Claude in Slack became Claude Tag on August 3, 2026. If you are looking for a Claude Tag alternative, you have probably hit one of three walls. The bill is a seat plus Opus 4.8 by the token, with a spend cap you have to set yourself. One shared @Claude holds every credential and acts for everyone in the channel. Or your company runs Microsoft Teams and Tag is a Slack-only beta.
This page starts from the bill and the identity, because those two things sort the alternatives better than any feature list. Five products answer the same question Tag does, a shared agent that lives in your chat and acts for the team, with a different bill or a different identity. Four things that appear on other lists are a different job. We build one of the five, Chickpea, so read that section knowing that. Every price below carries the date it was checked.
Key takeaways
- Claude Tag has no price of its own. You pay for a Claude Team or Enterprise seat, now from two seats at $20 a seat billed annually or $25 monthly, plus Opus 4.8 by the token for everything the one shared @Claude reads and writes.
- Five alternatives answer the same question with a different bill or a different identity: Junior, Viktor, PromptQL Tag, Runbear, and Chickpea. The table below has each one’s price and the date we checked it.
- Four things on other lists are a different job: Slackbot, the open-source clones, personal agents like OpenClaw, and support desks like eesel or Ada.
- Stay on Tag if your company is on Slack, uses Claude models by policy, and has set a spend cap it can live with. The last section says when.
What Claude Tag charges for
Anthropic’s help center describes the billing in one sentence: “spend is based on usage rather than the number of people.” The pricing model has three parts.
The seat. Claude Tag is “available on Team and Enterprise plans in beta.” Team seats are $20 a month billed annually or $25 monthly, Premium seats $100 or $125, and claude.com sells the plan “for teams of 2 to 150.” Enterprise is a seat price plus usage at API rates. The ten-seat minimum that most alternatives pages still carry, and that our own comparison page carried until today, is stale.
The meter. Tag runs Opus 4.8, and there is no published Tag rate, so we use the API rate, $5 per million input tokens and $25 per million output. A task that reads about 40,000 tokens of channel context and writes 6,000 costs around 35 cents. Two hundred of those a month is about $70 on top of seats. Two thousand is about $700. Work done in a channel bills the organization; work done in a DM bills the person who asked.
The cap. Admins can set a hard limit for the organization and one per channel, with alerts at 75 and 95 percent, and Tag declines work past the limit rather than billing it. What the default is depends on who you ask. TagCoworker’s guide lists the default at $1,000. One Team admin on Hacker News wrote that “every new feature has metered usage and ‘unlimited spending’ (aka no limit) enabled by default for our team org. So if I activate something (Claude Code, Claude Tag) and don’t actively go to the usage page to set a spending limit, there is no limit.” Either way, set the cap before you turn ambient mode on.
For a ten-person team on monthly billing, the arithmetic looks like this.
| Line | Amount | Source |
|---|---|---|
| Ten Team seats | $250 a month | claude.com/pricing, September 8, 2026 |
| 200 tasks at about $0.35 | about $70 | Opus 4.8 API rates, tagcoworker.com |
| 2,000 tasks | about $700 | same |
The meter is what people notice first. “One thing that wasn’t obvious in the post: it’s usage (token) based pricing, not including in your existing Claude subscriptions,” one reader wrote under the launch announcement. Another, a month later: “Claude Tag seems awesome, but a single query can run over a dollar.” The sharpest note came from a team that turned on ambient mode: “We spent $2 on just a few turns, doing basically nothing because Claude kept picking up on our messages as if we had summoned it.”
For the team Tag was built for, that bill is fine; Anthropic says Tag opens most of its own company’s pull requests. For a team that wanted a flat number, the meter is the reason to look elsewhere.
Who Claude Tag acts as
The second reason people leave is identity. Tag is one @Claude. Administrators “specify which tools and information the model should have access to, in which channels,” and Anthropic describes “separate Claude identities for different uses” scoped per channel. Those identities are real. They still sit under one organization identity that holds every connected tool, and only a Primary Owner or Owner can configure it. The Admin role cannot touch it. There is no approval step before an action, and with ambient mode on it joins conversations without being mentioned.
The complaints on the launch thread were about this. “To go after the non-developer side though they need to do a ton more work on governance & RBAC,” one commenter wrote. “You can’t even selectively enable or disable Cowork & Claude Code per member on the team plan, it’s all or nothing for everyone in the team.” Another put the platform problem in one line: “90% of F500 companies are on Teams, not slack.” Tag “works in Slack today,” in the help center’s words, and Anthropic has not announced a Teams version.
So the alternatives sort on four questions. Does it run in Slack or Teams? Can you pick the model? Who does it act as, and does it ask before it acts? Whose cloud does it run in, and who holds the keys?
How we sorted the alternatives
Same question means a shared agent that lives in your chat tool, connects to your systems, and does work for the team on request or on a schedule. Tag is one. Different job means a personal assistant, a support desk, a workspace that replaces Slack, or a summarizer, all of which show up on Claude Tag alternative lists because the keyword is new. We list the second group too, so you can rule them in or out. The keyword is ten weeks old and every page ranking for it is a vendor listing itself first, which is why the sort matters more than the order.
Five that answer the same question
| Option | Platforms | Models | Acts as | Asks first | Runs where | Pricing model | Price, checked |
|---|---|---|---|---|---|---|---|
| Junior | Slack, Teams | Claude family | one or more Juniors | yes, approval-gated | vendor cloud | credits, $100 steps | $100 to $20,000 a month; trial $100 credit on Sonnet only, Sept 8 |
| Viktor | Slack, Teams | vendor picks; OpenRouter key optional | one @Viktor per workspace | per-run approvals | vendor cloud | credits | $50 for 20,000 credits, quick task 100 to 300, Sept 4 |
| PromptQL Tag | Slack, Teams | any, incl. open-weight | the human who tagged it | vendor claim | vendor cloud, or your cloud on enterprise | usage | $0.14 per OLU, Aug 11 |
| Runbear | Slack, Discord, Teams | Claude, GPT, Gemini; BYOK on Business | named agents (@Support style) | not stated | vendor cloud | flat plus credits | $99 a month for 20,000 credits, Sept 8 |
| Chickpea | Slack | Anthropic, OpenAI, OpenRouter, Workers AI, per teammate | named teammates with own handles | yes, on sends, deletes, publishes | your Cloudflare account or Node host | free software | $0 plus your provider at list price, Sept 8 |
Junior
Junior is the approval-gated one. It runs in Slack and Teams, and its pitch against Tag is a single line on its own page: it “asks before it acts.” The trial is one Junior on Claude Sonnet with a $100 credit; the Standard plan runs from $100 to $20,000 a month in $100 steps, buys all models including Opus, and allows up to five Juniors and twenty concurrent sessions. Credits are “tied to compute units (AI tokens),” so it is a meter too, with a monthly ceiling you choose. Pick it over Tag if you need Teams or a human checkpoint before every external send. It is a hosted service, and the vendor holds the credentials, which is the same trade as Tag with a different vendor.
Viktor
Viktor installs from the Slack App Directory and runs in Slack and Teams, one @Viktor per workspace. It has 27 native integrations plus thousands more through managed connectors. It hands back PDFs, spreadsheets, and deployed web apps. Credits start at $50 a month for 20,000, and a quick task costs 100 to 300 of them, so the bill is a flat floor with a meter above it. When we ran the onboarding ourselves in August, a one-person workspace had to verify a card before the starter credits unlocked. Pick it over Tag for install speed and Teams. It is one shared identity on a vendor cloud, which is the identity question unchanged. We wrote up where it beats Chickpea.
PromptQL Tag
PromptQL Tag, from Hasura, answers the identity question differently. It “acts with permissions of the real human tagging it,” so what the agent can reach depends on who asked, and an auditor can see which person’s permissions each action used. It runs in Slack and Teams on any model, including open-weight ones, and the enterprise tier runs in your own cloud. Helio’s August comparison lists the price at $0.14 per OLU, a usage unit of Hasura’s. Pick it over Tag if per-user permissions and your own cloud are the requirement and a usage meter is fine. Of the five, it is the only one that changes the identity model rather than the bill, which is why it is here even though these are the vendor’s claims from its own comparison page and we have not run it.
Runbear
Runbear gives you named shared agents, @Support and @HR style, in Slack, Discord, and Teams, on Claude, GPT, or Gemini, with your own keys on the Business plan. Runbear prices the opposite way from Tag. It is $99 a month for 20,000 credits and five builders, $399 for 100,000, and one credit is one model call, so quick answers cost about five and deep work hundreds. The bill is flat with a ceiling, which Tag does not offer, and it is the reason to pick Runbear if you want a monthly number you can put in a budget. We have not run it, and the pricing page does not say whether agents get approval gates.
Chickpea
We build Chickpea, so this entry is longer on mechanism and shorter on adjectives. It gives a workspace named teammates, each a Slack user group with its own handle, instructions, memory, connected accounts, and model. @support can hold the Zendesk account while @revops holds HubSpot, and neither can reach the other’s. Credentials never enter model context; the gateway attaches the secret at the network edge. Sends, deletes, publishes, and bulk changes wait for a person unless the teammate’s saved instructions say otherwise. It reads nothing in the background; only a mention reaches a model. It deploys to your own Cloudflare account, on the Free plan, or to a Node host, and it is Apache 2.0. The bill is your model provider at list price.
The part I would not give up is the model pin. A triage teammate on a small fast model costs cents a day, the one that writes code review can run a frontier model, and you can move a teammate to another vendor without changing the product. Tag is Opus 4.8 for everything.
Where Tag is better. Tag reads whole conversations and joins unprompted, where Chickpea answers mentions and nothing else by design. Tag hands work to Claude Code; Chickpea reads repositories and runs an optional coding sandbox, and does not hand work to Claude Code. Anthropic hosts Tag; you deploy Chickpea.
Four that are a different job
Slackbot. Salesforce relaunched it in January 2026 as “your personal agent for work,” included at no extra cost on Business+ and Enterprise+. It is one agent per person, and it does not need a new vendor. If what you wanted from Tag was summaries and recaps, Slackbot is the cheaper answer.
The open-source clones. Three repositories call themselves the open-source Claude Tag, and they are three different things. CopilotKit’s OpenTag, MIT and about 1,200 stars, is a self-hosted triage bot for Slack and Teams, OpenAI-first, with a Linear issue “behind an approval gate.” Anil-matcha’s open-claude-tag, MIT and 11 stars, is a channel agent for Slack on any model through LiteLLM, four setup steps. fancyboi999’s open-tag, Apache 2.0 and 189 stars, runs as its own Slack-style workspace where coding agents are teammates, with no Slack app involved. Two are small and one is a different product. If open source is the requirement, those are the choices, and Chickpea is the fourth.
Personal agents. OpenClaw and Hermes Agent put one agent into Slack from a machine you run, where it reads files, runs shell commands, and browses. Their own docs describe one trust boundary per gateway, which is the problem we wrote about in OpenClaw alternatives for teams. They cover the personal half of what Tag does.
Support desks. eesel, Ada, and Freddy price per ticket or per resolution and live in your helpdesk. eesel’s own comparison says it: Tag “answers in a thread; it doesn’t run your Zendesk queue.” If the job is closing tickets, buy a ticket tool.
When to stay on Claude Tag
Stay if your company is on Slack and would not use a Teams version anyway, and if Claude models are the policy, because Tag on Opus 4.8 is the best-supported way to have them in a channel. Engineering-heavy teams should stay too. The Claude Code handoff is the point, and Anthropic runs its own product team on it. If you want Anthropic to host it and carry the operational risk, that is also Tag. Set a spend cap first. If the bill under it is fine, there is nothing to fix. Tag reads whole conversations, remembers across days, and logs everything @Claude did and who asked. None of the five above does all of that.
Leave if the meter is the problem, if one shared identity is the problem, or if Teams is the problem. Each of the five fixes at least one of those, and the table says which.
FAQ
How much does Claude Tag cost?
There is no Tag price. You pay for a Claude Team plan, from two seats at $20 a seat billed annually or $25 monthly, or Enterprise, and then Opus 4.8 by the token for the work Tag does, with admin caps per organization and per channel. A task that reads 40,000 tokens and writes 6,000 is about 35 cents at API rates.
Is Claude Tag free? Can I use it on Pro or Max?
No. It is in beta on Team and Enterprise plans only, and every plan bills the work it does by the token.
Does Claude Tag work with Microsoft Teams?
Not as of September 2026. The help center says Tag “works in Slack today,” and Anthropic has not announced a Teams date. Junior, Viktor, PromptQL Tag, and Runbear all run in Teams.
What happened to Claude in Slack?
Anthropic replaced it. Claude Tag launched June 23, 2026 with a 30-day migration window, and Claude in Slack switched over to Tag on August 3, 2026.
Is there an open-source Claude Tag?
Three projects use the name, and only two are Slack apps: CopilotKit’s OpenTag, about 1,200 stars, and Anil-matcha’s open-claude-tag, 11 stars. Chickpea is the fourth, Apache 2.0, built for named teammates in Slack rather than one agent.
I want Claude to remember my work across sessions. Is that Claude Tag?
No. Claude Code’s memory does that; Claude Code keeps per-project files and reads them alongside CLAUDE.md. Claude Tag is the shared agent in Slack. People confuse the two because Tag also keeps memory, but Tag’s memory is per channel and belongs to the organization.
Sources
Anthropic: What is Claude Tag? (help center, checked 2026-09-08), Introducing Claude Tag (2026-06-23), Plans and pricing (checked 2026-09-08). Pricing analysis: eesel, Claude Tag pricing (2026-07-17), tagcoworker, seat cost and token billing (June 2026), eesel, Claude Tag alternatives (2026-07-17). Users: rdli, isusmelj, thewebguyd on governance and on Teams (Hacker News, 2026-06-23), octember (2026-07-04), areganti on LinkedIn (June 2026). Alternatives: Junior pricing and Junior on Claude Tag (2026-06-30), Chickpea vs Viktor (2026-09-04) and our August onboarding run, Anthropic’s webinar on how it uses Claude Tag, PromptQL Tag vs Claude Tag (2026-07-04), Helio’s comparison (2026-08-11) for the PromptQL price, Runbear pricing, Salesforce on Slackbot (2026-01-13), OpenTag, open-claude-tag, open-tag (all checked 2026-09-08), Stilla AI alternatives after the Meta acquisition (2026-09-09). Chickpea: security model, authority and confirmation, requirements and costs, manage from Slack (verified 2026-09-04 to 2026-09-08), Chickpea vs Claude Tag (2026-09-08).
Facts checked September 8, 2026. Claude Tag is in beta and its plan terms have already changed once since launch. We re-check the register behind this page every 90 days.