Skip to main content

Documentation

Developer documentation for Acolyte, a terminal-first AI coding agent. Reliable by default, observable, and open source.

Acolyte is built around explicit runtime concepts: sessions, tasks, lifecycle phases, and tool contracts. Each piece has a defined boundary and a typed interface. The documentation below explains how these concepts fit together and how the system behaves during execution.

Overview

  • Features — shipped, user-visible capabilities
  • Why Acolyte — observable execution and full developer control over AI coding

Architecture

  • Architecture — headless daemon with typed RPC connecting CLI, editors, and custom clients
  • Workspace — workspace root resolution, sandboxing, and profile behavior
  • Lifecycle — how each task flows through resolve, prepare, generate, and finalize
  • Observability — local task traces, structured events, and lifecycle summaries
  • Errors — error contracts and runtime classes

Runtime

  • TUI — React terminal UI with fuzzy search, autocomplete, model picker, and code editing
  • Tooling — layered tool execution with contracts and effects
  • Sessions — chat context, message history, and session storage backends
  • Tasks — task lifecycle, queue policy, and state transitions
  • Memory — structured facts persisted across session, project, and user tiers
  • Cloud — portable agent identity via cloud-hosted memory and sessions
  • Context budgeting — per-call prompt planning, bounded payloads, and usage breakdowns

Reference

  • CLI — commands for chat, run, server, memory, config, logs, and trace
  • Configuration — settings for models, providers, memory, and runtime behavior
  • Protocol — transport-facing contract between client and server
  • Localization — translatable copy separated from protocol contracts
  • Updates — versioning, auto-update, and breaking change policy
  • Glossary — core terminology for sessions, tasks, phases, and effects
  • Testing — test types, naming, and execution