Overview
Droid is the AI coding agent from Factory AI, an “agent-native software
development” platform. Droid CLI is the terminal entry point: it takes a
natural-language task, plans it, writes and tests code across a repository,
and opens a pull request — “one prompt to PR.” The same agent is also
available inside VS Code, JetBrains, a browser, and Slack/Teams, so the CLI
is one surface among several rather than the whole product.
Key Features
- Model flexibility (BYOK) — no vendor lock-in: run tasks against Claude,
GPT, Gemini, or other providers per task; as of 2026 bring-your-own-key
support extends to Ollama, OpenRouter, Fireworks, and DeepInfra. - Codebase comprehension — agentic search that understands an entire
repository well enough to coordinate changes across hundreds of files in
one task. - Spec Mode — a planning step that produces a reviewable spec before the
agent starts editing, for tasks where you want to check the approach first. - Missions — longer-running, multi-step autonomous jobs (as opposed to a
single supervised edit), plus custom droids, skills, and plugins for
extending behavior. - Adjustable autonomy — a permission model ranging from fully supervised
(review every change) to autonomous execution, aimed at making the tool
usable in security-conscious enterprise settings. droid exec— a non-interactive/scriptable execution mode for running
Droid from CI or other automation, alongside native git worktree support
for running multiple tasks in parallel.- Droid Control (graduated from experimental to stable in 2026) — lets
the agent operate browsers and desktop apps to produce QA reports and demo
videos, extending it beyond pure code editing. - Automatic AI-powered code review, security scanning, and PR descriptions
on pull requests.
Performance & Pricing
- Ranked #12 on Terminal-Bench with 77.3% (±2.2) accuracy (Version 2.0,
tested 2026-02-24). - Freemium pricing: a free tier, with paid plans in the $20–200/month range.
Positioning
Droid competes directly with Claude Code, OpenAI Codex,
and other terminal-based coding agents. Its differentiators are explicit
multi-model support (no single-provider lock-in) and a broader “agent-native”
platform story — the same Droid agent works across CLI, IDE, browser, and
chat, rather than being CLI-first.