Essentials

Chat

Jalaran Chat is a multi-session AI chat workspace that lets you run several AI conversations side by side, each preserving its full context.

Open Chat →Create free account

Who Chat is for

For anyone who works on more than one problem at a time. A single-thread chat forces you to choose between losing context and polluting it — you either start over or drag an unrelated debugging session into a writing task. Chat is for people who noticed that and got tired of it.

What Chat does

Chat keeps every conversation as an independent session with its own message history, so the model that has been helping you debug a worker queue does not suddenly have to reason about your cover letter. Sessions are created, renamed, listed and deleted from the sidebar, and each one streams its responses token by token over SSE rather than making you wait for a finished block of text. You can truncate a session back to an earlier message when a conversation goes somewhere unproductive, which is faster than starting again and keeps the good part.

Parallel conversations, each with independent context
Token-by-token streaming over SSE
Truncate a session back to any earlier message
Self-hosted Gemma — no third-party AI vendor in the path

How Chat works

Start a session
Each session is its own thread with its own history. Name it after the problem, not the day — you will come back to it.
Talk to the model
Responses stream as they are generated. Input is capped at 4,000 characters per message, validated on the server rather than only in the browser.
Switch without losing your place
Move to another session and back; the full context is still there. Nothing is summarised away behind your back.
Truncate when a thread goes wrong
Roll a session back to any earlier message instead of abandoning it, so a bad turn costs you one exchange rather than the whole conversation.

What Chat does not do

There is no model picker, and that is deliberate: Jalaran is locked to one self-hosted model so behaviour stays predictable and no request can be silently upgraded to a costlier or less private backend. Chat does not browse the web, does not execute code, and does not read your other modules — if you want an AI that acts across your workspace, that is Agent, which exists precisely so Chat can stay a conversation. Messages are capped at 4,000 characters each.

Common questions

Does my conversation get sent to OpenAI or Anthropic?

No. Inference runs against a self-hosted Gemma model reached through the Jalaran backend. There is no third-party AI vendor in the request path, and no user-facing model picker that could put one there.

Are my messages logged?

The backend logs metadata only — user id, session id, token count — and never message content. That is enforced in the chat router itself rather than left to logging configuration.