Poe

What Poe is

Poe's creator documentation describes the product as a platform for creating and sharing AI-powered applications, and says it reaches millions of users.^[1] The underlying idea is that an AI application does not have to be a standalone website: a creator can assemble one on Poe, publish it there, and let the platform's users find and use it.

What distinguishes Poe's approach is that it offers several routes to the same goal rather than a single builder. They range from a few lines of written instructions to a program running on the creator's own server, and each route trades convenience for control in a different way.^[1]

The lightest option: prompt bots

A prompt bot is the simplest kind of application on the platform. It pairs custom instructions with a single base model: the creator writes the behaviour they want, chooses the model underneath, and the result is a bot that answers within those instructions.^[2] Poe pre-fills the bot's name with a randomly generated string of 4 to 20 characters, which the creator can keep or change.^[2] No code, no hosting, no API keys — just text and a model choice.

Canvas apps

Canvas apps are interactive web applications hosted on Poe with custom visual interfaces.^[5] Where a chat bot answers inside a conversation, a canvas app can present its own layout and controls. Poe's App Creator bot can build one without coding, so the interface is assembled by describing what is wanted rather than writing front-end code.^[5]

Script bots: orchestrating several models

A script bot sits between the no-code and the fully coded end. It orchestrates several models with custom Python logic, so a single bot can route different parts of a task to different models.^[4] Poe's Script Bot Creator writes that code from natural conversation, which lowers the barrier for creators who have an idea for a multi-model workflow but not the Python to implement it.^[4]

Server bots

Server bots run on the creator's own server and follow the Poe Protocol specification, which defines how Poe and the bot talk to each other.^[3] Because the logic lives outside Poe, the creator keeps full control over what the bot does and what it can reach. To make the protocol easy to adopt, Poe publishes a Python quick-start repository that a creator can use as a starting point.^[3]

API bots

An API bot connects an existing Chat Completions or Responses API endpoint to Poe.^[6] In this setup the creator does not implement the Poe Protocol; they point Poe at an endpoint they already run, and Poe charges the user based on the token usage that endpoint reports.^[6] The creator's own infrastructure stays in the loop, and the cost of a conversation is settled according to what the endpoint says it consumed.

External applications

Not every application has to live inside the Poe client. External applications run outside it, connect with a user's Poe API key, and spend that user's points.^[7] Poe also supports Sign in with Poe via OAuth, so an outside product can let people identify themselves with their Poe account.^[7]

Paying creators

Poe's creator monetization program pays creators per call to their bot.^[8] Payouts can be set either in code or as flat per-message rates on the site, so a creator can price a bot uniformly or vary the price across different kinds of calls.^[8]

On the cost side, Poe says it intends to cover all model inference costs and other significant per-message costs for bots running on the platform.^[9] Together these two arrangements address the usual obstacles to publishing an AI application: the creator does not have to fund the model calls up front, and there is a route to earning from the ones that get used.

Data handling

Poe states that it will not use a bot's responses to train LLMs, nor share them with a third party without an agreement preventing such use.^[10] For creators whose bots handle material they would rather not see used as training data, that statement is part of what makes the platform usable.

Where it sits

Poe works less like a single product and more like a distribution layer for AI applications. The platform supplies the audience and the account; the creator supplies the behaviour. That split is visible in the six creation routes: a prompt bot hands almost everything to the platform, a server bot hands almost nothing, and script bots, canvas apps and API bots sit in between.^{[1][3][4][5][6]}

The ladder from prompt bot to server bot maps onto a familiar trade-off. The more of the behaviour you hand to the platform, the less you have to build and run yourself; the more you keep on your own server, the more control you retain over logic and access. For users, the value is that many different bots and apps sit behind one account, one points balance and one sign-in.^[7] For creators, the value is reach plus a payment mechanism plus coverage of the model costs that would otherwise fall on them.^[1][8][9] Whether that combination holds up over time depends on how many of those millions of users actually call a given bot, which is exactly what the per-call payout is designed to reward.^[1][8]