Trunchbull

The product is a tool-native benchmarking platform that deploys tools from GitHub repositories and runs them against models via OpenRouter to evaluate performance, detect regressions, and inspect decisions under controlled budgets with full run traces. It sells to developers and B2B engineering teams building and evaluating AI agents and model-tool integrations. It is delivered as a SaaS platform using isolated Cloudflare Workers and container sandboxes with BYOK, versioned deployments, and enforced hard limits on steps, tokens, and spend.

Key features

• Deploy tools from GitHub URL
• Isolated Cloudflare Worker deployment
• Vercel AI SDK-compatible schema
• Versioned reproducible deployments
• Remote execution with hard timeouts
• Container isolation with resource limits
• File, command, process, terminal tools
• Image-backed repeatable environments
• BYOK through OpenRouter
• Per-model pricing and token accounting
• Hard limits on steps, tokens, spend
• Deterministic stop reasons
• Side-by-side model traces
• Tool-call and argument inspection
• Regression-ready run history
• Benchmark against any model