Skip to main content

Features

Acolyte combines a terminal-first client, headless daemon, lifecycle effects, persistent memory, and typed tools in one open-source coding agent.

CLI

  • interactive chat and one-shot run/skill commands
  • persistent daemon with automatic startup and lifecycle management
  • session resume by ID prefix with history
  • model picker that queries provider APIs for available models
  • fuzzy search and autocomplete for file paths, sessions, commands, and skills
  • file and directory attachments via @path
  • slash commands and skill invocation
  • engineering skills for structured workflows (plan, build, review)
  • an always-available skill roster the agent activates and deactivates on demand
  • project and user skills from .agents/skills
  • configurable locale
  • multi-line input
  • custom terminal renderer with React reconciler and structured output
  • syntax highlighting for fenced code and edit diffs
  • live status line with location, model, token, skill, and PR segments
  • auto-update on startup with progress UI
  • update flags to force or skip auto-update (--update, --no-update)
  • XDG-style global directories
  • one-line install script

Agent execution

  • single-pass lifecycle with resolve/prepare/generate/finalize phases
  • native end_turn completion (turn ends on a step with no tool calls)
  • pre/post-tool-call effect pipeline (auto-install deps, format, lint)
  • workspace profile detection with auto-detected install, lint, format, and test commands
  • configurable model reasoning level (low, medium, high) with provider-specific mapping
  • multi-provider support (OpenAI, Anthropic, Google, Vercel)
  • unified provider auth via acolyte auth — API key or subscription (OpenAI OAuth), global credentials, status and logout
  • provider rate limit awareness with sliding window pacing and exponential backoff
  • provider prompt-cache awareness with cached input token reporting
  • proactive token budgeting with system prompt reservation and priority-based allocation
  • step budget enforcement for cost protection
  • streaming progress output with real-time token usage
  • inline tasklist for multi-step tasks

Tools

  • find/search/read files with gitignore awareness; reads return the whole file, with an explicit line range for files over the token ceiling
  • edit/create/delete files
  • AST-based structural code editing
  • git status/diff/log/show/add/commit
  • GitHub CLI integration for PR and issue management (view/create/edit), auto-enabled when gh is installed
  • on-demand session search across conversation history
  • shell and test execution
  • web search/fetch
  • MCP client: connect to external services via stdio or HTTP MCP servers

Memory

  • three-scope persistent memory (session, project, user)
  • automatic observation via distiller (tool-based, runs after generation)
  • self-bootstrap project memory from codebase on first run
  • topic tags on observations for filtered recall
  • on-demand memory toolkit (memory-search, memory-add)
  • memory recalled on-demand via tools (not injected into the system prompt)
  • semantic recall with embeddings and cosine similarity ranking
  • hybrid retrieval scoring (cosine similarity + TF-IDF token overlap) with search result re-ranking
  • archive and restore retired memory with supersession lineage

Safety and control

  • workspace sandbox boundary enforcement for filesystem access
  • cooperative interruption and queued message handling
  • acolyte stop and restart refuse while work is unfinished, naming it; --force overrides

Diagnostics

  • lifecycle trace with SQLite-backed indexed queries
  • structured logs with level, session, and time filtering
  • token usage reporting with prompt and prompt-cache breakdown per turn
  • status command with JSON output
  • scoped debug logging with wildcard tag matching

Feature-flagged

Implemented but gated behind feature flags. See Configuration for setup.

  • syncAgents — sync AGENTS.md into project memory for on-demand recall
  • undoCheckpoints — session-level undo via write-tool checkpoints
  • workspaces — manage git worktrees and workspace-scoped sessions via /workspaces
  • cloudSync — portable memory and sessions across machines via acolyte login and acolyte logout
  • Postgres session storage backend and Postgres + pgvector memory backend (used by cloud tier)