Bang is an AI-native terminal with a real agent inside — it runs commands, reads and edits your files, and finishes the task. Bring your own models, or run them fully self-hosted. Own your backend and your data.
➜ bang "add rate limiting to the login route" ● agent reading src/routes/auth.js ● agent grep "loginHandler" → 2 matches ● agent editing src/routes/auth.js +18 −2 ● agent running npm test ✓ 42 passing ✓ Done. Added a 5 req/min limiter and a test. ➜ ▍
Works with OpenAI, Anthropic, Google, and any OpenAI-compatible endpoint — Ollama, vLLM, or your own gateway.
Bang closes the loop — it plans, runs tools on your machine, checks its work, and keeps going until the task is done.
Runs shell commands, reads files, greps your codebase, and applies edits — with your approval — then verifies the result. Real actions, not suggestions.
Point Bang at OpenAI, Anthropic, Google, or any OpenAI-compatible endpoint. Configure providers right in the UI — no JSON files to hand-edit.
Bang's harness is self-hostable. Run inference against models on your own infrastructure and keep every conversation on servers you control.
Built on a GPU-accelerated, Rust-based terminal. Blocks, workspaces, and a keyboard-first UI that keeps up with you.
Accounts, workspaces, and shared configuration — so your whole team runs the same agent against the same guardrails.
Every command the agent wants to run is shown for approval. Nothing touches your machine without a green light.
Frontier APIs when you want the best. Self-hosted models when you need privacy, cost control, or offline. Switch anytime — Bang doesn't lock you in.
Describe the task in plain language, right in your terminal.
Bang plans, then shows each command and edit before it runs anything.
It executes, checks the result, and iterates until the job is finished.
We're onboarding teams in waves. Drop your email and we'll send your invite.