Loading ToolBench
77 operations for vector search, TurboQuant vector compression (PolarQuant + QJL), NoSQL, dedicated PostgreSQL management, files, events, RLHF, and persistent memory for AI agents with enterprise security. All tools annotated with readOnly/destructive/idempotent hints per MCP spec 2025-03-26. Production-ready MCP integration for AI assistants.
Server provides 4 core embedding/memory tools with strong naming conventions (verb_noun pattern) and comprehensive schemas. Tool descriptions are clear and contextual, ranging from 168-270 characters, within the 10-1024 baseline and well above the 194-char median. All parameters include type definitions and descriptions. However, output schemas are not explicitly documented in the source code, and error handling lacks recovery guidance. The server claims '77 operations' in package.json but only 4 tools are visible in the provided source excerpt. Tool composition is good (each does one thing), but the incomplete source view limits full assessment.
Generate embeddings and store in ZeroDB in one step. Auto-detects dimension based on model and routes to correct storage. Use when you want to embed text and persist the vectors in a single call. Combines zerodb_generate_embeddings + zerodb_upsert_vector.
Generate embeddings using specified model (384/768/1024 dimensions). Supports BAAI/bge-small-en-v1.5 (384-dim, default), BAAI/bge-base-en-v1.5 (768-dim), BAAI/bge-large-en-v1.5 (1024-dim). Use when you need raw embedding vectors without storing them. Does not persist data — for embed + store in one step, use zerodb_embed_and_store instead.
Semantic search using text query (auto-embeds query text). Searches vectors in the same dimension as the model used. Use when searching vectors using a text query (auto-embeds the text). Unlike zerodb_search_vectors which requires a raw vector, this accepts plain text. Unlike zerodb_search_memory which searches agent conversation memory, this searches the general vector store.
Store agent memory in ZeroDB for persistent context. Use when the agent needs to persist conversation turns, decisions, or observations for later recall. Stores with role, session, and agent metadata for filtered retrieval.
Output schemas not documented. Tools define input schemas clearly but lack documented return value structures. LLMs need to know what fields to expect in responses to plan downstream calls.
Error handling lacks recovery guidance. No evidence of actionable error messages or suggestions for next steps when tools fail (e.g., 'Invalid model' errors do not guide LLM toward valid enum values).
Source code incomplete. Package.json claims '77 operations across 11 categories' but only 4 tools are visible in index.js excerpt. Cannot verify 73 additional tool definitions. Full server assessment requires complete source.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | B | 72 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 0 | - | v1 |
Pagination not visible. zerodb_semantic_search accepts 'limit' parameter but source does not show handling of large result sets, cursor management, or total_count return. No evidence of pagination pattern implementation.
Batch operations not offered. zerodb_generate_embeddings processes arrays but tools like zerodb_store_memory do not expose batch variants. N sequential calls waste tokens and latency.