Co‑founders Willy and Dan unveiled Twill.ai, a service that runs coding CLIs such as Claude Code and Codex inside isolated cloud sandboxes. Users submit tasks via Slack, GitHub, Linear or a web UI, and the platform returns a pull request, a code review, or a diagnostic query. The system only interrupts the developer when clarification is required, keeping control in human hands in production.

They built Twill to solve three pain points they hit with local agents: inability to parallelize tasks that touch the same Docker configuration, loss of state when a laptop sleeps, and security concerns over giving an autonomous model full filesystem access. By spawning a fresh sandbox per job, the service offers persistent environments, port isolation, and runtime‑injected secrets for CI pipelines.

Pricing starts with a free tier that grants ten credits per month—each credit equals one dollar of AI compute at cost—followed by plans from $50/month for fifty credits and support for bring‑your‑own‑key on higher tiers. An open‑source SDK, agentbox‑sdk, lets developers run any lab‑native CLI in the cloud, positioning Twill as a vendor‑agnostic bridge for automated PR generation for engineering teams.