OpenChamber: The Cockpit OpenCode Never Shipped
OpenCode has a thriving SDK and no real cockpit, so the community built one. OpenChamber landed on the Hacker News front page today: an agentic development environment that wraps OpenCode with the stuff you actually need for long-running agent work, across desktop, browser, phone, and VS Code. It is open source, free, and explicitly not affiliated with the OpenCode team.
The feature that matters most is Session Goals: you set an objective, and after every turn OpenChamber checks the result and keeps the agent working until the goal is done, blocked, or hits a limit you set β even with the app closed. Add Multi-run, which hands the same task to up to five different models in parallel sessions and lets you fuse the best pieces of each, plus cron-scheduled prompts, a changes walkthrough that groups big diffs into ordered explained steps, and GitHub integration that goes from issue to merged PR without leaving the app. Code and sessions stay local; remote access is password-protected.
The pattern here rhymes with what T3 Code is doing on the multi-harness side: the model layer and the harness layer are getting commoditized, and the new competition is the surface you actually live in β goals, scheduling, review, mobile access. That an independent project can assemble this on top of someone else's SDK in months tells you where the leverage in the stack has moved.
At openchamber.dev, code at github.com/openchamber/openchamber.
← Back to all articles
The feature that matters most is Session Goals: you set an objective, and after every turn OpenChamber checks the result and keeps the agent working until the goal is done, blocked, or hits a limit you set β even with the app closed. Add Multi-run, which hands the same task to up to five different models in parallel sessions and lets you fuse the best pieces of each, plus cron-scheduled prompts, a changes walkthrough that groups big diffs into ordered explained steps, and GitHub integration that goes from issue to merged PR without leaving the app. Code and sessions stay local; remote access is password-protected.
The pattern here rhymes with what T3 Code is doing on the multi-harness side: the model layer and the harness layer are getting commoditized, and the new competition is the surface you actually live in β goals, scheduling, review, mobile access. That an independent project can assemble this on top of someone else's SDK in months tells you where the leverage in the stack has moved.
At openchamber.dev, code at github.com/openchamber/openchamber.
Comments