Developer Overview

Docket is built around adapter contracts. You can extend it by implementing these contracts — no need to fork or modify core code.

What you can build

• LLM adapters — Add support for new providers (Groq, Together, local models)
• Store adapters — Add new databases (MySQL, MongoDB, Qdrant)
• Blob adapters — Add new storage backends (Azure Blob, GCS)
• Queue adapters — Add new job queues (BullMQ, SQS, RabbitMQ)
• Connectors — Ingest from external sources (GitHub, Notion, email)

The contract model

Core knows nothing about your adapter. It only knows the interface.

Your Adapter  ──implements──▶  Interface Contract  ◀──called by──  Docket Core

This means:

• Your adapter lives in its own package or directory
• Core never imports from your adapter
• You can swap adapters at runtime via config

Two types of extensions

1. Adapters (infrastructure)

Replace how Docket talks to LLMs, databases, blob stores, or queues.

Example: adding a new vector database.

2. Connectors (data sources)

Ingest data from external systems into Docket.

Example: a GitHub connector that ingests issues and PRs.

Connectors are not adapter contracts. They are standalone packages that use Docket's API (or direct adapter injection) to push data in.

Getting started

1. Read the Adapter Contracts page
2. Pick the contract you want to implement
3. Use the implementation template from AGENTS.md
4. Write contract tests
5. Submit a PR or publish as a separate package

Plugin packages

Plugins can be published as npm packages with the naming convention:

@docket/{category}-{provider}

Examples:

• @docket/llm-groq
• @docket/store-qdrant
• @docket/connector-github

Core dynamically loads adapters by package name from config.