search
Search stored knowledge using hybrid keyword and semantic fusion, auto-strengthening memory recall with every access.
Instructions
Unified search tool. Uses hybrid search (keyword + semantic + convex combination fusion) internally. Auto-strengthens memories on access (Testing Effect).
Input Schema
TableJSON Schema
| Name | Required | Description | Default |
|---|---|---|---|
| limit | No | Maximum number of results (default: 10) | |
| query | Yes | Search query | |
| concrete | No | Force literal/concrete search. Skips semantic expansion, FSRS reweighting, spreading activation, and cognitive side effects. Auto-enabled for quoted strings, env vars, UUIDs, paths, and code identifiers. | |
| source_id | No | Investigation filter: a specific source record id (issue number / ticket id). Pair with source_system to disambiguate across systems. | |
| tag_prefix | No | Optional tag-prefix filter. When set, only results carrying at least one tag whose value starts with this prefix are returned (case-sensitive). Example: tag_prefix="meeting:" matches memories tagged 'meeting:standup', 'meeting:1-on-1', etc. Applied as a post-filter; combine with a larger 'limit' if you expect heavy thinning. | |
| source_type | No | Investigation filter: source record type, e.g. 'issue', 'comment'. | |
| detail_level | No | Level of detail in results. 'brief' = id/type/tags/score only (saves tokens). 'summary' = default 8-field response. 'full' = all fields including FSRS state and timestamps. | summary |
| token_budget | No | Max tokens for response. Server truncates content to fit budget. Use memory(action='get') for full content of specific IDs. With 1M context models, budgets up to 100K are practical. | |
| exclude_types | No | Node types to exclude from results (e.g., ['reflection']). Reflections are excluded by default to prevent polluting factual queries. | |
| include_types | No | If set, only return nodes of these types. Overrides exclude_types. | |
| min_retention | No | Minimum retention strength (0.0-1.0, default: 0.0) | |
| source_author | No | Investigation filter: the source author/reporter (not assignee). | |
| source_status | No | Investigation filter: 'any' (default), 'valid' (currently-valid records only), or 'tombstoned' (records no longer visible upstream, kept for audit). | any |
| source_system | No | Investigation filter (#57): only memories ingested from this external system, e.g. 'github' or 'redmine'. Post-filter — non-connector memories are excluded. Combine with a larger 'limit' if thinning is heavy. | |
| context_topics | No | Optional topics for context-dependent retrieval boosting | |
| min_similarity | No | Minimum similarity threshold (0.0-1.0, default: 0.5) | |
| retrieval_mode | No | precise: top results only (fast, token-efficient, skips activation/competition). balanced: full 7-stage cognitive pipeline (default). exhaustive: maximum recall with 5x overfetch, deep graph traversal, no competition suppression. | balanced |
| source_project | No | Investigation filter: only memories from this source project/repo, exact match (GitHub 'owner/repo', Redmine project id). | |
| source_updated_after | No | Investigation filter: only records whose source was updated at/after this RFC3339 timestamp (inclusive). | |
| source_updated_before | No | Investigation filter: only records whose source was updated at/before this RFC3339 timestamp (inclusive). |