A Space is Finch's way of grouping a category of work together. By default, all conversations are mixed together and sorted by time; a Space lets you carve out a project or an ongoing area of work so Finch only remembers the rules and history relevant to it, kept separate from everything else.
Two kinds of Spaces
When creating a Space, you choose what it's anchored to:
| Not bound to a directory | Bound to a folder | |
|---|---|---|
| Best for | Ongoing work without specific files, e.g. "daily admin," "research on a long-running topic" | Projects with actual code/docs on disk |
| Files Finch can access | No fixed directory — you provide paths manually in conversation | The folder itself, readable and writable automatically |
| Where rules live | A note file Finch maintains internally | AGENTS.md inside the project folder, versioned with Git, shareable with the team |
A directory-bound Space is like giving a project its own dedicated Finch: open it, and Finch automatically knows which folder to work in and reads the rules file already written for that project. A Space without a directory is more like opening a dedicated channel for a category of work — good for things without specific files where you still want long-term context.
Rules: tell Finch how this Space should operate
Every Space can have its own set of rules that supplement or override Finch's general behavior — e.g. "commit messages in this project must be in Chinese," or "keep replies short in this Space, no emoji." Write the rule once, and every conversation in that Space follows it automatically — no need to repeat yourself.
For a directory-bound Space, the rules are the project's own AGENTS.md file, managed alongside the project and shared by the whole team. For a Space without a directory, rules are stored separately on Finch's side, visible only to you.
Memory: a Space remembers its own things
Beyond rules, a Space also builds up its own memory automatically — project details, agreements, and lessons learned discussed within it. Finch distills these in after each conversation, and they're automatically carried into future conversations in that Space, kept separate from other Spaces and regular conversations. This memory is generated automatically — you don't write it by hand, but you can always ask Finch to review, edit, or forget any part of it.
A Space can hold many conversations
Opening a Space doesn't mean a single conversation — you can start many independent conversations within the same Space. They share the Space's rules and memory, but their chat content doesn't interfere with each other. You can also "fork" a new conversation from any reply, exploring a different direction without affecting the original.
Spaces don't affect each other
Permission approvals are also tracked separately per Space: approving "always allow" for a command in one project Space doesn't automatically apply to another — unless the action itself is project-agnostic (e.g. certain general-purpose tools). This keeps authorization scoped correctly as you switch between projects.
Custom appearance
Each Space can have its own icon and color, so you can tell at a glance which Space you're in from the sidebar — especially useful when juggling multiple projects.
Relationship with automation tasks
When creating a scheduled task, you can specify that it runs "in a given Space" — every conversation the task triggers then carries that Space's rules, memory, and directory permissions. It's effectively "run this Space's work on a schedule," rather than starting fresh in a blank conversation.
When to create a Space
A project you're maintaining long-term, a recurring category of work, or anything where you'd rather not re-explain the background every time — all of these are worth a Space. For the occasional one-off question, a regular conversation is enough; you don't need a Space for everything.