‚ÄëModel & Planner uses MuZero++ latent dynamics, a Meta‚ÄëAgentic Tree Search demo demonstrates recursive agent rewrites and an Experience‚ÄëFirst loop supports curriculum generation. Memory Fabric combines dense vector search with causal graph edges (pgvector + Neo4j) and a SQLite fallback. The repo ships a gallery of 14 demos (Insight, Marketplace, Finance, ASI world‚Äëmodel, self‚Äëhealing CI, business demos), offline browser demos via Pyodide and GPT‚Äë2 assets, Llama/gguf local model support and tooling for building an offline wheelhouse. Deployment and CI automation include Docker Compose, Helm charts, signed container verification, pre‚Äëcommit hooks, tests, and scripts to fetch and verify demo assets.
Use Cases
For researchers and developers the project provides a complete sandbox to design, run and evaluate multi‚Äëagent experiments and meta‚Äëagent workflows across industries. It lets teams simulate agent populations, run zero‚Äëdata insight search loops, benchmark world‚Äëmodel planning, and prototype domain agents (finance, biotech, supply‚Äëchain, security, etc.). Features such as offline inference, wheelhouse builds, service worker browser assets and deterministic CI make it usable in restricted or air‚Äëgapped environments. Observability integrations (Prometheus, OpenTelemetry, Grafana), governance primitives (MCP envelopes, attestations, audit ledger) and sandbox limits support safer experimentation. The demos, notebooks and one‚Äëclick Docker/Helm recipes accelerate reproducible evaluation and deployment while optional ADK and Agents SDK adapters enable cross‚Äëorg federation and hybrid local/cloud runtimes.

Please fill the required fields*