Libero

The architecture is remarkably clean for an open-source solution. Running a shared session per channel reduces noise, and the split between the gateway and the tool proxy ensures that tool credentials remain isolated. The use of mutual TLS on a private network adds a layer of confidence that is often missing in similar products. Having the admin surface as a file in git makes version control straightforward, allowing for clear diffs and easy rollbacks without needing a complex web interface. The ability to self-host with model-agnostic support fits perfectly into existing infrastructure. The append-only audit log provides transparent visibility into every action, which is crucial for compliance. Overall, the design prioritizes security and deterministic enforcement without sacrificing the flexibility needed for modern workflows.

It is nice to see a tool that focuses on clear rules rather than complex settings. The team sheets are easy to read.

Libero’s architecture provides robust isolation through its per-channel SQLite database files, ensuring that channel history and skills remain strictly scoped. The use of mutual TLS between the gateway and proxy adds a layer of security that is often overlooked in similar tools. The ability to bring your own model is a significant advantage for enterprises with existing licensing agreements. The deterministic enforcement via team sheets ensures that the AI cannot bypass permissions through prompt engineering. While the admin surface relies entirely on TOML files, this approach ensures that configuration is version-controlled and reviewable, providing a clear audit trail for compliance purposes.

The team sheet approach sounds clever, but it feels a bit rigid. Since the admin surface is just a TOML file in git, you really need to check the terms before booking. If your team isn’t comfortable managing configurations via code, this might not be the smooth experience you expect.

Libero offers a refreshing approach to AI in Slack by keeping governance deterministic rather than relying on the model’s good behavior. The split between the gateway and the tool proxy is a smart design choice that keeps credentials secure. I appreciate that approvals are explicit and require a human click before destructive actions. However, the lack of a web admin interface might feel dated to some, even though the git-based team sheets are powerful. The append-only audit logs provide excellent traceability. While the pre-1.0 status suggests some format changes may occur, the current stability and clear security model make it a strong candidate for teams needing strict control over their AI interactions.