microagents-cli 0.1.0

Batteries-included terminal-based agent built on top of the microagents framework
docs.rs failed to build microagents-cli-0.1.0
Please check the build logs for more information.
See Builds for ideas on how to fix a failed build, or Metadata for how to configure docs.rs builds.
If you believe this is docs.rs' fault, open an issue.

microagents-cli

A batteries-included, terminal-based AI agent built on top of the microagents framework.

What it is

microagents-cli (binary: microag) is a Rust CLI that gives you an interactive coding agent inside your terminal. It combines a local vector search index with an LLM-powered agent that can read, write, edit, search, and execute shell commands in your workspace.

Key features

  • Interactive TUI — Chat with the agent in a minimal, keyboard-driven terminal UI built with Ratatui. Supports multi-line input, scrollback, and session resume.
  • Headless mode — Run a single prompt and get JSON-RPC output for scripting or piping.
  • Local semantic search — Automatically indexes files in your workspace using dense + sparse embeddings (Model2Vec + BM25) and stores them in an on-disk Qdrant Edge vector database.
  • Smart file ingestion — Parses code with AST-aware chunking (via astchunk) and unstructured documents (PDFs, Office files, images) via liteparse.
  • Built-in tools — The agent can:
    • read — Read any file (including PDFs, Word docs, images via OCR)
    • write — Create or overwrite files
    • edit — Make precise text replacements
    • search — Semantic search across the indexed workspace
    • shell_execute — Run shell commands with a safety regex filter
  • Session persistence — Save conversation history to JSONL or SQLite and resume later.
  • Skills system — Auto-discover or explicitly load agent skills from the microagents framework.

Usage

Interactive TUI

microag

Headless / single prompt

microag --prompt "Refactor the auth module to use async/await"

Resume a session

microag --session-id <uuid>

Options

Flag Description
--model LLM model to use
--provider Provider (defaults to openrouter)
--storage jsonl or sqlite (default: jsonl)
--skill Load specific skills (repeatable)
--prompt Run in headless mode
--verbose Print initialization info

Workspace indexing

On startup, the CLI scans your current directory (respecting .microagentsignore), computes a diff against the last run, and incrementally updates the local vector index stored in .microagents/. This lets the agent search and reason over your entire codebase.

Architecture

  • src/main.rs — CLI argument parsing, agent builder, TUI/headless dispatch
  • src/tui/mod.rs — Full Ratatui chat interface with streaming output
  • src/init_env.rs — File walking, diffing, ingestion, and vector store setup
  • src/processing.rs — Chunking (AST + text) and embedding (dense + sparse)
  • src/search.rs — Hybrid dense/sparse vector search with RRF fusion
  • src/tools.rs — Tool implementations: read, write, edit, search, shell_execute

License

MIT