| I built Browser Terminal Use for AI Agents. Project: https://github.com/chaokunyang/browser-terminal-use The goal is simple: let an Agent run iterative loops from local code, but execute commands inside a browser-hosted cloud terminal. With this tool, an Agent can: - send commands from local CLI
- execute inside a bound browser terminal tab
- stream output in real time
- get the exact remote exit code
- use queue / timeout / cancel for stable loops The key value for me is building a cloud-side verifiable execution environment for local LLM Agents: - execution happens in a remote/browser terminal context
- results are observable and auditable from local workflows
- command lifecycle is structured, deterministic, and script-friendly It is also very useful when using frontier models locally to debug cloud GPU issues (driver/toolchain/env mismatches, flaky jobs, setup
regressions) while keeping a tight local Agent loop. Would love feedback on Agent workflows, reliability across terminal vendors, and ideas for stronger verification primitives. |