Pilot
#1 Terminal Benchmark 2.0 — AI that ships your tickets.
Last verified:
What is Pilot?
Pilot is an autonomous development pipeline AI that ships software tickets while developers focus on architecture. It picks up tickets from GitHub Issues, GitLab Issues, Linear, Jira, Asana, Plane, or Discord, plans implementation using codebase context, writes code with Claude Code, runs quality gates, and opens pull requests for human review and merge. With 260+ features production-ready, it achieved #1 on Terminal-Bench 2.0 with an 82% pass rate across 445 real-world software tasks.
Key features include ticket-driven workflow (not prompt-driven), context intelligence that reduces token usage by 92%, autopilot CI loop with auto-merge and failure feedback, self-hosted deployment with hot upgrades, multi-platform support, Telegram bot for mobile execution, quality gates with automatic retry, self-review code review before PR push, epic decomposition for complex tickets, model routing for cost efficiency, and a terminal UI dashboard showing token usage and costs.
Pilot is designed for solo developers handling backlogs, team leads managing multiple tickets, and organizations wanting to codify engineering standards. It works best on well-defined tasks like CRUD endpoints, React components, test coverage, documentation updates, bug fixes with clear reproduction steps, dependency updates, and straightforward database migrations. It is not intended for architecture decisions or novel algorithms requiring engineering judgment.
Pilot pricing
Pricing model: Freemium
Pilot is source-available under BSL 1.1 license. It is free for internal use and self-hosting. There is no separate paid tier - you use your existing Claude subscription (Pro/Max) or Anthropic API key. No separate API key is required if you're already logged into Claude Code. You can inspect the code, audit what it does, and contribute improvements.
Pilot pros
- #1 on Terminal-Bench 2.0 with 82% pass rate across 445 real-world tasks
- 260+ features production-ready out of the box
- Picks up tickets within 30 seconds of labeling
- Works with existing ticket systems (GitHub, GitLab, Linear, Jira, Asana, Plane, Discord)
- No prompting or babysitting required
- Context intelligence reduces token usage by 92%
- Autopilot CI loop with auto-merge when tests pass
- Self-hosted so secrets never leave your environment
- Hot upgrades itself without downtime
- Uses your existing Claude subscription (Pro/Max) - no separate API key needed
- Single Go binary - easy deployment
- Epic decomposition automatically splits complex tickets into subtasks
- Self-review code review runs before PR push
- Model routing uses fast models for trivial tasks, deep reasoning for complex ones
- Telegram bot allows chat, research, and task execution from mobile
- 40% token savings on self-review via session resume context continuation
- Quality gates include test, lint, and build with automatic retry
- Multi-repo polling manages multiple repositories from single instance
- Notifications via Slack, Telegram, email, webhooks, and PagerDuty
Pilot cons
- Requires Claude subscription (Pro/Max) or Anthropic API key
- Self-hosted only - no cloud SaaS option available
- BSL 1.1 license restricts commercial redistribution
- Not suitable for architecture decisions or novel algorithms
- Requires GitHub/GitLab token configuration
- Desktop app requires starting Pilot daemon in terminal first
- Only labels with 'pilot' trigger automatic execution
- Autopilot modes require CI infrastructure to be set up
- Limited to well-defined, bounded work tasks
Frequently asked questions about Pilot
What is Pilot?
Pilot is an autonomous development pipeline with 260+ features production-ready. It picks up tickets from GitHub Issues, GitLab Issues, Linear, Jira, Asana, Plane, or Discord — plans the implementation using your codebase context, writes code with Claude Code, runs quality gates, and opens a pull request. You review and merge. No prompting or babysitting required.
What AI model does Pilot use?
Pilot works with your Claude subscription (Pro/Max) or Anthropic API key. It uses Claude Code to write code. No separate API key is required if you're already logged into Claude Code. Pilot also features model routing where trivial tasks use fast models and complex tasks use deep reasoning.
What makes Pilot different from other AI coding tools?
Pilot is ticket-driven, not prompt-driven. Other AI tools need you typing prompts and reviewing suggestions, but Pilot works from your existing ticket system. Write the ticket like you would for a junior developer and Pilot handles the rest. It also has context intelligence reducing token usage by 92%, autopilot CI loop with auto-merge, and is self-hosted with hot upgrades.
What tasks is Pilot best at?
Pilot works best on well-defined, bounded work: CRUD endpoints and React components, test coverage for existing code, documentation and README updates, bug fixes with clear reproduction steps, dependency updates and security patches, and database migrations with straightforward schema changes. It is not for architecture decisions, novel algorithms, or anything requiring engineering judgment.
How does the autopilot CI loop work?
In autopilot mode, Pilot monitors CI, auto-merges when tests pass, and automatically creates fix issues when CI fails — targeting the original branch for fast iteration. There are three modes: dev (fast), stage (CI required), and prod (human approval required). Pilot doesn't stop at creating a PR.
Is Pilot secure?
Yes, Pilot is self-hosted on your infrastructure. Your secrets never leave your environment. It runs as a single Go binary. Pilot is source-available under BSL 1.1, so you can inspect the code, audit what it does, and contribute improvements. PR review provides human control.
How does context intelligence work?
Pilot's built-in context engine reduces token usage by 92% while following your architecture, patterns, and conventions. It auto-detects your project's .agent/ directory and uses it for strategic documentation loading, giving Pilot deep codebase understanding rather than generic code generation. Session resume provides 40% token savings via context continuation.
Can I use Pilot with multiple repositories?
Yes, Pilot supports multi-repo polling, allowing you to manage multiple repositories from a single instance. You can configure it to poll GitHub or GitLab repositories and handle tickets from any of them.
How do I upgrade Pilot?
Pilot has hot upgrade capability - it self-updates without downtime through in-process binary replacement. Press 'u' in the dashboard and the binary replaces itself in-place. This is one of the 260+ production-ready features.