memory_ingest
Convert ephemeral insights into permanent long-term memory; automatically chunks text, generates embeddings, and stores for semantic recall.
Instructions
Promote content into permanent long-term memory (demo.marsvault_chunks). Automatically chunks the input text, generates vector embeddings, and stores each segment for semantic recall. Use this to preserve important insights, decisions, patterns, or knowledge that should survive beyond the current session. This is the primary path from ephemeral to permanent memory.
Input Schema
| Name | Required | Description | Default |
|---|---|---|---|
| content | Yes | The insight content to promote to long-term memory. Be specific and self-contained — future recall depends on the quality of this text. | |
| source_file | No | Logical file path or identifier for the source (e.g. "sessions/2026-06-10-session-notes") | |
| section | No | Optional section label prefix to organize chunks within the source | |
| tags | No | Categorization tags (e.g. ["decision", "architecture"]) | |
| type | No | Content type label (e.g. "insight", "observation", "decision") | insight |
| date | No | Date for this content in YYYY-MM-DD format. Defaults to today if omitted. | |
| visibility | No | Access level: "private" = this profile only, "shared" = cross-profile readable, "global" = system-wide | private |
| origin | No | Origin marker | warp-demo |
| source_memory_id | No | Link this promotion to a specific short-term memory ID (for traceability) | |
| source_session_id | No | Session ID where this insight originated (for provenance tracking) | |
| source_tool | No | The tool/platform where this insight was originally captured | |
| source_user_note | No | Brief note explaining why this memory was selected for promotion | |
| agent_body | No | The persona/body this memory belongs to (e.g. "coco", "toto") | |
| environment | No | Environment label (e.g. "production", "staging") | |
| max_chunk_chars | No | Maximum characters per chunk (default 1200). Adjust for finer or coarser granularity. |