kaeru memory
Server Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| KAERU_VAULT_PATH | No | Override the vault path. Default is platform-specific: Linux $XDG_DATA_HOME/kaeru, macOS ~/Library/Application Support/ai.lamantin.kaeru, Windows %LOCALAPPDATA%\ai.lamantin.kaeru. | |
| KAERU_MCP_UPDATE_CHECK | No | Set to 0 to turn off the daily GitHub update check entirely. | |
| KAERU_MCP_HYGIENE_ENABLE | No | Set to 0 to turn off automatic background hygiene. On by default since 0.7.3. | 1 |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {} |
| resources | {} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| contextC | Restore project rules, current records, open work and saved reasoning. Records reminder delivery metadata. Use view=graph for compact authorized structure across selected or all accessible initiatives; bodies remain in read. |
| searchB | Find scoped memory using literal text or explicit filters; report currentness and matched scopes. |
| readB | Read a record with currentness, material doubts and a saved reasoning summary by default; expand named graph views when needed. |
| noteC | Save an observation, decision, standing constraint, plan or outcome. Optional slot names the current record for a project purpose. |
| referenceC | Preserve supplied source text or a verified artifact; a URL alone records a citation without fetching it. |
| taskC | Create an open task; optional deadline and relation are captured with it. |
| task_updateB | Update task status or deadline. Completion preserves the deadline and stops active debt; reopening uses the retained date. |
| claimC | Create an open hypothesis, or save an already reached verdict together with its evidence in one call. |
| verdictA | Record supported, refuted or inconclusive plus optional evidence in one call. The rare record_check action attaches evidence without changing status. |
| reviseB | Patch the same record, preserving identity and tier. Omitted fields remain unchanged; immutable source bytes are replaced through replace. |
| replaceA | Create an explicit successor, supersede the old record atomically and report dependents needing review. |
| relateB | Connect, disconnect or reweight a typed directed edge. A supersedes edge marks succession but does not itself retract a node. |
| reviewA | Open or close review flags, or link an answer to a question. Closing flags does not delete substantive counterevidence. |
| trailB | Save a reasoning summary and ordered route; explicitly refresh or extend an existing trail and expose summary staleness. |
| pinB | Keep a record in view or release the pin; does not assign a role or supersede other knowledge. |
| reflectB | Read unresolved work, stale summaries, scope fragmentation and maintenance candidates; no automatic semantic edits. |
| lifecycleC | Move a record to operational or archival tier with full content and preserved graph identity lineage. |
| synthesiseB | Create a supplied synthesis with links to its sources; does not silently close or retire the sources. |
| forgetB | Retract a record and its live incident edges while preserving inspectable history. |
| historyC | Read recorded versions and audit events, labeling fields whose prior values are not stored. |
| retentionC | Set a recall layer or inspect/run metadata hygiene; tier and semantic closure remain explicit. |
| initiativesA | List accessible initiative names; filtering does not create or merge aliases. |
| initiative_manageC | Rename, merge, delete or attach membership; occasionally set/clear a current-record slot. Every action has explicit scope and effects. |
| board_configA | Manage board columns and explicit terminal flags; reordering columns does not change completion semantics. |
| policyB | Read or set sharing policy and allowed clouds independently. Empty clouds explicitly removes the restriction. |
| exportA | Export current knowledge as a downloadable Markdown ZIP (format markdown) or portable AMG file (format amg). Select exactly one initiative, several initiatives or all. No import; no history backup. Destination stays inside the host artifact root. |
| cloudsB | List configured clouds. |
| shareC | Share a node using existing sharing gates. |
| unshareC | Withdraw a node from the named cloud. |
| cloud_initiativesB | List initiatives in one configured cloud. |
| cloud_recallB | Read a full shared cloud record without importing it by passing id, or list/search using the existing paging contract. |
| pullC | Copy one cloud node into the local initiative. |
| link_cloudC | Create a soft reference to a cloud record. |
| cloud_linksC | Resolve existing cloud references of a local node. |
| sync_reviewC | List local sharing candidates under existing policy. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 35 tools
Many tools target related memory operations, and several boundaries are fuzzy: claim vs verdict both can save verdict evidence, note/reference/synthesise all create records, and context/read/search overlap in retrieval. Detailed descriptions mitigate much of the confusion, but an agent still has to reason carefully to choose the right tool.
All names are snake_case, but conventions are mixed: some are single verbs or nouns (task, claim, note, read), some are verb_noun (task_update, link_cloud), and others are noun_verb or noun_noun (initiative_manage, cloud_recall, board_config). The set is readable but does not follow a predictable pattern.
With 35 tools, the surface is heavy for a memory/knowledge server and includes many niche maintenance, sharing, and cloud-sync operations. While the domain is broad, the count exceeds a well-scoped set and likely burdens tool selection.
The surface covers an extensive memory lifecycle: creation, revision, replacement, relations, review, trails, lifecycle tiers, retention, audit history, search/read/context, initiatives, export, and cloud sharing. Minor gaps remain, such as no import counterpart to export and no obvious initiative creation action beyond management.