DeepSeek Models In One Private Workspace
Multi-model AI means one interface in front of several models. ChatOTL is powered by DeepSeek and the model is chosen per message, so a single conversation can move between deepseek-v4-flash, deepseek-chat, deepseek-reasoner and your own compatible endpoint while keeping the same history.
Models are not interchangeable. A fast model is cheap enough to run twenty times; a reasoning model works step by step through a hard problem. Choosing per task beats being locked to one default.
How switching works
The model selector applies to the next message you send. Prior turns stay in the thread and are replayed as context to the new model, so you keep the conversation instead of restarting it.
When switching is worth it
- Draft with one model, then ask a stronger reasoning model to critique the draft.
- Move to a larger context window when you attach a long document.
- Attach a screenshot or scanned page: ChatOTL transcribes it with a dedicated vision pipeline and passes that text to DeepSeek.
- Fall back to a cheaper model for routine reformatting.
Branching plus switching
Branching forks a reply into its own thread. Combined with model switching, that gives you two answers to the same prompt from two models, side by side, without destroying either.
Bring your own endpoint
DeepSeek is the default. Any endpoint that speaks the same OpenAI-compatible protocol can also be registered, which covers self-hosted gateways — protocol compatibility only, ChatOTL does not use OpenAI. The available model list is driven by the server-side registry, so adding a model does not require a frontend change.
Limitations
- Capabilities differ per model: attaching an image only works on vision-capable models, and a text-only model receives a description instead.
- Long histories cost more tokens on every switch, because context is replayed.
- No live web access on any model in the workspace.
Frequently asked questions
Does the conversation reset when I change models?
No. The existing turns are sent to the new model as context.
Can I compare two models on the same prompt?
Yes — branch the reply, then send the same prompt on a different model in the new branch.
Which models are supported?
The DeepSeek models, plus any compatible endpoint you configure.
Switch models mid-conversation
Open a thread, ask a question, then answer it again with a different model and compare.
Start chatting