By Craig Jellick, VP Engineering & Co-founder of Obot AI
v0.25.0 is out, and it closes the loop on something we've been working toward for a few releases: governing not just what flows through Obot, but what AI agents do on developer machines before a request ever reaches a gateway.
Device management in v0.22.0 gave admins inventory. v0.24.0 added audit logs via Obot Sentry. v0.25.0 adds enforcement: administrators can now define an allowlist of approved tools and MCP servers, and Obot Sentry checks every tool call against it through the coding agent's pre-tool hook before it runs. Fails closed. Every decision is logged.
But that's just the headline. This release also adds:
- MCP Tunnels — reach MCP servers on private networks via an outbound WebSocket connection, no inbound firewall changes required
- Agent Auth Scopes — scoped, expiring API keys for autonomous and headless agents that can't do a browser sign-in
- Obot Community tier — all auth and model providers now available without a n enterprise license
Full breakdown in the release notes below.
📓 Full release notes · 🐙 GitHub v0.25.0
Originally published on Obot AI
Top comments (0)