Project-Scoped Chat
When you're chatting "inside" a project, several things change vs a regular conversation. Most are silent and helpful.
What changes when you open a project
| Outside a project | Inside a project | |
|---|---|---|
| Conversation list shown | All your conversations | Only this project's conversations |
| New conversations | Saved at the top level | Saved in this project |
| System prompt | None (model defaults) | Project's system prompt prepended to every message |
| Preferred model | Your global pick | Project's pinned model (if set) |
| Memory recall scope | All your conversations | Biased toward this project's conversations + global memories |
| Artifacts you save | Top-level artifact vault | Project's artifact subfolder |
How to tell which mode you're in
The chat header shows the active project's name + icon when you're inside one. Click the × next to the name to leave the project and return to top-level.
In some chassis (Ridgeline, Quant, Apsis) the chat surface tints slightly to indicate project mode. Others (Matrix, Paper, Garden) keep the visual identical and rely on the header text.
The system prompt in action
When you send a message inside a project, the request body looks like:
{
"project_id": "abc-123-...",
"model": "anthropic/claude-sonnet-5",
"messages": [
{ "role": "system", "content": "<the project system prompt>" },
{ "role": "user", "content": "<your message>" }
]
}The system message is prepended automatically. The model sees the system prompt before your first user message in every conversation in the project.
For multi-turn conversations within the project, the system prompt stays at index 0, then all prior turns, then the new user message. The model gets full conversation context plus the project's framing on every turn.
Memory recall in projects
The memory layer recalls relevant past chunks before each message. Inside a project:
- Memories from THIS project are weighted higher
- Memories from OTHER projects still appear if highly relevant
- Memories from the same time window are clustered
This means you can pick up where you left off — a research project with 50 prior conversations will surface relevant chunks naturally as you continue.
Switching projects mid-conversation
You can't. A conversation belongs to exactly one project (or none). To "move" a conversation to a different project, use drag-drop or right-click → Move to project (see Moving Conversations).
Pinned model behaviour
If a project has a preferred model set, every new conversation in that project uses it by default. You can still override per-conversation via the model picker — that override applies only to that conversation, not the project.
If the preferred model is unavailable (e.g. you set Anthropic but your BYOK key was removed), the request falls back to your global model with a small warning toast.
Sharing patterns
Currently: there's no way to share a project with another user. Each project is private to the account that created it. Multi-user / shared projects are on the roadmap but not shipped.
If you need to collaborate, the workaround is:
- Export a representative conversation as a JSON or Markdown
- Send it to your collaborator
- They paste it into a fresh project on their account
Where to next
- System Prompts — what to put in the prompt
- Moving Conversations — drag-drop and per-row
- Memory articles (Phase 4) — how recall works across projects