What’s new in ControlClaw
Every release of the software that runs your agents and their firewall, newest first.
Subscribe to the feed- Auto approve rules are now kept and checked on your firewall, and a change that lets more through needs your firewall's code.
- Calls your agents make in connected apps and MCP servers show in Activity again, with their rule counts.
- The agent's control service moves to a new port, so a firewall and an agent can run on one machine.
- Sites a model provider or channel needs are allowed by rules ControlClaw adds and removes by itself, so they keep working when you block every other site.
- Activity hides routine model and channel traffic by default; a switch shows it.
- The firewall can be updated, and updates itself at night, before you connect a channel.
- Pairing requests show the person's name.
- Agents and firewalls run only the new faster build of the ControlClaw software.
- Clearer setup steps in the agent templates.
- Web search no longer asks for approval, and an action you approved is not asked about again.
- Auto approve rules now apply to the agent they are set for.
- Connected apps show their logos on the firewall's pages.
- The firewall and agent software is updated to version 1.1.0.
- The firewall and your agents now run on the new, faster Rust build.
- Two test organizations move their agents' manager to the new Rust build.
- Two test organizations move their firewall's traffic filter to the new Rust build.
- The firewall's AI Log Review no longer flags your agents' own backups as data leaving the box.
- The test organizations' firewall can update the brain's AI settings again.
- Two test organizations move their firewall to the new Rust build.
- The firewall and agents can now run the new, faster Rust builds; they stay off until we turn them on for your organization.
- An agent's health check no longer returns the list of processes running on it.
- Meeting agents no longer leave a Google Meet a second after joining.
- Realtime phone calls work again on the newest OpenClaw.
- Meeting settings with a list where one choice is expected are refused when saved.
- Team chat messages and calls to an organization's MCP servers are always checked by the content and Integrations rules, however they are written.
- A website login's PIN can only go into the field that holds its stand-in.
- A new OpenClaw release is installed only after the agent has checked it; a failed install is rolled back.
- An agent no longer stays offline after an automatic update fails: OpenClaw is started again, and the agent restarts it by itself if it stops for any other reason.
- Refreshing an agent's firewall certificate no longer fails with 'fetch failed'.
- The firewall now checks the agents' traffic to ControlClaw and to backups, so these connections can no longer be used to reach other sites.
- The firewall's content rules can no longer be skipped with compressed, oddly formatted or oddly encoded messages, on Telegram and the other channels.
- Calls to an app's tools are checked even when they are written in an unusual way.
- New agents build again: the meeting step now accepts OpenClaw 2026.9.35.
- Meetings no longer end a few seconds after joining when Google opens the meeting at a /meet/ address.
- The firewall checks that each connection goes only to the site it names, and puts in a saved key only for that site.
- The Overview and each agent's Doctor tab list what is wrong, such as a refused model key, used-up AI credit or a disconnected channel, with a button to fix it.
- Owners and admins get at most one email a day when something new goes wrong; it can be turned off in Settings.
- The Firewall page shows which Security rules are at work: how often each one matched and what it did, with links to edit them.
- Every rule on Security rules shows how many times it matched in the last 7 days.