Docket Documentation

Docket is an open-source, self-hosted Second Brain as a Service. It is built for applications that need more than simple document search: a memory system that models knowledge after human cognition, reasons across time, and retrieves the right context when it matters.

Unlike flat RAG pipelines, Docket stores memories as a structured graph with sectors, validity over time, salience scores, and adaptive decay. You can ask what was true last month, surface the most relevant memory for a task, and build agents that remember context across long-lived conversations.

Choose your path

Whether you are running Docket for your own apps or extending it with custom adapters, the documentation is split into two tracks.

What makes Docket different?

• Cognitive Memory Model — Memories are classified into sectors (episodic, semantic, procedural, emotional, reflective) instead of being dumped into a single vector index.
• Temporal Knowledge Graph — Every memory has a lifetime, so you can ask time-bound questions like “what did we decide on March 1st?”
• Composite Retrieval — Combines vector similarity, graph traversal, salience, recency, and temporal filters into one ranked result set.
• Adaptive Decay — Per-sector forgetting curves surface what matters and fade what does not, mimicking real memory.
• RBAC Access Control — Resource-based policies protect every memory, making multi-tenant and multi-agent deployments safer.
• Pluggable by Design — Swap LLMs, embedders, vector stores, blob providers, and queues without rewriting your application.

Get started in minutes

New to Docket? Start with the quickstart to install the CLI, configure your first adapter set, and run your first query.

Community

Docket is open source and actively developed. If you run into issues, want to contribute adapters, or discuss architecture, open an issue or pull request on GitHub.