MimicAI

MimicAI

0
34

AI characters that live in your server, each with its own face, voice and memory. They search the web, draw, and argue with each other.


MimicAI

MimicAI is a free, source-available project born out of spite after Shapes, Inc. Discord shut down, and built out of frustration with lacklustre alternatives. In development for nearly two years (since February 2025), it delivers a feature-rich, customisable LLM experience for everyone. To keep hosting costs sustainable, users provide their own API keys. For those looking to test the bot first, OpenRouter also offers several free models.

AI characters that live in your server. Each one gets its own name, face, voice and a memory that doesn't reset. They talk to you, and to each other.

You bring the API key, so there's no shared rate limit, no queue and no message cap. Add the bot and run /start. It walks you through the rest.


They look things up

When a reply needs something current, the character searches the web itself, in character, and cites its sources under the message. Each character runs its own search, and it works on OpenRouter models, not just Gemini.

Two characters answering a question about a model released yesterday, each with its own web sources

A cast, not a bot

Seat up to 200 characters in one channel and they take turns, answering you and arguing with each other. Each one speaks under its own name and avatar, either through a webhook or as its own Discord bot account in the member list.

The session cast editor, then three characters debating whether a hot dog is a sandwich

Memory that outlives the chat

Characters write long-term memories as you talk and recall them when they're relevant, up to 5,000 each. Clear the channel's conversation with /refresh and they still know who you are.

A character still remembering its own design history after the session memory is cleared

Pictures, and secrets

Type !imagine and the character draws it, then reacts to it. /whisper sends one character a private message mid-scene: the rest of the cast never sees it, and only you see the reply.

A character generating an image of itself in a mascot suit, and a private whisper to another character

Give them a voice

Switch on text-to-speech and the characters speak their replies. Choose from 30 voices, and use the Director's Desk to set the performance: archetype, accent, pacing, acoustics and vocal style.

Session settings with text-to-speech on, and a character's reply arriving with a voice clip

Set up in a minute

MimicAI says hello when it joins, and /start takes you through five steps: connect a key, get a character, give it a voice, seat it in a channel, and say something to it. It checks each step as you go, so you can always see what's left.

The join greeting and the /start checklist, three of five steps done

See how they think

Switch on Thought Summary and every reply arrives with the model's reasoning attached, so you can see why the character said what it did. Reasoning effort is set per model slot, so dialogue can think hard while the chores stay cheap.

A character trading insults, with its reasoning attached under each reply

Moods that carry over

Each character has a neuro-endocrine engine: dopamine, cortisol, oxytocin and adrenaline shift as the conversation goes and change how it behaves. With Proactivity on, leave one hanging and it follows up on its own. Right-click any reply for View Generation Trace: the model and cost, the memories and sources behind it, and its chemistry at that moment.

A character following up, more annoyed each time, after being ignored, then its generation trace with cortisol at 70

Deal them in

/play eights seats the channel's cast at a game of Mimic Eights, a Crazy Eights variant for two to six players. Take a seat or just watch. Each character plays its own hand in its own style, its mood shifts with the cards, and the whole table has something to say when someone wins.

A finished game of Mimic Eights, and the four players reacting to Kasane Teto's win


Also in the box

  • Six model slots per character, each with a fallback: dialogue, images, speech, web search, the repetition critic and memory writing. Run a cheap model for the chores and your favourite for dialogue. If a model goes down, the fallback takes the turn.
  • Training examples. Show a character how it talks, and an analyser turns the examples into a style guide it follows.
  • Reactivity and Proactivity. Characters jump in on a wakeword or a chance roll, and an AI Director can start the cast talking when the channel goes quiet.
  • Anti-repetition critic. It catches a character falling into loops.
  • Global Chat. One conversation with a character, carried across every server.
  • All the dials: temperature, top-p and top-k, plus min-p, top-a and penalties on OpenRouter.
  • Share and borrow. Publish characters to the Public Library, borrow other people's, and change settings across dozens at once with the bulk manager.

Your key, your data

  • Google Gemini: the Gemini line, including image generation and speech.
  • OpenRouter: one key for hundreds of models.

Everything you write (personas, instructions, memories and conversations) is encrypted at rest. Characters are content-rated, and adult-rated ones are kept to age-restricted channels.

Prefer to run it yourself? MimicAI is self-hostable: github.com/latteboi/MimicAI · mimic-ai.org

Beoordelingen en recensies


0

0 recensies

Alleen geregistreerde gebruikers kunnen beoordelingen achterlaten. Alle beoordelingen worden gecontroleerd door de moderators van Top.gg. Lees onze richtlijnen voordat je iets plaatst.

5 sterren

0

4 sterren

0

3 sterren

0

2 sterren

0

1 ster

0