← All features

Spaces

A Space is a long-lived workspace built around a project: a codebase, a research question, an ongoing piece of writing. It bundles instructions, reference files, memory and pinned model defaults. Every chat created inside a Space inherits that context automatically.

What a Space holds

• Instructions — a system prompt that overrides the global custom instructions when set. The Space is the user explicitly opting in to space-level guidance.
• Reference files — text files and PDFs are inlined into the system prompt of every chat in the Space; images ride along with the user turn so vision-capable models can see them.
• Memory — optional auto-extracted one-line facts about the user (see below).
• Default provider and model — pinned per-Space, so a “code review” Space can always start in a different model than a “writing” Space.
• Default generation parameters — temperature, top-p and the rest, layered between model defaults and per-chat overrides.

Creating and editing

Create a Space from the Spaces sidebar tab or from the in-app library tile. Each Space has a detail view with its own tabs for chats, instructions, files, memory and model defaults. Editing any field updates the context for every chat inside the Space the next time they are sent.

Chat rows in the Space’s Chats tab expose the same per-row … menu as the main sidebar — Pin / Unpin, Rename, Move to archive, and Delete — so you can manage a Space’s chats without leaving the Space view.

Deleting a Space cascades: the chats inside it, the reference files and the auto-saved memories all go with it. Move any chat you want to keep out of the Space first.

Space Memory

Optional per-Space auto-memory. After every assistant reply in a Space, Loach fires a one-shot call against the same provider and model the user is chatting with and asks it to extract durable, single-sentence facts about the user from the new turn. The surviving facts are deduplicated and persisted as memories.

You will see a small Saved to memory pill briefly appear after every save, showing the new fact — so you always know what landed in long-term context.

• Memory tab — review, edit or delete any auto-saved row. You can also add facts manually.
• Per-Space toggle — turn extraction off without wiping the existing rows. Stored memories keep riding along in every chat; only new writes stop.
• Cancelled or errored turns never trigger extraction, because the assistant’s text is incomplete.

Memories are silently injected into the system prompt of every chat inside the Space as a short --- Space memory --- bulleted list.