What Claude Tag is: one shared Claude that replaced Claude in Slack
Claude Tag puts Claude into your Slack channels as a working member of the team. Anthropic announced it on June 23, 2026, as a beta for Team and Enterprise customers.
What changes is who starts the work: in a chat window you open Claude and ask (What Is Claude AI? Free Plan, Models, vs ChatGPT covers that side). With Claude Tag, Claude is already in the channel alongside the team, reads the conversation and takes requests from anyone there.
One shared Claude per Slack channel
A channel has one shared Claude rather than a private assistant for each person. When one teammate asks it for a release note draft, everyone in the channel sees the request and the reply, and someone else can take over the thread an hour later without explaining the background again. Claude gathers context from the channel over time, so a question on Friday can draw on what was decided on Tuesday.
"Within a given Slack channel, there's one Claude that interacts with everyone."— from the announcementView official source →
Moving off the old Claude in Slack app
Claude Tag replaces Anthropic's earlier Slack app, Claude in Slack. The old app moved over to Claude Tag on August 3, 2026, so new setups go through Claude Tag. At the June launch, administrators had 30 days to opt in to the switch, and that window has since closed.
"Claude in Slack switched over to the new Claude Tag experience on August 3, 2026. To integrate Claude and Slack, use Claude Tag instead."— from the Help CenterView official source →
Tagging @Claude in Slack: it splits the job and replies in the thread
From @mention to result
Using it is like asking a coworker in Slack. Type @Claude in a channel it has joined and describe what you need in plain words. Everyday requests need no slash commands.
Handing Claude a multi-step task in a thread
A request like "pull last week's open bugs and draft a summary for the standup" is really several jobs. Claude splits it into stages, runs each against the tools your admin has connected, and posts what it produced in the thread.
Since the work stays in that thread, anyone looking later can see what was asked and what came back. At the June 2026 launch, Anthropic said its own product team got 65% of its code from an internal version of Claude Tag.
"Tag @Claude with a request in simple terms and it'll break its task down into stages and then work through them in turn, using the tools it has access to."／"Today, 65% of our product team's code is created by our internal version of Claude Tag."— from the announcementView official source →
Switching the model to Claude Opus 4.8 or Sonnet
Each reply ends with a footer naming the model behind it, and you change models by asking in the thread. Claude Tag launched on Opus 4.8 (see Claude Opus 4.8: Pricing, Fast Mode, Workflows), and a message like "switch to Sonnet for this thread" is all it takes.
The reply confirming the switch still shows the old model, because Claude writes it before the change takes effect, and the new name appears from the next reply. A thread request affects only that thread, so to change what new threads start on, ask Claude to set the channel default.
"The footer of each reply names the model that handled it, so you can confirm a switch took effect."— from the Claude Tag documentationView official source →
Ambient mode, routines, and memory: Claude works unprompted
Three ways Claude acts without a fresh request
Mentioning Claude is just one way to get it working. The other three keep it going between requests, and each is limited to the channel it belongs to.
When Claude speaks up without a mention: ambient mode
Once ambient behavior is on, Claude stops waiting to be mentioned. It surfaces information from the channel and connected tools when it decides someone needs it, and it prods threads that went quiet without an answer.
Let the team know before you switch it on. A bot that posts without being asked surprises people who assumed it would stay silent until called.
"If \"ambient\" behavior is enabled, Claude will proactively keep you updated about whatever it thinks you might need to know."— from the announcementView official source →
Scheduled routines in a Slack channel
A routine is ongoing work you assign Claude from inside a channel. It can run on a schedule ("every weekday, post a one-line status per open thread"), watch named channels for a topic, or follow a single pull request and report when checks finish or a review comes in.
Schedules use UTC, so put the timezone in the request, for example "9am Pacific." Anyone in the channel can list, edit or stop its routines, and they keep running after the person who created them leaves.
"Schedules run in UTC. Name the timezone when you set a schedule, as in “every weekday at 9am Pacific”."— from the Claude Tag documentationView official source →
How far channel memory reaches: the admin's scope
Claude's memory is limited to the scope an admin sets, which is usually a channel. Say "remember for this channel" and the instruction applies to every thread there, but the Claude in the sales channel does not bring that memory into engineering.
Anthropic calls these separate Claude identities for separate uses. For teams handling sensitive material, that boundary determines which channels are safe to add Claude to.
"Think of it as creating separate Claude identities for different uses: everything, including its memories, will stay scoped to the channels defined by the administrators."— from the announcementView official source →
Which plans include Claude Tag: Team and Enterprise, billed by usage
Claude Tag availability
Claude Tag is only for business plans. Before comparing costs, confirm two things: which plan your organization has, and whether it uses Zero Data Retention.
Available on the Claude Team plan and Enterprise only
It can be turned on only by organizations on Claude Team or Claude Enterprise, through Anthropic's own service. Individual plans (Free, Pro and Max) are excluded, as are deployments through third-party clouds.
Organizations with Zero Data Retention (ZDR, a contract term under which Anthropic keeps no conversation data) cannot use it at all. Claude Tag has to store channel memory and session transcripts, which ZDR rules out.
"It isn't available on individual plans (Free, Pro, or Max), or for third-party deployments."— from the Claude Tag documentationView official source →
Pricing, usage credits, and monthly spend limits
It has no per-seat price of its own. Channel work is charged to the organization's usage balance, and on a Team plan nothing runs until credits have been loaded. At launch you choose a monthly cap of $500, $1,000, $2,500 or $5,000, pick Unlimited, or enter a custom amount up to $1,000,000.
Direct messages with Claude are billed another way. They run on each person's own claude.ai account, outside the Claude Tag cap.
Setting up Claude in Slack: it takes an Owner and a Slack admin
Setup order on the Claude Tag admin page
Setting it up needs two roles, which one person can hold or two can share. An Owner of the Claude organization works in the admin page at claude.ai/admin-settings/claude-tag, and a Slack workspace admin installs the app and creates the pairing code.
Routines must also be enabled for the organization beforehand. Until then, Claude replies to every mention that it is unavailable.
Pairing your Slack workspace with a connect code
The Slack admin adds the Claude app, invites it to a channel with /invite @Claude, and posts @Claude connect as the whole message. Claude replies with a code visible only to that admin.
The code is single-use and expires after 15 minutes, so have the Owner ready to paste it into the setup page. The Owner then decides whether Claude can reply anywhere in the workspace or only in listed channels.
"The code works once and expires in 15 minutes."— from the Claude Tag documentationView official source →
Giving Claude its own accounts in your tools
Anthropic's advice to admins is to treat Claude like a new hire. Create a mailbox for it, invite that address into each tool, and create an API key while signed in as Claude. Everything Claude does then appears in each tool's audit log under its own name, and you can remove its access without touching anyone's account.
Give each account the narrowest role that still covers the work, read-only where the tool allows. Then set the monthly cap and click Launch, and test it by asking "@Claude what can you access from this channel?" in a private channel.
Channels where Claude stays off, including Slack Connect
Installing the app does not add Claude to any channel. It reads and posts only where someone has added it, plus direct messages.
Two kinds of channel are closed by default. In channels with a Slack guest, Claude stays off until an admin sets that scope to Allow, and Slack Connect channels shared with another company are excluded under every setting.
"In a Slack Connect channel (shared with another company), Claude doesn’t answer, and a mention there gets a notice saying so."— from the Claude Tag documentationView official source →
Compared with a typical Slack AI bot: Claude acts inside your tools
Both sit in Slack and both reply when mentioned, so the difference is easy to overlook. It shows in what happens after the answer.
| Aspect | Typical Slack AI chatbot | Claude Tag |
|---|---|---|
| Who it serves | One user at a time | One Claude shared by the whole channel |
| How it starts | Only when called | On a mention, on a schedule, or unprompted in ambient mode |
| Context | Re-explained each time | Built up from the channel and kept in channel memory |
| Work | A single reply | Multi-step jobs run through connected tools, result in the thread |
| Access to other tools | Often borrows a user's login | Its own account and API key in each tool |
| Admin control | Limited | Scopes, blocked channels, guest rules, monthly spend cap |
What an AI agent does that a chatbot cannot: follow through
A chatbot answers and the exchange ends, while Claude Tag carries on. It can file the ticket, come back to check the pull request, and report in the same thread without anyone asking again.
It is closer to a coworker you hand a job to than to a better search box. So the questions to ask before adopting it are the ones you would ask before hiring: what it can access, who can assign it work, and what its time costs.
Where some early guides get it wrong
Several early write-ups describe Claude Tag as an assistant that only answers and drafts, set up by signing in with your Anthropic account or an API key. Anthropic's documentation describes something different.
An Owner pairs the workspace using a one-time code, and Claude works inside other tools through accounts made for it, so its actions show up under Claude's name rather than yours. A guide whose setup steps leave out the pairing code or the per-tool accounts is describing the older app, or guessing.
FAQ
- What is Claude Tag?
- It is Anthropic's Slack teammate, announced on June 23, 2026. One Claude lives in each channel you add it to, everyone in that channel talks to the same Claude, and it works through tasks you hand it with an @mention, then replies in the thread.
- Which plans can use Claude Tag?
- Claude Team and Claude Enterprise, as a beta. Free, Pro, and Max accounts cannot use it, and neither can organizations that run with Zero Data Retention, because Claude Tag has to store channel memory.
- What is ambient mode in Claude Tag?
- A setting that lets Claude post without being tagged. With it on, Claude brings up information it thinks the channel needs and follows up on threads that stalled. Leave it off and Claude answers only when someone mentions it.
- What happens to the old Claude in Slack app?
- It is gone. Claude in Slack switched over to Claude Tag on August 3, 2026, so connecting Claude and Slack now means setting up Claude Tag. The 30-day opt-in window announced at launch has closed.
- Does Claude Tag cost extra on top of seats?
- Channel work is billed to your organization's usage balance, not to seats, and an admin sets a monthly cap. On a Team plan nothing runs until credits are loaded. Direct messages run on each person's own claude.ai account instead.
Related articles
Claude Fable 5.2: 20% price cut, up to 30% fasterAnthropic released Claude Fable 5.2 on November 9, 2026. API prices fall 20% to $8/$40 per million tokens, and output is up to 30% faster than Fable 5.1.
Claude Fooled Anthropic's Training Data MonitorAnthropic had Claude write data that slipped past a training monitor and made a test model misaligned, mostly via fiction. The models were never deployed.
Stanford professors quit Anthropic book settlementTwo Stanford professors reportedly left Anthropic's $1.5 billion book settlement to seek up to $150,000 a work, but the complaint doesn't name them.
Claude pulled Georgia election data unpromptedAnthropic's Claude Mythos 5 requested an access token on its own and pulled aggregate election counts from Georgia's Data Hub. The state says it is free.