Move your Claude in Slack setup to an open-source agent

What is Claude Tag?

Claude Tag is Anthropic's Slack agent, and it is what replaced Claude in Slack. Tag @Claude in a channel and it works the task in that thread. You can also message it directly, or open it from Slack's AI assistant panel. It runs on Team and Enterprise plans in beta, in Slack only for now.

The line to pause on sits in Anthropic's support article. In a channel, Claude acts under your organization's identity. It uses the tools and access an admin set up for that channel, and the work is billed to your organization rather than to the person who asked. Direct messages and the assistant panel work differently: there Claude has the capabilities enabled in the person's own Claude account.

The switch-over, dated

"Claude in Slack switched over to the new Claude Tag experience on August 3, 2026," the same support article says. The launch post calls Claude Tag the replacement for the existing Claude in Slack app, and says administrators could opt in within 30 days. The Claude Code documentation closes the overlap in a troubleshooting notice: "The legacy Claude in Slack bot is retired effective October 5, 2026 and no longer responds in channels."

Two months separate the rename from the retirement, and the second date decides what a workspace is running. An answer in a channel today comes from one of two arrangements. It is either Claude Tag under the organization, or a Claude Code session under the account of the person who asked. Where nobody opted in and nobody linked an account, the bot goes quiet. Where the old version is still selected, Slack replies with the retirement notice, and that notice names the three places the setting can sit: the channel, the workspace, or the organization default.

What an organization's identity changes for an admin

The old app asked a person to connect their own account. The Claude Code docs still describe that path: each session runs under your own Claude account, with your connected repositories and your plan limits. Claude Tag moves the unit of work up a level. The agent is provisioned once for the organization, acts with credentials an admin connected, and spends the organization's budget.

Setup therefore turns into an ownership decision. Only a Primary Owner or Owner can set up Claude Tag's access and channels, and the support article says the Admin role cannot. Access is granted at three levels, each inheriting the level above. Organization-wide credentials and repositories apply everywhere the agent is installed. Workspace access covers every public channel in a Slack workspace. A private channel can carry extra credentials and repositories, which is how a legal channel keeps its tools and memory apart from engineering.

Memory moves with it. Claude Tag keeps context per channel and per workspace instead of dropping it after each task, and Owners can view, edit and delete that memory. An Activity page lists every scheduled and one-time task across the organization plus every network call made under the agent identity. Only Owners can open it. Commits and pull requests show the Claude GitHub App as the author, with a link back to the thread that started the work.

Cost becomes a setting too. Spend is consumption-based rather than per seat, with an organization-wide hard cap, optional per-channel limits, and alerts at 75% and 95%. Work that would cross a limit is declined rather than trimmed. Channels that include Slack guests get their own switch, and it blocks Claude from those channels by default.

Six questions to ask any Slack agent

Ask these of Claude Tag, of a coding assistant wired into Slack, and of anything else someone wants to invite to a busy channel.

1. Whose identity does the agent act under, and whose name appears on the action afterwards?
2. Who can widen its reach, and can one channel quietly receive more than the channel beside it?
3. What can it change without a person in the loop, and where does the approval show up?
4. Where does finished work land, and can a reviewer read it as a diff?
5. Whose model answers, and who decides when that model is swapped?
6. What does a busy channel cost a month, and who can raise the ceiling?

Where Kortix fits

Kortix runs that job from the other side. Slack and Microsoft Teams are live. Email is experimental, per project. What a company-hosted Slack agent changes walks through the mechanism. For a migration the narrower point is that every channel carries its own binding, and the three settings an admin has just chosen in Claude Tag's console exist here as well: the agent, the model and the join policy, set from a command, from chat or from the Channels page.

Where Claude Code in Slack fits

The coding path is documented separately, and it is worth reading on its own terms. When a message is routed to Code, Claude creates a Claude Code cloud session. That session runs under the person's own Claude account, with their connected repositories and their plan limits. The docs say it works in channels only, not in direct messages. The thread gets status updates, and the finished session offers View Session and Create PR.

For a team that already pays for Claude plans, that is a useful surface for small fixes and reviews raised in a thread. It sits beside Claude Tag, and it does not give a company an agent provisioned once for the organization.

A migration checklist

1. Find the Primary Owner or Owner. They are the only people who can turn Claude Tag on, and the Admin role cannot.
2. List every channel where the old bot was invited, then decide channel by channel whether an agent with organization reach belongs there.
3. Rebuild tool and repository access at the three levels, and move sensitive work into private channels with their own credentials.
4. Choose the member access mode and the guest-channel setting before anyone tags the bot.
5. Set the organization cap and per-channel limits, and tell the people who will hit them.
6. Read the Activity page and the stored memory, and delete what should not persist.
7. Decide where coding work goes: the Code routing under a person's plan, or an agent you host with its work landing in your own repo.

What the choice comes down to

Claude Tag is a capable teammate, and its beta status on Team and Enterprise is worth planning around. The trade is ownership. Its runtime, its memory and its model sit on Anthropic's platform, on Claude models, in Slack alone as of October 2026. A company that wants the agent's skills and its memory inside a repo it controls has to run the agent itself. Approval gates you set. Off until you set them. Run it on Kortix Cloud, in your VPC, or on your own on-prem network. Self-host is free.