Introduction

Pimsy is an agent platform. The language model is one component inside it — the reasoning engine — surrounded by a runtime that plans work, retrieves information, calls tools, operates software, writes and executes code, and verifies results before returning them.

This documentation describes the system as it is designed. Sections that are implemented, in preview, or still conceptual are labelled at the top of each page. Where an API surface has not been finalized, examples are marked conceptual rather than presented as a shipped interface.

The difference in one diagram#

Conventional LLM application
  prompt ──▶ reason ──▶ response

Pimsy agent runtime
  goal ──▶ understand ──▶ plan ──▶ research ──▶ select tools
                                                    │
                            ┌───────────────────────┘
                            ▼
                     execute ──▶ observe ──▶ verify
                            ▲                  │
                            └──── iterate ◀────┤
                                               ▼
                                           complete

A conventional deployment produces text. Pimsy produces an outcome: a tested repository, a research memo with citations, a filled-in form on a real website, a simulated and approved transaction, or a structured dataset — along with the trace of how it got there.

What you can build#

How the documentation is organized#

Start here#