What is Bullet?
Bullet is a coding agent built for speed, designed to eliminate the frustrating latency that slows down AI-assisted development. Instead of making models smarter, Bullet makes the entire loop around them leaner—routing, searching, and executing with one purpose: keeping up with you. It works with your existing Claude Code or Codex subscription, API keys, or even an on-device model, and it's currently available for macOS and Linux in private beta with free access.
Who it's for
- Professional developers who spend hours daily in agent loops and want that time slashed without changing their workflow.
- Teams already using Claude Code or Codex who want a drop-in speed boost without abandoning their current subscriptions or API setups.
- Performance-focused engineers who care about benchmarks and measurable gains, like Bullet's 95.8% on SWE-bench Verified and 119s/task average.
Key features
Smart routing and escalation
Bullet automatically picks the right model and reasoning level for each prompt. Straightforward tasks go to fast models, complex ones escalate to deeper reasoning, and the hardest missions get full attention. You never manually juggle model choices—the agent decides in the moment.
Targeted code search
Instead of embedding your entire repository, Bullet uses targeted search and file reads to find only what's relevant. It acquires the target and ignores the noise, which means faster context building and less wasted compute on irrelevant files.
Parallel execution
Independent tool calls—search, read, verify—run simultaneously rather than in sequence. Bullet also intercepts duplicate calls and stuck loops before they waste another second. No dead time, no queuing what could run together.
CLI access
The same router, tools, and agent loop are available in your shell via npm install -g @trybullet/cli. It's free, requires no key to start, and runs on macOS and Linux with Node 18+.
What stands out
Don't let your code hold you hostage.
That's the core philosophy: Bullet isn't about magic—it's about momentum. The founders built it out of genuine frustration with slow agent runs, and it shows in every design choice. The 95.8% SWE-bench Verified score (top 3) proves the speed doesn't come at the cost of quality, and the 119s/task average makes it one of the fastest agents on the market. Backed by Y Combinator and built by Yale CS grads with experience at AppLovin and Citadel, Bullet has the pedigree to back up its claims.
Worth checking out if…
You're tired of waiting on agent runs and want a faster loop without switching models or subscriptions. If you already use Claude Code or Codex and feel the machinery around the models is the bottleneck, Bullet is a direct answer. It's also worth a look if you're benchmark-driven and want a coding agent that ranks in the top tier on SWE-bench Verified while being dramatically faster per task. And since it's free during the private beta, there's little reason not to try it.
