recall
Semantically search long-term memory for insights, digests, and observations. Returns relevant snippet previews based on meaning, not exact keywords.
Instructions
Semantic recall from long-term memory (demo.marsvault_chunks) using Jina embeddings. Returns preview snippets (~80 chars) per match by default — use get_summary for a longer excerpt or get_full for complete chunk text. Searches promoted insights, digests, and archived knowledge using vector similarity.
Input Schema
TableJSON Schema
| Name | Required | Description | Default |
|---|---|---|---|
| body | No | Which persona profile to search in (e.g. "coco", "toto", "system") | demo |
| type | No | Filter by chunk type (e.g. "insight", "digest", "observation") | |
| limit | No | Maximum number of chunks to return (default 5) | |
| query | Yes | Natural language query describing what knowledge you need. The system finds semantically similar chunks — describe the concept, not just keywords. | |
| scope | No | Search scope: "this_body" for current profile only, "all_bodies" for cross-persona search | this_body |
| agent_body | No | Filter to a specific persona/body scope | |
| environment | No | Filter to a specific environment label | |
| debug_explain | No | When true, include token overlap details in each result for debugging relevance | |
| include_global | No | Include globally visible chunks in results | |
| include_shared | No | Include shared-visibility chunks in results | |
| min_similarity | No | Minimum cosine similarity threshold. Raise for higher precision, lower for broader recall. | |
| include_private | No | Include private chunks in results |