Integration Overview
UAICP integrates as a reliability control layer under your existing orchestration runtime.
This website provides complete integration guidance for v0.3 implementation.
Polyglot Support
UAICP provides official adapters in three languages:
| Language | Framework | Package | Tests |
|---|---|---|---|
| TypeScript | LangGraph | @uaicp/adapter-langgraph | 12 |
| Python | Microsoft AutoGen v0.4 | uaicp-adapter-autogen | 23 |
| Rust | Rig | uaicp-adapter-rig | 17 |
Integration Target
You keep your existing framework for orchestration and tool execution.
UAICP adds deterministic gates for:
- identity validation
- evidence gating
- verification gating
- policy-gated write actions with rollback safety
- multi-agent swarm hierarchy tracking
- secure read-only UX streaming
- replayable audit context
Canonical Integration Flow
framework request/context
-> map to UAICP envelope (include parent_trace_id if swarm)
-> flush UX partial streams safely (if applicable)
-> execute tools and collect evidence objects
-> run verifier checks (verifyGates)
-> run policy gate (enforcePolicy - must include rollback_action for writes)
-> deliver OR fail_safe (execute rollback)
Visual Architecture References
For component and framework-level diagrams, see:
Quick Start
Install published packages:
TypeScript
npm install @uaicp/core @uaicp/adapter-langgraph
Python
pip install uaicp-core uaicp-adapter-autogen
Rust
cargo add uaicp-core uaicp-adapter-rig
For contributor/source validation in this monorepo:
# TypeScript
cd libs/typescript && npm install && npm test
# Python
cd libs/python && pip install -e core/ -e adapter-autogen/ && pytest adapter-autogen/tests/
# Rust
cd libs/rust && cargo test --workspace
Definition of Complete Adapter Integration
An adapter is complete when it can:
- map framework state to UAICP envelope fields, including
parent_trace_idfor nested swarms. - enforce the
streamPartialboundaries to ensure UI streaming works without exposing write operations. - persist required evidence objects before delivery.
- block delivery on failed verification or missing evidence.
- enforce policy gate semantics on high-risk writes by enforcing
rollback_action. - emit reason-coded outcomes (
allow,deny,needs_review).
Implementation Locations
Contributor Links
Post-v0.3 enhancements are tracked in: