What’s new in ControlClaw
Every release of the software that runs your agents and their firewall, newest first.
Subscribe to the feed- The first Telegram connection on a new agent no longer fails with a 'superseded' error.
- Apps rules are now Integrations rules and also cover every MCP server your agents use, tool by tool.
- An Integrations rule can carry a sentence, which the firewall's AI checks against each call.
- Without the AI accepted, MCP tools that do not say what they do count as destructive, so the default rule asks before each call; accept the AI on the Integrations tab to change that.
- Add MCP servers in Integrations and give them to your agents: the firewall keeps their keys and sign-ins, and the agents never see them.
- Agents see the tools of the apps you give them again.
- When an agent has only one connection to an app, the firewall picks it for the agent.
- The firewall now checks every app call an agent makes (GitHub, Notion, Slack and the rest) before it runs, with Apps rules to allow, ask or block each action.
- App calls are checked by your content rules, and each one shows in Activity with the app and the action.
- Every agent gets a ControlClaw skill that explains why the firewall stopped something and links to the right console page.
- The web scraping skill no longer tries to get around a site your firewall blocks.
- AI Log Review is separate from the AI Firewall, with its own switch and who gets an email about its findings.
- Agent cards and the agent's Home tab show a small picture of the agent's screen, refreshed every 30 seconds.
- Each agent's Home tab shows what it is doing now, its last actions and the title of each task; the Overview card says what it is working on.
- Heartbeats and scheduled runs show as Scheduled on the timeline and no longer count as tasks.
- Meeting instructions are now written on the agent's Meetings tab in the console, and each change is confirmed with a code from your firewall.
- Each agent's Home tab shows its CPU and memory over time, its activity of the last 12 hours, and a few counts.
- The Brain page has an Overview with how many pages the brain keeps, by type, and which agents use it.
- Team chat has an Overview with how many messages people and agents sent, and in which channels.
- A request to email someone new can now be answered with "Always allow this recipient for this mailbox", confirmed with a code.
- When an email waits for your approval, the agent is told why, with the request and the recipients, also on Gmail.
- Agents can check whether their request was approved, denied, expired or revoked.
- On your firewall's Content rules page, Back to console now opens your Messages rules.
- Content rules can now be written as a sentence, such as "Don't share salaries", and the firewall's AI checks every message against it, in chat, email, websites, webhooks, meetings and calls.
- When the AI check cannot run, messages a stop or hold rule covers are stopped, and the person writing is asked to send again later.
- Content rules: your organization can stop or hold messages your agents send or receive on Telegram, Slack, email, websites, webhooks and team chat.
- Content rules also cover meetings and calls: a blocked question never reaches your agent, and a blocked answer is not spoken.
- Held and stopped messages are listed on your firewall's own page, with their text kept on the firewall.
- In meetings, follow-up requests to a task are no longer dropped after the agent's acknowledgement.
- In meetings, the agent always acknowledges a request with a short word before it works on it.
- A follow-up to a task started in a meeting now continues that task, with what it did so far.
- Meeting lookups no longer repeat their instructions each time, so they stay small in long meetings.
- HTML widgets and pages the agent makes now open in its dashboard. Only signed-in members of the organization can see them.
- In meetings, the agent's lookups now carry only what was said since its last answer, so they stay small in long meetings.
- Groundwork for showing HTML widgets in the dashboard, signed in like the dashboard itself (not turned on yet).
- Meetings: the agent's lookups run on a fast model you can choose in its Meetings settings, so it answers sooner.
- Meetings: the agent no longer comments on the conversation by itself, opens and shows a page when asked to show a site, can browse during a meeting, and leaves when asked.
- Security: optional content rules on the firewall stop, hold for approval or record Telegram messages that match them; leaking secrets is always checked. Every change needs a confirmation code.