What’s new in ControlClaw
Every release of the software that runs your agents and their firewall, newest first.
Subscribe to the feed- Meetings: the agent answers questions in a meeting faster, on a fast model you can choose in its Meetings settings.
- Meetings: the agent leaves a meeting when someone asks it to, and presents its screen only when it was asked directly.
- Meetings: an owner can let the agent present its browser in a meeting (confirmed with a code); ask it to share its screen or stop.
- Meetings: each meeting's page links every session the agent ran for it to that session in the agent's dashboard.
- Agents on the included AI tokens reuse their cached prompt between replies, so Claude models answer faster and cost less.
- Meetings: an owner can let the agent do real work when asked in a meeting (Full agent), confirmed with a code; it says "ok, on it" and tells the room the result.
- Meetings: the agent answers only when it is sure it was addressed, says a short "ok" before it looks something up, and keeps its answers shorter.
- Confirmation codes sent on Telegram have number buttons: tap the number your console shows instead of typing the code.
- When you replace the organization's recovery key, every agent's disk is re-sealed to the new key, and you are told once the old key is no longer needed.
- Email rules: a new outgoing mode, Send to anyone, lets your agents email any address without asking you, up to your hourly and daily limits. Switching to it needs a confirmation code.
- An agent that takes a long time to start, for example after AgentMail is set up, is no longer restarted while it starts.
- Shared notes: your organization's notes now reach your agents, are kept inside each agent's encrypted disk, and your agents search them.
- Add storage: you can make an agent's own storage larger from its page while it runs.
- A new agent's page says Pairing its disk until your firewall can open it, and asks before a restart that would leave the disk locked.
- After applying a template you can test the agent, let Doctor fix a step that failed, and let an AI write the steps Save as template cannot read.
- Templates can add scheduled messages and MCP servers, and can use bundles; a bundle can also be applied to an agent that already runs.
- Skills and shared notes can be installed, updated and removed on agents that already run, with one confirmation code for several agents.
- A skill an agent wrote can be saved into your organization's skills.
- Code messages from your firewall say what is asked on the first line and show the code on the second.
- An Update no longer fails or waits when Google's download servers for Chrome are slow.
- Update your firewall before your agents. Update everything does this in the right order.
- Applying a template to an agent that is already running now asks for a confirmation code from your chat channel.
- Code messages from your firewall now show the code and the firewall's own facts first, and any text typed in the console separately below.
- Each request your firewall sends to an agent now works once and only for that agent, so a copied request cannot be reused.
- New agents start with the skills your organization marks for them on the Skills page.
- Update your firewall before your agents: an updated agent does not accept requests from an older firewall. Update everything does this in the right order.
- An agent's page has a Skills tab: what it has, where each skill came from, and a switch for each.
- A template can carry skills, and you can save an agent's setup as a template of your own.
- The agent page shows CPU, memory and disk use.
- When you apply a template to an agent that already has instructions, you can review and edit them on the agent.
- The disk figure now reads the way df does, and the fuller of the server's disk and the agent's own storage is shown.
- Your organization can keep its own agent templates and create or set up agents from them.
- A template's file steps write only inside the agent's workspace; anything else is a script you can read before confirming.
- A template can be applied to an agent that already runs: you see what would change first, then confirm once.
- Agent templates: a new agent can start as a Personal assistant, Software factory, Chief of staff or Support agent, set up for you when it is created.
- The agent page lists what is left for you to connect after a template, and ticks items off as you do them.
- Agent templates, first part: an agent can be set up from one template file with cc-runbook. Nothing in the console uses it yet.
- Signing in to Google with a saved login works again; it was blocked by mistake in 1.4.3.
- A request that asks the site to treat it as a different kind of request never gets a saved password, and the firewall's login page explains what a sign-in address must be.
- A saved website password is now put in only at the site's sign-in, in the password field. A login needs its sign-in page, and anything else that carries the stand-in is blocked.
- Meeting notes are written on an agent that has a mailbox, and a failed notes run is tried again later.
- An agent no longer freezes during heavy web research: it can hold many more connections, and it recovers by itself if it gets stuck.
- Meetings: the agent can listen the whole time without a wake word, set reminders when asked, join from calendar invites, join signed in to its Google account, and act on a meeting afterwards by your instructions.
- Your firewall's pages (Logs, Doctor, Website logins) have the ControlClaw look, and the page an agent sends you to add a login works well on a phone.
- Screenshots of a browser tab that is not in front no longer time out, and unused search providers are switched off.