Claude Code and ChatGPT, live in your meeting
Today we are proud to launch your own Claude Code and ChatGPT inside Tana meetings. With them, Tana becomes the first agentic meeting platform for teams. Bring the agent you already use into the call and the work gets done while you are still talking. It has the live transcript, the shared screen and your own connectors, so it acts on what is being discussed during the call, not after it.
We built this because we were done leaving every meeting with homework. The decisions were made in the room, and then the work waited for whoever had time after. Now it starts while the conversation is still going, and it is done before the meeting ends.
Your agent works during the meeting, not after it
Plenty of meeting tools now let Claude read your transcripts once the meeting is over. Tana is the first where your own Claude Code or ChatGPT works during the meeting, with the live transcript, the shared screen and the connectors you have already set up.
It is the Claude Code or ChatGPT you already run on your computer, signed in with your own account. Search your email through your connectors, work in a folder or a repository, build a slide deck, a dashboard or a working prototype, and keep going on it while the conversation moves on. It brings your skills, your connectors and everything you have taught it.
It hears the meeting as it happens
Your agent already knows how to draft, build and research. Until now it only heard about a meeting afterwards, as a summary or a list of follow-ups. In Tana it gets the call live: what people are saying, what is in the chat, and what is on the screen someone is sharing. When a customer walks you through the export that breaks, you ask your agent to file the bug, and it works from what was said and shown.

Ask it in the chat, or say “Hey Claude” out loud while you talk. It also listens on its own and suggests what it could do for you, and you approve with a click. Nobody on the call has to type a prompt for the work to start.
It hears the call in whatever language people speak. Transcription is multilingual and detects the language per speaker, so a call that moves between Norwegian and English still transcribes everyone correctly. What Tana writes from it, the summary, the outcomes, the generated documents and their titles, comes out in the content language your organization picks in Settings, from 33 languages, whatever language the conversation was in. The Tana interface itself is English.
Done before the meeting ends
Most meeting tools leave you with a summary and a list of things to do. With your agent in the meeting, the list gets done before the call ends, so there is no need to circle back. Here is what that looks like with the connectors your agent already runs.
- The dashboard check. Someone shares the retention chart and asks whether 46% is good. Your agent pulls last year’s number from PostHog or Amplitude, checks the industry baseline on the web, and answers in the chat before the next slide.
- The follow-up, sent. You say “I’ll follow up with Priya on the two decisions.” Your agent drops the drafted email into the chat. You read it, say send, and it goes out from your Gmail before the meeting is over.
- The PR, started in the meeting. The PM walks the engineer through the feature with the design on screen. The engineer’s Claude Code has the repo, the transcript and the screen share, so it starts the branch while they are still talking through edge cases. The call ends with a draft PR open and the Linear issue linked, not a ticket to interpret on Monday.
- The design, updated while you look at it. The Figma file is on screen. “What if the filter sat above the list, and we add an empty state?” Your agent makes the change in Figma while everyone watches, so the review turns into a working session and you leave with the design you agreed on.
- The plan, updated, and the team told. You agree to push the launch two weeks. Your agent updates the dates and owners in the project plan in Google Sheets, creates the two new issues in Linear and assigns them to whoever said “I’ll take it”, then posts the change in the team’s Slack channel. It is done before you move to the next item.
- The pricing question, answered on the call. The prospect asks what 40 seats on an annual plan would cost with SSO, and whether you do that for other companies their size. Your agent pulls the price list, checks the deal history in HubSpot and what their champion said last call, and drops the number and two references in the chat. You answer in the meeting instead of saying “I’ll get back to you”, and the deal does not lose a week.
These are examples. Your agent works with any app you have already connected to it, with the same setup, skills and tools it has on your computer.
No bot joins the call
Tana runs on your machine. For a Zoom, Google Meet or Microsoft Teams call, a small Tana window opens next to the call with the meeting’s timeline, so everyone keeps their usual setup. Nothing joins the call, and nobody else on it sees what your agent is doing. Meetings hosted in Tana work the same way.
Tana records nothing until you start it. It also detects when you join a Zoom, Teams, Meet or Slack call that never made it onto your calendar and asks whether to capture it, so an ad hoc call gets the same agent, transcript and outcomes as a scheduled one.
Private until you share it
What your agent makes stays with you until you share it. The chat in every meeting has two sides:
- Agents is your private tab. Your own agent answers there, and the suggestions from the meeting arrive there for you to accept, dismiss or hand to a different agent.
- Meeting is the shared tab. Everyone in the call sees it, guests included, with the timeline and screenshots. Docs holds the documents pinned to the meeting, and Transcript is beside them.
- Everyone brings their own. One person on Claude and another on ChatGPT work the same meeting side by side, each with their own connectors. The team is already talking about the work, and their agents pick up what they can take.
- You choose what the room sees. When something is ready, choose Share with everyone under its answer and Tana posts it to the Meeting tab under your name and your agent’s. Don’t share keeps it to yourself.
Your agent does not act on its own. Before it changes anything in Tana it asks you, and every change it makes is a proposal you review before it lands.
Tana is built for teams. Every meeting adds to your team’s shared context, so decisions and the thinking behind them reach the people who were not in the room, and access control decides who sees what. On the Business plan, SAML single sign-on and groups let you manage that across the whole organization.
Your agent runs on your own computer through your own Claude or ChatGPT account, and Tana never uses your meeting data to train a model. Tana has completed its SOC 2 Type 2 examination, and you can choose the EU as the region where your data is processed and stored. Read more in Your own Claude or ChatGPT.
Set it up before your next call
You need the Tana desktop app on Mac, Windows or Linux, and the Claude or ChatGPT app (Claude Code on Linux) installed on the same computer, since that is where your agent runs. For Claude, your account needs access to Claude Code, which is what Tana runs.

- Open Settings in Tana and go to External agents.
- Choose Sign in next to Claude or ChatGPT and finish signing in with the account you already use. If Tana cannot find the app on your computer, it says what to install, and the troubleshooting guide walks you through it.
- Open a meeting. In the Agents tab, pick your agent under Who answers here, or say “Hey Claude” during the call.
Until you connect one, Home offers it as your next step, with a button for each agent that takes you straight to the same settings.

Claude and ChatGPT are available today, with more agents on the way. Don’t have one yet? Tana’s own meeting assistant works out of the box, and your Claude or ChatGPT can come later.
The Tana Team

Turn meetings into the most productive part of your day.
Explore further
All updates



