Developer docs
Product concepts and API reference for Copepod.
This docs site is organized around the actual platform surface: what the product does, the concepts behind it, and the REST endpoints exposed by the services in this repository.
Overview
What Copepod is built to do
Live context sync
Queue-Consolidate-Resolve keeps concurrent agent writes consistent and tenant-isolated.
Progressive enrichment
Knowledge graph, LLM, and web-grounded enrichment are enabled by plan and policy.
Security by default
Tenant scoping, audit logs, encryption, rate limiting, and step-up controls are built in.
Knowledge structures
Schema-driven knowledge objects let you store partial data and enrich it over time.
Developer access
API keys, auth flows, and role management support real integrations and team workflows.
Libraries & SDKs
Call Copepod from curl, Python (cme-sdk, private beta), and — soon — Node and Go.
MCP Server
Claude Desktop and MCP clients can search and store memories directly from conversations.
LOCOMO Benchmark
Measure retrieval quality with Recall@10, MRR, and NDCG scores across search modes.
Concepts
Terms used throughout the product
Tenant
Every request is scoped to a tenant. The backend uses tenant IDs, row-level isolation, and request middleware to keep data separated.
Memory
A memory is the primary tenant-owned record. It includes content, tags, sensitivity, versioning, contradiction state, and enrichment metadata.
Knowledge structure
A schema that defines the fields for a knowledge object and the enrichment rules for each field.
API key
API keys are scoped by permissions and can be rotated or revoked. Sensitive operations use step-up authentication.
MCP Server
A Model Context Protocol server that provides CME tools to Claude Desktop, OpenClaw, Cursor, and other MCP clients. Tools include cme_status, cme_create, cme_get, cme_update, cme_delete, cme_list, and cme_search with graph and explain support.
LOCOMO Benchmark
A quality benchmark runner that measures CME retrieval performance using Recall@10, MRR, and NDCG@10 metrics. Supports ablation studies across vector-only, hybrid, and hybrid+graph search modes.
API reference structure
The API reference is split into service-specific pages so you can focus on the surface area that matters to your integration.
Quick start
curl -H "X-Tenant-ID: <tenant-id>" \ -H "Authorization: Bearer <token>" \ "http://localhost:8082/memories/search?q=context&k=10"
Use the Memory API for application integrations and the Portal facade when you are extending the browser UI itself.