Core concepts

Core memory

Core memory is the stable project briefing. It should contain the facts an AI assistant should know every time it works in the project.

Notes memory

Notes are durable records such as decisions, constraints, preferences, references, summaries, and general notes.

Events

Events record changes over time. They make sync, audit, and agent workflows easier to inspect.

Recall

Recall turns memory into searchable context. Local recall can be keyword-based. Cloud or provider-backed recall can use vector search and reranking.

Graph memory

Graph memory stores entities and relationships. It helps answer questions like “what depends on this?” or “how are these two decisions connected?”

Context package

A context package is the structured payload sent to an AI model or tool. It can combine core memory, notes, recall results, graph context, and source metadata.

Package architecture

TekMemo is published as a family of packages:

Package	Purpose
@tekbreed/tekmemo	Core runtime — filesystem store, graph memory, recall, and cloud client
@tekbreed/tekmemo-cli	CLI for local, cloud, and hybrid memory operations
@tekbreed/tekmemo-mcp-server	Self-hosted MCP (stdio) server for IDE/agent integration
@tekbreed/tekmemo-connectors	Connector framework — ingest external sources (GitHub, Notion, …) into .tekmemo/
@tekbreed/tekmemo-adapter-ai-sdk	Vercel AI SDK adapter — memory tool, runtime bridge, and context builders
@tekbreed/tekmemo-adapter-openai	OpenAI embedding provider adapter
@tekbreed/tekmemo-adapter-voyage	Voyage AI embedding and reranking adapter
@tekbreed/tekmemo-adapter-transformers	Local, zero-API-key embedding adapter
@tekbreed/tekmemo-benchmark-kit	Benchmark kit, workloads, and runners
@tekbreed/tekmemo-testing	Shared contract tests, fixtures, and fakes (internal/contributor use)

For application code, import the Tekmemo class from @tekbreed/tekmemo. See the API Reference page for the full module listing.

Glossary

Term	Meaning
Agent session	A temporary coding workspace managed by AgentFS, tracking the lifecycle of an agent's task.
Cloud client	The cloud client module of @tekbreed/tekmemo, handling API transport and auth.
Context package	A structured payload combining core memory, notes, recall results, and graph data to be sent to an AI model.
Core memory	The stable project briefing (core.md). Contains facts the agent must know every time it starts working.
Graph memory	Nodes (entities, concepts, files) and edges (relationships) that help tools answer architectural questions.
Hybrid runtime	A runtime mode that combines local and cloud, routing every read and write through read/write policies (local-first, cloud-first, local-only).
MCP	Model Context Protocol. An open standard used to expose TekMemo memory to IDEs and agents. Available as a hosted cloud-only endpoint (https://mcp.memo.tekbreed.com/) or a self-hosted stdio server (@tekbreed/tekmemo-mcp-server) for local/cloud/hybrid memory.
Note	A durable memory record (e.g., a decision, constraint, or summary) saved to notes.md.
Recall	The process of retrieving relevant memory for a query. Can use keywords (local) or embeddings (cloud/provider).
Snapshot	A point-in-time backup bundle of project memory, used for rollbacks or migrations.
Sync	The mechanism of resolving memory state between the local .tekmemo/ filesystem and the TekMemo Cloud database.