Modes
The Agentic Memory OpenClaw plugin supports two runtime modes. You can switch between them anytime without reinstalling the plugin.
capture_only
In capture_only mode, the plugin captures conversation turns and makes memory search available, but it does not inject custom context blocks into LLM prompts.
- Turn capture is active via the context-engine slot.
- Memory search works inside the OpenClaw memory slot.
- No additional context is added to prompts.
- Safest default for operators who want memory without changing prompt behavior.
openclaw agentic-memory setup --mode capture_onlyaugment_context
In augment_context mode, the plugin does everything in capture_only plus it resolves and injects relevant Agentic Memory context into each turn.
- Turn capture is active.
- Memory search works inside the OpenClaw memory slot.
- The backend assembles context blocks for prompt augmentation.
- Requires the backend's context engine pipeline to be healthy.
openclaw agentic-memory setup --mode augment_contextComparison
Memory capture and search are active, but custom context blocks are not injected into prompts.
- Turns are captured and stored in the knowledge graph
- Memory search works inside the OpenClaw memory slot
- No additional context is injected into LLM prompts
Which should I choose?
Start with capture_only. Once you validate that turns are being captured and memory search works, switch to augment_context to get the full experience.
The doctor command will tell you if the backend is honestly ready for augment-context before you switch.