Chat and Sessions
Start a chat
Section titled “Start a chat”$ dreadnodedreadnode> /newStarted a new sessiondreadnode> Hello!dreadnode> /export ./kickoff-demo.mdExported conversation to ./kickoff-demo.mdSession commands
Section titled “Session commands”| Command | Arguments | Description |
|---|---|---|
/new | — | Start a fresh session |
/reset | — | Reset the current session state |
/sessions | — | List available sessions |
/use <id> | — | Switch to an existing session |
/delete <id> | — | Delete a session |
/rename | <name> | Rename the active session |
/export | [path] | Export the current conversation to a file |
/compact | [guidance] | Summarize older messages to reduce context size |
/agent | <name> | Change the primary agent for future turns |
/model | [provider/model] | Change the model for future turns |
Session runtime binding
Section titled “Session runtime binding”Each local chat session binds to exactly one server-side session runtime. The runtime stays stable while the TUI can change which primary agent and model it sends on future turns.
- Pick an agent before the first turn, or switch the primary agent later with
/agent <name>. - Pick a model with
/model <provider/model>and future turns in the current TUI process use it. /newcreates a fresh server session immediately./resetclears the current session state without changing the active session./api/chatruns against the current server session and can supply the agent and model for that turn.- Changing the primary agent does not create a new session; it changes which agent future turns use.
- Changing the model does not mutate stored server-side session metadata; it changes the model used on future turns.
@agent ...remains a one-off routed message and does not change the primary agent.- Per-session model restoration only lasts for the current TUI process.
Compaction
Section titled “Compaction”Long sessions accumulate context that can exceed model token limits. Compaction summarizes older messages while preserving recent history intact.
- Manual: Run
/compactto compact the active session on demand. Optionally provide focus guidance:/compact focus on the API design work. - Automatic: The runtime automatically compacts when token usage exceeds 75 % of the model’s context window. A flash notification appears when auto-compaction occurs.
- Compaction uses the same model as normal turns — it resolves from the agent definition first, then falls back to the session model.
- A compaction marker is persisted in the session history so restored sessions show that compaction occurred.
Tips for everyday use
Section titled “Tips for everyday use”Organize by project
Section titled “Organize by project”Prefix session names with a project slug (for example acme-kickoff) so /sessions stays tidy.
Switch between active sessions
Section titled “Switch between active sessions”Use /sessions to list what is available, then /use <id> to hop between threads without
starting over.