Back to Index
Commands
Slash commands · messaging + TUI
/model <name>
Switch active model. Use provider:model for cross-provider (e.g. /model deepseek:deepseek-v4-flash).
/new
Start a fresh session. Safest escape hatch when model is dead/404ing.
/reasoning <mode>
Set reasoning visibility. hide | show | none | low | medium | high. DeepSeek: none/low/med/high only.
/think
Toggle thinking mode on/off.
/voice
Start voice mode (CLI/TUI only — not messaging).
/stop
Stop the agent mid-response. Cancels all running background subagents.
/memory-status
Show memory gradient report — layers, entries, budget.
/rollback
Roll back to the previous turn.
/plan <prompt>
Create a structured implementation plan before executing.
/skill <name>
Load a specific skill into context.
/compact
Compress current session context to free token budget.
/help
Show available commands and usage.
Important Notes
• Slash commands are processed locally by the gateway before any API call — they work even when the active provider is dead/404ing.
• /new is the safest escape hatch on messaging platforms — wipes the dead session, starts fresh.
• /reasoning hide is INVALID for DeepSeek — use 'none' instead. 'hide' defaults to medium.
• /model with bare name resolves against current thread's provider. Use provider:model for cross-provider.
Commands · SourceMEM.com · 12 slash commands