Agent profile · Enterprise & productivity
Tana Agentic Meetings
Provider: Tana
Tana integrates existing Claude and ChatGPT agents into live meeting transcripts, chat and screen shares. Results stay private until deliberately shared; external agents run through the local desktop app and their own credentials.
Original evidence-based decision map
Tana Agentic Meetings — Decision and approval boundaries
Product scope, supervision, and open questions before live agent actions.
Overview & direct answer
What is Tana Agentic Meetings — Overview & direct answer
Tana Agentic Meetings is an agent-enabled meeting platform rather than a new foundation model or a standalone universal work agent. Its November 3, 2026 launch brings existing Claude and ChatGPT agents into ongoing calls. The company positions the experience beyond post-call summaries: an assistant can read the live transcript and meeting chat, inspect shared screens and begin preparing work before the meeting has ended.
The product should also be distinguished from the older Tana Outliner. In the meeting experience, participants use their existing agent account, connectors and skills. This provides relevant working context but also creates a trust boundary between meeting information and whatever external systems a personal agent can access. A guest or employee joining a call does not by itself determine the authorization of those connected applications.
Capabilities & workflow
Tana Agentic Meetings — Capabilities & workflow
Tana provides native meetings, transcripts, documents, shared chat and private responses from attached agents. According to the company, Claude and ChatGPT can work from what is said and shown during the call. Examples include preparing a customer follow-up, researching metrics, creating a task or drafting a document. The underlying agent must already have the relevant skills and connectors; the mere presence of an example on the product page is not evidence of universal access.
External agents operate through Tana's desktop application on macOS, Windows or Linux, alongside an installed Claude or ChatGPT client and an eligible account. Guests can join through a browser but do not automatically bring their own agent. The launch names Claude and ChatGPT as supported integrations at that time, while other assistants remain on the roadmap. Availability and subscription requirements should be checked for each participant.
Practical use cases
Tana Agentic Meetings — Practical use cases
A low-risk pilot can use a meeting with fictional business details to draft a decision memo. An organizer compares the draft with the actual transcript, corrects the responsible owners and deliberately shares the final output in the meeting channel. This reveals whether the assistant resolves ambiguities correctly and whether private suggestions remain private. No messages should be sent to external recipients in the first test.
A second experiment can ask the agent to prepare a ticket as a draft in a test project. The advantage is avoiding the need to restate the conversation in another tool. The risk is that an assistant may misinterpret who promised what or expose sensitive details from a screen share. Keep payment systems, real customer records and live source repositories outside the initial permission scope, even if connectors technically exist.
Availability, pricing & limits
Tana Agentic Meetings — Availability, pricing & limits
Tana says guests may join meetings in a browser, but personal Claude and ChatGPT integrations require the desktop app and the corresponding locally installed agent client. Having access to a Tana meeting does not remove external AI subscription costs or imply unlimited transcription quotas. The platform's own plans include meetings with differing allowances. Enterprises should confirm seat limits, usage, meeting retention and agent eligibility in their actual account.
The company states it captures conversations only when recording is started. Organizations still need appropriate consent and notice for the participants and jurisdictions involved. Tana advertises an EU processing and storage option and settings that control meeting retention, but these options require configuration and do not establish where every connected external model processes its data. The interface is English, although multilingual transcription and output-language settings are described.
Privacy, security & trust
Tana Agentic Meetings — Privacy, security & trust
Tana makes an important distinction between a user's private Agents tab and the shared Meeting tab. It describes proposed actions that require a person's review before changes are applied in Tana, and the user chooses what to share with the room. This is useful, directly scoped product behavior; it does not prove that all actions performed through an external Claude or ChatGPT connector are independently blocked by the Tana approval interface.
The provider advertises SOC 2 Type 2, an EU hosting option and no training on meeting data. These are vendor claims, not independently inspected audit evidence for this review. More importantly, organizations must evaluate the separate data path to the external AI account and its tools. Screen shares, participants' comments and transient transcripts may be sensitive even when the final shared document is harmless. Meeting permissions and tool permissions need independent treatment.
See AI agent security for the risks of external tool permissions and sensitive context.
Comparison & decision
Tana Agentic Meetings — Comparison & decision
Compared with a conventional transcription tool, Tana aims to have agents act while the call is in progress, not simply produce a summary afterwards. Compared with a generic chatbot, it provides live meeting context without manually copying every relevant comment and screen. That closeness to the discussion can improve timeliness but also increases the risk of a tool taking ambiguous conversational language as an instruction.
Tana is worth considering where teams routinely leave meetings with substantial follow-up work and already use compatible personal agents. A simpler meeting recorder may be enough when no external action is required. Highly sensitive or regulated meetings require a separate assessment of consent, retention, guest access and third-party model processing. AgentenCode's fit judgment is based on documented behavior rather than comparative benchmark testing.
Relevant comparisons: Compare the meeting context with Claude Cowork as a separate work agent and Notion Custom Agents for knowledge workflows. Their data and approval boundaries differ.
Editorial deep dive
Tana Agentic Meetings — Editorial deep dive
Start with three restricted meetings containing only synthetic data. Verify exactly when recording begins, what the attached agent sees, who can view a private suggestion and whether a mistaken task assignment is reversible before being shared. Intentionally show sensitive-looking but fictitious information during a screen share to test the group's review discipline. Track whether responses cite the correct exchange rather than simply sounding convincing.
Only after the first review should a connector prepare one non-production action behind explicit approval. Document local client status, external account permissions, prompt and tool traces, what Tana can and cannot revoke, and how output is removed from the shared meeting space. The vendor's demonstrations are not independent reliability statistics. This profile offers a source-backed initial assessment and a test design, not certification that all meeting-agent actions are safe.
Frequently asked questions
Frequently asked questions — Tana Agentic Meetings
Is Tana itself an AI foundation model?
No. It is a meeting product with built-in assistance and integrations for existing Claude and ChatGPT agents.
Can anyone bring an agent through the browser?
No. Guests can join a meeting in a browser, while bringing a personal external agent requires the desktop application.
Does Tana approve every external connector action?
Its documented approval behavior concerns Tana actions and private sharing; each connected agent's external privileges must be evaluated separately.
Original sources & method
Original vendor sources and evidence scope
Product claims are limited to the linked first-party sources and their stated scope. No independent hands-on test, audit, or compliance certification was performed. Unverified controls remain unknown, not negative findings.
- Tana – Claude Code and ChatGPT live in meetings (3 Nov 2026) — https://tana.inc/blog/your-own-claude-or-chatgpt-in-the-meeting
- Tana – Tana Calls product information — https://tana.inc/tana-calls
- Tana – Official product and deployment boundaries — https://tana.inc/