Factory Droid: Review and Setup Guide (2026)

Factory Droid has been quietly building momentum among developers who want something more than a chat-box bolted onto an editor. It ships as a full-screen terminal UI, runs headless in CI, supports multi-agent orchestration through a feature called Missions, and lets you swap between Anthropic, OpenAI, Google, and a growing roster of open-weight models mid-session. The result is an agent that feels less like a coding assistant and more like a programmable colleague you can leave running overnight.

The platform has matured considerably in 2026. BYOK support now covers Ollama, OpenRouter, Fireworks, DeepInfra, and more. The Droid Control plugin, which lets the agent operate browsers and desktop apps to produce QA reports and demo videos, graduated from experimental to a stable plugin. And the pricing model was simplified around rolling rate limits that are easier to reason about than the per-seat token buckets common elsewhere.

Last updated: June 10, 2026

What Factory Droid Actually Is

Droid is the agent layer inside Factory's platform. You interact with it through three surfaces: the Droid CLI (a full-screen TUI you run in any terminal), the Factory desktop app (which wraps the same agent with a GUI), and droid exec for headless non-interactive runs in CI or cron jobs.

The CLI is the heart of the product. When you run droid inside a repo, the agent reads your codebase, proposes changes with a diff view, and waits for your approval before touching anything. You stay in full control: accept, reject, or edit each change before it lands. The agent follows conventions you document in an AGENTS.md file at the repo root, so it learns your team's patterns rather than imposing its own.

Beyond single-session work, Factory Missions add a planning layer on top. You describe a large project, collaborate with Droid to build a feature-and-milestone plan, and hand off execution to an orchestration layer that manages a team of worker subagents. Missions are not fire-and-forget - you act as project manager, monitoring progress and redirecting workers when they get stuck.

Quick-Start Setup

Install takes under two minutes on macOS or Linux:

# Install
curl -fsSL https://app.factory.ai/cli | sh

# Linux only: install xdg-utils if not present
sudo apt-get install xdg-utils

# Navigate to your project and start
cd /path/to/your/project
droid

On first launch, a browser tab opens for authentication. Once signed in, you are dropped into the full-screen TUI. From there, type a task in plain English and press Enter. A few useful commands to know on day one:

• /settings - configure models, permissions, and defaults
• /model - switch models mid-session
• /review - trigger an AI-powered code review workflow
• /missions - open the Missions planning interface
• /limits - check your current rate limit status

Teach Droid your repo conventions. Drop an AGENTS.md in your project root with your stack, coding standards, test commands, and anything Droid should know before touching code. Workers in Missions also inherit this file automatically.

Model Routing and BYOK

Factory manages a broad model roster across Anthropic, OpenAI, and Google, each with a cost multiplier relative to your plan's base usage. A few highlights from the current lineup:

Switch models mid-session with /model. For Missions, you can pin a strong model to the orchestrator and a lighter model to workers - a common cost-quality tradeoff since planning and validation need more reasoning than routine edits.

BYOK adds any OpenAI-compatible provider. Add entries to ~/.factory/settings.json under customModels:

{
  "customModels": [
    {
      "model": "qwen3:27b",
      "displayName": "Qwen3 27B (local)",
      "baseUrl": "http://localhost:11434/v1",
      "apiKey": "${OLLAMA_API_KEY}",
      "provider": "generic-chat-completion-api"
    }
  ]
}

API keys stay local and are never uploaded to Factory servers. Factory warns that models under 30 billion parameters show significantly lower performance on agentic coding tasks, so local BYOK works best for experimentation rather than production use.

Factory Missions: Multi-Agent Projects

Missions are the standout feature for larger work. The workflow has five stages: you run /missions, collaborate with Droid to define features and milestones, let it pull in relevant skills and develop new ones for the work, approve the plan, and then let Mission Control manage execution.

The planning phase matters most. Factory is explicit about this: a well-scoped plan with clear milestones produces dramatically better results than jumping straight into execution on a vague goal. Droid will ask clarifying questions and push back until the scope is solid.

Missions run headless in CI too:

droid exec --mission -f mission.md

# Tune model/reasoning per agent type
droid exec --mission \
  --worker-model claude-sonnet-4-6 \
  --worker-reasoning-effort medium \
  --validator-model claude-opus-4-7 \
  --validator-reasoning-effort high \
  -f mission.md

Missions require Extra Usage to be enabled (see pricing below). Rate limits pause the mission rather than cancel it, so a runaway multi-agent run will not silently drain your account.

Droid Control: Browser and Terminal Automation

Droid Control is an optional plugin that gives the agent eyes and hands. Install it with:

droid plugin marketplace add https://github.com/Factory-AI/factory-plugins
droid plugin install droid-control@factory-plugins --scope user

It adds three commands:

• /demo pr-1847 - records a before/after comparison video of a pull request, rendered with title cards, transition effects, and optional keystroke overlays
• /verify "ESC cancels streaming in bash mode" - tests a specific behavior claim and returns a CONFIRMED, REFUTED, or INCONCLUSIVE verdict with evidence
• /qa-test https://app.example.com -- login, create a project, invite a member - drives a web or Electron app through an end-to-end flow and reports pass/fail per step

The agent is framed as an investigator, not an advocate: if a claim is false, that is a valid finding. For web and Electron targets, install the browser driver separately:

agent-browser install   # downloads Chromium

Pricing (Verified June 2026)

Factory uses rolling rate limits across three windows: 5-hour, weekly, and monthly. Hitting any one cap blocks new requests until that window resets. All windows are rolling from first use, not calendar-based.

When Standard Usage runs out, two fallback options kick in. Droid Core is a free tier of open-weight models (Kimi K2.6, DeepSeek V4 Pro, MiniMax M2.5, and others) with their own separate rate limits. Extra Usage is a prepaid credit balance starting at $10 minimum that draws down as you use models - credits never expire, and the toggle is sticky so you do not need to re-enable it each billing cycle.

Teams and Enterprise plans are not subject to rate limit changes, which matters for teams running long autonomous workflows.

BYOK usage has a free allowance on all plans. After the allowance, BYOK usage is charged per your specific plan.

Who Should Skip This

Droid is worth trying if you spend meaningful time in the terminal, work across large or multi-service repos, or want CI-integrated code review without wiring up a separate service. The Missions workflow is genuinely useful for scoped projects that benefit from upfront planning.

Skip it if you want a lightweight inline suggestion tool. Droid is a full-screen TUI agent, not an autocomplete layer. VS Code users who primarily want inline completions will find the overhead unnecessary - dedicated extensions handle that better.

Also skip it if you need a firm token budget guarantee. The rolling rate limits are clean, but if you have strict cost requirements and are on a Pro or Plus plan, it is possible to exhaust quota mid-Mission. Enterprise and Teams plans avoid this, but at a price point that requires a sales conversation.

Finally, the Missions feature requires Extra Usage to be active. If you are uncomfortable with open-ended prepaid credits, Missions as a workflow is off the table until that changes.

FAQ

How is Factory Droid different from Claude Code or Cursor?

Factory Droid is a standalone terminal agent with its own TUI, model routing layer, and headless execution mode. Claude Code is tightly coupled to Anthropic's model family. Cursor is editor-first and built around inline suggestions. Droid sits closer to Claude Code in the agentic style but adds multi-model support, Missions orchestration, and the Droid Control browser automation layer that neither competitor ships natively.

Can I use my own Anthropic or OpenAI API key?

Yes. BYOK is supported on all plans with a free allowance. Add your key to ~/.factory/settings.json using the customModels array and the anthropic or openai provider type. Your keys stay local and are not uploaded to Factory servers.

What happens when I hit a rate limit mid-session?

The current request finishes, then subsequent requests are refused. You can run /limits in the CLI to see your status and toggle Droid Core or Extra Usage without reopening the session. For Missions, the orchestrator pauses and surfaces the error rather than silently stopping.

Is the AGENTS.md file required?

No, but it is worth the ten minutes it takes to write. Droid reads it before every session and passes it to Mission workers automatically. Without it, the agent falls back to inferring conventions from the codebase, which works but produces less consistent results on team projects or repos with strong stylistic conventions.

Sources

• Factory documentation: https://docs.factory.ai/welcome
• Droid CLI quickstart: https://docs.factory.ai/cli/getting-started/quickstart
• Plans and pricing (verified June 10, 2026): https://docs.factory.ai/pricing
• Available models: https://docs.factory.ai/models
• BYOK overview: https://docs.factory.ai/cli/byok/overview
• Factory Missions: https://docs.factory.ai/cli/features/missions
• Droid Control: https://docs.factory.ai/cli/features/droid-control