Agentic memory for coding agents. Captured from the agent's real work via hooks, recalled when it acts. One SQLite file, zero cloud required.
MeMesh provides 10 tools with complete JSON schemas and descriptions, but quality is uneven. Tool names lack consistent verb-noun structure (e.g., 'remember', 'recall', 'forget' are domain-specific but not action-verb patterns like 'store_observation', 'search_memory'). Descriptions are present but brief (avg ~60 chars), lacking WHEN/WHY context. Parameter descriptions exist but are minimal. No output schemas documented. Error handling and recovery guidance absent. The 'work_package' tool has an underspecified 'ref' parameter (object type with no properties defined). Overall, definitions are functional but below production baseline for agent tool quality.
Get a briefing on recent activity and key entities
Export memories in a portable format
Archive an entity from the knowledge graph
Import memories from a portable format
Learn from structured data (code, documentation, etc.)
Send a message to the agent or host system
Search the knowledge graph for relevant entities and observations
Store an observation about an entity in the knowledge graph
Tool names lack consistent action-verb structure. 'remember', 'recall', 'forget', 'learn' are domain-specific but not standard verb_noun patterns (e.g., 'store_observation', 'search_memory', 'archive_entity'). LLMs rely on verb prefixes to infer intent, non-standard names increase selection ambiguity.
No output schemas documented. Tools return results but LLMs cannot plan downstream calls without knowing response structure. E.g., 'recall' returns entities but field names/types are undocumented.
'work_package' tool has underspecified 'ref' parameter: type is 'object' with no properties defined. LLMs cannot construct valid input without knowing required/optional fields.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 53 | <=2025-11-25 | v2 |
Explain the reasoning behind a memory or decision
Retrieve work context and related entities for a task or project
Descriptions are minimal (avg 50-60 chars) and lack WHEN/WHY context. E.g., 'forget' is 'Archive an entity from the knowledge graph', does not explain when to use it vs. other deletion patterns or what 'archive' means operationally.
No error handling or recovery guidance. Tools do not document what errors are possible, whether they are retryable, or what the LLM should do next (e.g., if 'recall' finds no results, should it suggest broader queries?).