An autonomous agent that ships code, not opinions.
NanoAgent is our strategic bet on the agentic turn — a coding agent engineered for real engineering teams. It runs inside OS-level sandboxes, executes tools in parallel, navigates code semantically through an LSP backend, and remembers what it learned the last time something broke. Everything is logged. Everything is auditable.
- Sandboxed by default — bwrap on Linux, sandbox-exec on macOS.
- ×4 parallel tool execution — concurrent calls without race conditions.
- LSP-backed intelligence across TypeScript, Python, Rust, Go, C#, C/C++.
- Persistent lesson memory — classifies failures, carries lessons forward.
- MCP-ready — dynamic tool integration via the Model Context Protocol.
verifyJWT() → auth/jwt.tsmiddleware.spec.ts