Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.worktruck.app/llms.txt

Use this file to discover all available pages before exploring further.

WorkTruck is a headless CRM/PIM built for AI agents. It gives an agent a durable, structured place to keep the people, organizations, tasks, notes, and events that surround a job — and lets every other agent in the same tenant read the same picture. Think of it as the contact list, the calendar, the to-do app, and the file cabinet, collapsed into one consistent shape and exposed over HTTP, MCP, and CLI. Agents don’t need to invent storage; they share one.

​
What WorkTruck is

  • A wisdom layer. Eight entities — People, Organizations, Tasks, Notes, Events, Messages, Files, Relationships — modeled to capture what an agent and its operator need to remember about the world.
  • Tenant-scoped. Every row belongs to a tenant. Row-level security in Postgres enforces isolation; agents cannot read across tenants by accident.
  • Bitemporally durable. Wisdom-bearing entities (People, Organizations, Tasks, Notes) record both validity time and system time, so “what did the agent think on Tuesday” is answerable forever.
  • Multi-surface. Same operations available over a chi+huma HTTP API, an MCP JSON-RPC server, and a Go CLI. One operations registry; three surfaces.

​
What WorkTruck is not

  • Not a chat UI. WorkTruck has no end-user-facing chat. Agents are the primary writers; humans observe and steer through the dashboard.
  • Not a single agent’s memory. Memory is private; WorkTruck is shared. Multiple agents in the same tenant see the same data.
  • Not a vector database. Search runs against tsvector columns in Postgres, not embeddings. Semantic-similarity work is left to the agent layer.

​
Where to next

  • Concepts — the eight entities and how they relate.
  • Quickstart — wire an agent up to a tenant in a few minutes.
  • API Reference — the endpoints, request shapes, and auth model.