Project memory & context
Workspace folder
Each chat can be assigned a workspace folder (folder icon in the chat header). The agent sees this folder's structure and works only with its files.
Conversation and context compression
- Regular chats — history is compressed automatically as the model context limit approaches (toggle in the chat context panel); manually — via the command palette (
Ctrl+K). - Autonomous runs — once the working context exceeds ~96K tokens, older messages are folded into a summary automatically ("Agent context auto-compression" toggle in settings): the agent keeps the task context while token spend no longer grows with every step.
Compressed originals are not deleted: the summary card in the chat lets you browse the archive and restore messages.
File mentions (@)
Type @ to pick a specific file from the workspace. This is more reliable than hoping the agent finds it: the file is passed to the model explicitly.
What the agent remembers between sessions
- Settings, profiles and templates — locally and in the cloud (via your license).
- Chat contents — locally (and in the export backup).
- AI memory — the agent keeps facts across chats: general ones via
save_memory, workspace-specific facts viasave_memory_project(visible only in chats of that folder). Manage records on the "Memory" tab in settings; plan limits — Free 100 / Lite 150 / PRO 250 records; prompt-injection protection is built in.