Loading ToolBench
A knowledge graph and memory management system with hybrid search (FTS5 + vector embeddings + graph traversal), code symbol indexing, and graph analytics for note-taking and knowledge base management
My-Memory demonstrates solid schema completeness and thoughtful parameter design with enums, ranges, and progressive context loading. Tool names follow verb_noun convention (memory_search, memory_write_note). However, descriptions are verbose (200-400+ chars, exceeding the 10-200 char sweet spot) and written in Portuguese, creating friction for English-speaking LLM agents. Several tools lack output schema documentation. Error handling and recovery guidance are absent from visible code. The server shows domain expertise (RRF, graph traversal, code symbol indexing) but needs tighter descriptions and explicit error patterns.
Retorna o sub-grafo (in/out edges) de um code_symbol via CTE recursivo até depth hops (CA-11, ADR-004 + ADR-047)
Busca estruturada em code_symbols (function/method/class/interface/struct/etc.) com boost RRF 2x por qualified_name match exato (CA-10, ADR-047)
Audita a integridade do grafo e tabelas de notas, identificando dead links, notas órfãs, self-loops e calculando o Health Score
Exporta um subgrafo centrado em node_id no formato aberto JSON Canvas 1.0 (.canvas) para visualização espacial no Obsidian
Retorna os nós com maior centralidade de conexões (God Nodes / Hubs de conhecimento) no grafo de notas
Retorna conexões latentes e surpreendentes (Surprising Connections) entre notas com alta similaridade sem links diretos no grafo
Descriptions are verbose (200-400+ chars) and in Portuguese. LLM agents expect concise English descriptions (50-200 chars) for efficient tool selection. Portuguese descriptions create friction for English-speaking LLM clients.
Output schemas are not documented in visible code. LLMs cannot plan downstream tool calls or extract required fields (e.g., what does memory_search return? field names, types, pagination info?). This violates the documented output schema requirement.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 58 | 2026-07-28+ | v2 |
Retorna nós e notas vizinhas no grafo relacional a partir de um node_id via CTE recursivo
Busca híbrida com Reciprocal Rank Fusion (RRF) combinando texto exato FTS5/tsvector, vetores k-NN e expansão de grafo
Grava ou substitui uma nota atômica em Markdown no vault com frontmatter estruturado e conexões tipadas, disparando indexação cirúrgica imediata no grafo
No error handling or recovery guidance visible. Tools lack descriptions of failure modes, retryability, or actionable error messages. E.g., memory_search should document: 'Returns empty array if no matches found. Try broader query or different mode (fts vs vector).'
memory_write_note has incomplete schema: 'path' and 'title' are documented, but the description does not specify: required frontmatter fields, markdown format constraints, or what 'indexação cirúrgica imediata' (immediate surgical indexing) means to the LLM. Vague descriptions invite misuse.
memory_export_canvas has optional 'output_path' parameter but no guidance on default behavior or file permissions. Does it write to disk? Return JSON? Overwrite existing files? Agents need explicit behavior documentation.