An agent platform that onboards like a new hire, not a chatbot

What it does

It’s an open-source (Apache-2.0), self-hostable enterprise agent platform: Python 3.12 backend, Vue 3 frontend, Docker deployment, multi-LLM support, and ten agents out of the box — Product Owner, Tech Lead, QA, an office assistant, even a white-box pentest expert. The pitch is agents as employees rather than apps: they ingest your Confluence pages, PRDs, and API docs into structured project knowledge, plan tasks into trees, execute under individual identity and permissions, and leave auditable trails.

The interesting bit

The repo claims to be maintained by its own agents — the Product Owner and Tech Lead agents running on the live site handle requirements and PRs for this very repository. Which means you can contribute without writing code: chat with the Product Owner agent, and it hands work to Tech Lead. The dogfooding is the demo.

Key highlights

• Knowledge ingestion is deliberate, not vector-search roulette: primary files load fully into context, with [[slug]] cross-references and two-level archiving for API docs; ambiguous cases get clarified with the user instead of guessed
• Experience compounds: skills, workflows, and knowledge entries turn one person’s tuning into shared, inheritable project assets
• Agents collaborate via @mentions — plug one into a conversation as needed
• Multi-database support (SQL Server, PostgreSQL, MySQL, SQLite), OAuth SSO for production, per-user encrypted model keys
• Scheduled tasks and an Agent-as-a-Service angle round out the enterprise checklist

Caveats

• Default deployment ships with AUTH_BYPASS_ENABLED=true — fine for local play, but you must switch it off and configure SSO before anything serious
• First Docker build runs 10–30 minutes (Playwright plus a pentest toolchain), and defaults to Chinese mirror sources — expect friction on non-Chinese networks
• The “no-code contribution” flow depends on the hosted site being up and behaving

Verdict

Worth a look if you want agents with identity, permissions, and institutional memory instead of a smarter chat box — the knowledge-onboarding loop is genuinely different from the RAG norm. If you just want a quick local coding agent, this is a heavier, organization-shaped machine.