Working with chats

A chat is the workbench: the place where your saved knowledge gets used. Everything below happens on your machine — the thread is a local file you own, every turn saved as it happens.

Attach vs load — the two gears

There are two ways to connect knowledge to a chat, and the difference matters. Attach a topic or KB and it joins the chat's retrieval scope: local vector search pulls in just the passages relevant to each question. Cheap, fast, and it scales to big libraries. Load a topic and its full text enters the context window: the model sees the whole document at once — right for close reading, rewriting, or when every clause matters.

Rule of thumb: attach for "answer from this when relevant", load for "work on this whole thing with me".

Comparing sources

Because a chat can hold several topics at once, it's where comparison happens: load one vendor's brief, attach the other's KB, and ask how their security postures differ. The answer draws on both, and cites each side.

Citations and verification

Answers grounded in your knowledge carry citations — click one to open the topic it came from. When you need more than a cited answer, build a verification report in the thread: sources scored against your own rubric, evidence attached.

Editing the thread

Messages aren't frozen. Edit any message, delete the ones that don't belong, and the thread stays clean. You can also put the AI to work on the thread itself: extract the useful parts of a long conversation, reformat them, or distil scattered turns into something finished — a bio, a summary, a brief.

Turning a chat into a topic

A good conversation is knowledge. Convert a chat into a topic and the thread joins your KB as source material — searchable, attachable to future chats, citable in future answers. This quarter's working session becomes next quarter's reference.

Export and import

Chats are portable files. Export a thread to back it up or hand it to a teammate; import one to pick up where someone else left off. No platform lock, no copy trapped in a cloud account.

Under the hood: markdown

Every message is stored as markdown inside the chat's JSON — the same markup language models natively read and write. That's why threads render cleanly, why exports are legible in any editor, and why chat content reuses so well as topic material.

Sessions

Each chat runs in its own session space — kick off a long research run in one thread and keep working in another; nothing waits. See the multi-session page for how concurrency works.