Skip to content
Documentation ▾

AI

Connect any model

AtomCut is not tied to one AI company. Connect the models you already pay for — or a free one on your own computer — and switch between them in the middle of a chat. Media jobs (images, video, voice, sound) route to whichever provider you pick for each.

Add a connection

Agents ▸ Connect a model (or ⋯ ▸ Connections). Pick a provider, paste its key, press Test. A green check means a real request made the round trip — key, endpoint, model and browser access all at once. Providers that list their models refresh the list live, so new models appear without an AtomCut update.

Chat models

ProviderKey fromGood for
Anthropic (Claude)console.anthropic.comThe strongest agent for long, careful, multi-step edits. Opus 5.5 is the default; Sonnet 5 for speed.
OpenAIplatform.openai.comGPT-6 — plus images (GPT Image), video (Sora), voices and Whisper transcripts on the same key.
Google Geminiaistudio.google.comGemini — plus Nano Banana images, Imagen, Veo video and Gemini voices on the same key.
OpenRouteropenrouter.ai/keysHundreds of models behind one key.
xAI · DeepSeek · Mistral · Groq · Togethertheir consolesGrok, DeepSeek, Mistral, fast open models (Groq also transcribes).
Ollama · LM StudiononeFree and private — models on your own computer. For Ollama allow the site: OLLAMA_ORIGINS=https://app.atomcut.net ollama serve
Custom endpoint—Anything that speaks the OpenAI API: vLLM, LiteLLM, Azure, a company gateway.

A Claude or ChatGPT subscription is not an API key — API usage is billed by the provider separately. Smaller local models are weaker at long, tool-heavy tasks; a text-only model is told when it cannot see a render instead of being sent one.

Media providers

ProviderWhat it adds
falEvery media job on one key: Nano Banana Pro, FLUX.2, Veo 3.1, Kling 3, Sora 2, background removal for images and video, Topaz upscaling, ElevenLabs voices, sound effects and music.
ElevenLabsEleven v3 voices, Scribe transcripts, sound effects, music, voice clean-up and lyric timing (forced alignment).
OpenAI · GoogleImages, video and voices on your chat key (above).

A media-only connection never becomes the chat model — pair it with a chat connection.

Route each job

Each connection's page has Also use it for: switch on the jobs it should do — images, video, voiceover, transcripts and captions, sound effects, music, background removal, voice clean-up, upscaling, lyric timing. Pick a model per job, or leave Best fit for each job: it chooses the model that takes what the job hands it — an edit model to change a picture, an image-to-video model to animate a still.

Where your keys live

On this device, in the browser's storage, sent only to the provider they belong to. AtomCut's servers never see them.

Next: AI actions & Generate.