// the find
caipe-io/ai-platform-engineering
CAIPE is an open-source AI platform for building, governing, and operating AI agents and agentic workflows for platform engineering and beyond.
CAIPE is a CNCF-adjacent platform for building and governing multi-agent systems in a platform-engineering context — think agent orchestration, identity/authz, auditing, and chatbot integrations (Slack/Webex) all under one roof. It's aimed at platform teams who want to run their own agent infrastructure rather than wire together point solutions themselves.
The monorepo is genuinely modular: dynamic_agents (runtime, routing, auth), autonomous_agents (scheduling, webhooks, Mongo-backed state), audit_service, and a cron-runner are separable services each with their own pyproject.toml, uv.lock, and test suite. CI is taken seriously — per-component workflows (ci-rag, ci-audit-service, ci-dynamic-agents, etc.), plus prebuild pipelines, zizmor for GitHub Actions security linting, and a pinned-deps check. It plugs into real infra standards instead of reinventing them: Keycloak for identity, OpenFGA for authorization, MCP for tool/agent interop.
The README is almost entirely community boilerplate — meeting times, Slack invite, star history chart — and tells you nothing about the actual architecture or how the pieces talk to each other; you have to go spelunking through source or an external docs site to understand what you'd be running. The surface area is huge: auth, authz, auditing, scheduling, RAG, a skill-scanner, two chat bot integrations — adopting this means operating a small platform, not dropping in a library. It reads as governance-first (CODEOWNERS, a separate governance repo, weekly Webex calls) before it reads as solving a sharply defined problem, which tends to slow down iteration speed on the actual agent logic.