Yutha AI Agent

Yutha

A framework-agnostic control plane for agent swarms. Identity, capability, accountability, and norms — for agents from any framework, on any backend.

Multi-agent systems work in demos and break in production. The reason is almost always the same: there's no shared substrate for who an agent is, what it's allowed to do, what it actually did, and which norms govern the swarm it lives in. Each framework reinvents a fragment of that and stops.

Yutha is the substrate. It runs in front of agents you've already built — in LangGraph, CrewAI, OpenAI Agents, Microsoft Agent Framework, or anything else you'd like to write an adapter for — and gives them passports, signed receipts, attenuated capabilities, declarative constitutions (Cedar+) with four-stage enforcement, and an optional cryptographic verification layer (Sui anchoring) when you need to prove what happened to a third party.

Full documentation

→ yutha.ai — concepts, operator guide, developer guide, and worked examples.

The doc site is the canonical reference. Start at the landing page; pick the operator guide if you're running a swarm, the developer guide if you're building agents that join one.

What's here

/spec        — wire & artifact specs (RFC-governed)
/crates      — Rust workspace: control plane, registry, capability, transport, receipts, cedar+ engine
/backends    — Pluggable backends: Postgres receipts, Sui anchoring
/sdks        — Framework adapters (Python: LangGraph, CrewAI, OpenAI Agents, MAF)
/contracts   — Move package for Sui receipt anchoring
/docs        — Source for the doc site at yutha.ai

Install the Python SDK

pip install yutha                       # core SDK
pip install 'yutha[langgraph]'          # + LangGraph adapter
pip install 'yutha[crewai]'             # + CrewAI adapter
pip install 'yutha[openai-agents]'      # + OpenAI Agents adapter
pip install 'yutha[maf]'                # + Microsoft Agent Framework adapter

Python 3.11+. The control plane is a Rust binary — clone the repo and cargo run -p yutha-control-plane to bring one up, or follow the operator quickstart for the longer playbook.

Quickstart

The fifteen-minute joiner path (developer) and thirty-minute initiator path (operator) live on the doc site under Developer Guide → Quickstart and Operator Guide → Quickstart respectively.

If you want to poke locally without the doc site:

• End-to-end LangGraph example. A customer-support swarm with capability-gated messaging, operator-driven eviction, and a verifiable audit trail. Runnable demo at sdks/python/examples/s1_support_queue.py; walkthrough at docs/developer/langgraph.md.
• Constitution + four-stage enforcement. Four runnable demos across four adapters: code_review.py (LangGraph), ap_invoice.py (CrewAI), research_crew.py (OpenAI Agents), devops_incident.py (Microsoft Agent Framework). Each exercises the Cedar-based constitution layer + detect → coach → quarantine → evict.
• Conformance suite. cargo test -p yutha-conformance runs the in-process scenarios covering the receipt log, send-path enforcement, operator revocation, and constitution evaluation.

Contributing

See CONTRIBUTING.md. The project is stewarded by a single maintainer (@abhinavg6); guidelines are intentionally light.

Spec changes go through the RFC process. Vulnerability reports go to GitHub private security advisories — see SECURITY.md.

License

Apache License 2.0. See LICENSE.