Why Chat Agent Relay Exists
The Story
Every team building AI-powered chat follows the same path:
- Start with a Slack bot + OpenAI integration. 200 lines. Ship it.
- PM asks for Teams. Double the integration code.
- Compliance needs audit logs. Build a logging system.
- CEO wants to try Claude instead. Rewrite the agent integration.
- Each change multiplies the codebase. Each platform leaks into your logic.
The Insight
The layer between chat platforms and agents is always the same pattern — receive, decide, route, invoke, respond, deliver. Chat Agent Relay makes this pattern explicit.
Architecture Overview
architecture.txt
Chat Platforms CAR Middleware Your Agents ┌──────────┐ ┌──────────────────────────┐ ┌────────────────┐ │ Slack │ │ Canonical Event Chain │ │ Internal agent │ │ Teams │────>│ Policy + Governance │────>│ LangChain agent│ │ Discord │ │ Routing + Dispatch │ │ Self-hosted LLM│ │ Web │ │ Audit Ledger │ │ Any HTTP agent │ └──────────┘ └──────────────────────────┘ └────────────────┘
Core Design Principles
Protocol-first
Canonical event model defines the contract. Implementations follow.
Zero lock-in
Channel adapters and backend adapters are pluggable interfaces.
Auditable by default
Every decision is an immutable event in an append-only ledger.
Governance built-in
Policy evaluation happens before agent invocation, not after.
Developer-first
TypeScript, Bun, JSON Schema, conformance test suite.
What Chat Agent Relay is NOT
- Not a bot builder (use Botpress for that)
- Not an AI framework (use LangChain for that)
- Not a customer support tool (use Chatwoot for that)
- Chat Agent Relay is the middleware that connects these pieces
Technical Details
- 12 packages, 222 tests, TypeScript strict mode
- Built on Bun for speed
- SQLite for durable event storage
- JSON Schema for contract validation
- Conformance test suite for adapter validation