droid

Start an interactive session

$ droid

$ droid exec "[analyze this code and fix bugs]"

$ droid exec --auto high "[fix bug, test, commit, and push]"

$ droid exec -f [prompt.txt]

$ droid exec -m [claude-3-opus] "[your prompt]"

$ droid exec -o json "[your prompt]"

$ droid exec -s [session_id] "[continue the task]"

droid [options] [command]

Droid is Factory.ai's enterprise-grade AI coding agent for the terminal. It handles end-to-end development workflows including code analysis, feature implementation, bug fixing, code review, and automated testing.
By default, Droid runs in spec-mode where the agent only executes read-only operations. Use --auto to enable edits and command execution with configurable risk tiers. The tool supports custom sub-agents (Custom Droids) defined in .factory/droids/ as markdown files with YAML frontmatter.
Droid achieves state-of-the-art performance on Terminal-Bench benchmarks, outperforming other CLI agents. It provides a professional trial plan with access to Claude and OpenAI models upon account creation.

-v, --version

Output the version number

Display help for command

-o, --output-format FORMAT

Output format (default: text)

Input format: stream-json for multi-turn sessions

Read prompt from file

Autonomy level: low, medium, or high

Skip ALL permission checks (dangerous)

Continue an existing session

Model ID to use

Reasoning effort level

Start in spec mode (read-only operations)

Enable specific tools

Requires Factory.ai account authentication. The --skip-permissions-unsafe flag allows all operations without confirmation and should only be used in isolated environments like Docker containers. High autonomy mode (--auto high) can make destructive changes.

Droid was developed by Factory.ai as an enterprise AI coding agent. It has achieved leading performance on Terminal-Bench with a score of 58.75% using Opus, surpassing other CLI agents including Claude Code and Codex CLI.

claude(1), codex(1), aider(1), opencode(1)