What Is a Claude Bot: Three Things the Phrase Can Mean, and How to Tell Them Apart

Anthropic publishes no product called a Claude bot. There is no Discord bot, no Telegram bot, and no generic bot you install somewhere and point at a workspace. What Anthropic does publish is a set of products with specific names — Claude Code, Claude Cowork, Claude Duet, Claude Tag, Claude Science and Claude Security — and the word bot appears in none of them. That absence is the first useful fact in this piece, because almost everything confusing about the phrase follows from it.

The second useful fact is that bots built on Claude do exist, in numbers, and none of them is Anthropic's. A developer can call Claude models through Anthropic's API and stand that call behind any chat surface that accepts a bot. That product belongs to the developer. Anthropic supplies the model, and very little else about the arrangement is Anthropic's to describe.

So the question has three honest answers, and they are different enough that picking the right one decides what you do next. This piece separates the meanings, gives you the checks that tell them apart, and is specific about credentials, because that is where the question stops being academic.

The word is doing three jobs at once

Bot is a chat-era word. It describes software that holds a place in a conversation: an account that answers, a service that reacts to a message, a participant that is not a person. That is useful for describing what you see in a channel and useless for saying who made the thing, who runs it, or what it can reach.

Anthropic's products do not use the word, and their names do not follow a pattern that lets you guess. Code is for engineering, Cowork for knowledge work, Duet is proactive, Tag lives in Slack, Science is a desktop research app, Security scans. A person who has seen one of these and is looking for another has no vocabulary for it except the generic one, which is why the phrase gets typed into a search box.

Read literally, it could mean any of three things: one of Anthropic's own surfaces, described in the user's own words; an integration somebody else built on Anthropic's API; or a product that uses the name with no relationship to Anthropic at all. The categories are not academic. Each implies a different operator, a different set of terms, and a different answer to the question that matters when something goes wrong, which is who you ask.

The first meaning: a surface Anthropic actually publishes

Anthropic's navigation lists these product surfaces: Claude, Claude Code, Claude Cowork, Claude Duet, Claude Tag, Claude Science and Claude Security. Alongside them it lists capabilities rather than products — Artifacts, Design, Connectors, Plugins, Skills, Extensions, Claude in Chrome, Claude for Google Workspace and Claude for Microsoft 365. None of the sixteen is called a bot.

The closest match for most people asking the question is Claude Tag, which is Claude inside Slack. Its availability line is exact: available for Claude Enterprise and Team customers in Slack. If the thing you are looking at lives in Slack, works inside channels, and requires your organisation to be on Enterprise or Team, that is the official surface, and everything else about it is described on Anthropic's own product page.

Claude Code is the other one people arrive at sideways, and it is not a chat bot at all. Anthropic describes it as built for software engineering, running in a terminal, a desktop app, a web interface and a mobile app, with integrations for GitHub and popular editors. Claude Cowork is the general-purpose one, and its page is direct about the offer: give it a goal and it works across your files and tools, and you come back to polished work for review. The same page says it runs on web, desktop and mobile, works in folders you choose, opens a browser in its own side panel rather than taking over yours, and that deleting anything needs your approval.

Notice what these descriptions have in common. Each names a surface you go to, a plan that qualifies, and a job it holds. That is what an official product page does, and its shape is the first thing to compare against when you are looking at something that has only a name.

What Anthropic does not publish

Anthropic publishes no Discord bot, no Telegram bot, and no product called a Claude bot. This is worth stating flatly because the absence is easy to mistake for a gap in research. It is not. A vendor with a Discord integration lists it beside its Slack one. Anthropic lists one chat-platform product, in Slack, and it lists no bot anywhere.

There is also no single page titled Claude bot to read. The description in the section above had to be assembled from the product navigation and the individual product pages, which is a fair illustration of the problem: the phrase names a category the vendor does not use, so there is no canonical page that settles it. If you want an official answer, you have to start from a product name, not from the word.

And that is a statement about what Anthropic publishes, not about what exists. Bots built on Claude exist. That is the second meaning.

The second meaning: somebody else's product on Anthropic's API

Anthropic's developer platform lets software call Claude models. Any developer can wrap that call in a chat interface and ship the result as a bot on whatever platform accepts bots. When that happens the model is Anthropic's and the product is not. The operator is the developer, the hosting is theirs, the terms of service are theirs, the support queue is theirs, and what gets logged — and for how long — is their decision.

This piece will not name one. Naming a specific third-party bot in an explainer of this kind reads as a recommendation, and a vendor-neutral article is in no position to recommend an integration it cannot audit. What can be said without naming anybody is what the arrangement implies for the person deciding whether to use one.

Three things move when you cross from an official surface to a third-party one. Support moves: you are asking a developer, not a model vendor, and their ability to fix a problem ends where their code ends. Data moves: your messages now pass through an operator who is not Anthropic, under a policy Anthropic did not write. Billing moves too, in the common case where the operator charges for access to a model you could otherwise call yourself. None of those is disqualifying on its own. All of them are worth knowing before the bot is sitting in a channel with your team in it.

The third meaning: not Claude at all

The third category is the one that gives the phrase its bad reputation. It contains products and accounts that use the word Claude with no relationship to Anthropic: lookalike services, wrapper pages of unclear provenance, and material that borrowed a recognisable name in order to be found. The word is also an ordinary given name, which means a bot called Claude may be nothing more than a bot called Claude.

This category is hard to talk about because from the outside it can look identical to the second one. A legitimate third-party integration and a name-borrowing one both live on somebody else's domain, both describe themselves in their own words, and both may claim to use Claude models. You cannot settle the difference by reading the marketing copy, because the marketing copy is the thing in question. What you can do is check the things a real operator has to disclose and a borrowed name usually does not.

Four checks that separate the categories

The checks are ordered by how hard they are to fake. None of them needs technical knowledge, and all of them can be done on the page where the bot is offered, before you install anything.

• Domain. An official Anthropic surface is described on Anthropic's own site. If a page presents a bot as official and lives somewhere else, that is your answer regardless of how polished the page is. A third-party integration on its own domain is not pretending to be official — it is simply third-party — and that rules the first meaning out.

• Operator. A legitimate product names the entity that runs it: a company, terms of service, a support address. A bot offered by nobody in particular, with no terms and no contact, is asking for trust it has not described.

• Credentials. This is the sharpest of the four. An official Anthropic product signs you in the way Anthropic signs you in everywhere else. A third-party integration needs a key or a grant that you deliberately issue to it. No legitimate integration needs your Anthropic account password, and a bot that asks for one in a chat window is asking for your account rather than for access to a model. An API key is a bearer credential: whoever holds it can spend against the account it belongs to, up to whatever limits are attached to that key. Pasting one into a conversation is a different act from configuring it into software you control, and it is the act that turns a curious click into a bill.

• Terms. Whose policy governs what the bot does with what it reads? If that question has no answer on the page, then the bot's provenance is whatever you assumed it was.

Run the four in order and most candidates fall out at the first or second. What survives is a third-party product with a named operator, real terms and a credential model you can control. That is a decision you can make on the merits. What the checks do not do is turn an unofficial bot into an official one: the categories are about who operates the software, and diligence does not change that.

Why the question is being asked now

Google Trends puts worldwide interest in the exact query claude bot at an index of 32 in the seven days to 23 October 2026, up 20% on the period before (Google Trends, worldwide, past seven days, read on 23 October 2026). That is attention rather than adoption, and it is the kind of number that grows when a product family expands faster than the vocabulary around it does.

The vocabulary point is the real explanation. Six product names, none containing the word bot, none following a shared pattern, several of which sound like features rather than products. Someone who has watched an assistant answer in Slack, or seen a demo of an agent working through a folder of files, reaches for the only noun available. The vendor's naming and the user's naming have come apart, and the search query is where the gap shows up.

If what you want is a bot of your own

One reading of the phrase belongs to neither of the three categories: somebody who wants a bot of their own, in their own server, with their own rules. That is a build, and the build is more legible than the market is. You need a model behind an API, a surface to put it on, and credentials that stay under your control.

The model is the part that is easy to buy. OrcaRouter hosts models, not Anthropic's apps or anyone else's agent, so a Claude bot run by a third party is not something our endpoint produces. What the endpoint does is put 200-plus models behind one OpenAI-compatible interface and one key, at provider list price with no markup, with automatic failover moving a call when a provider has a bad minute. If you are assembling your own bot, that is the half of the problem that arrives solved.

The other half is what the checks above are really about. A key you issue to software you run is a key you can rotate, scope and revoke. A key you paste into somebody else's chat window is not, and good intentions on their side do not change the mechanics.

The short answer, and the one that matters

A Claude bot is not a product. It is either one of Anthropic's surfaces described in the wrong vocabulary, a third party's product built on Anthropic's API, or something that borrowed the name. Sorting a specific bot into those categories takes about a minute: the domain, the operator, the credential request, the terms. The question that decides whether any of it matters to you is the third one, because a bot that wants your Anthropic credentials is asking you to hand over your account and your spend along with the access.

Everything else follows from that. If it is official, the product page tells you what it does and which plans qualify. If it is third party, the operator's terms tell you what happens to what it reads, and the operator's reputation is the only guarantee you have. If it is neither, there is nothing to evaluate. The word is the problem, and the checks are the fix.