Overview
On August 10, OpenChamber (https://openchamber.dev/) publicly launched as an open-source "agent development environment," positioning itself as a complete replacement for developers' three parallel choices: IDE, TUI, and Web UI for AI coding. The project frames itself around one rule: OpenCode is the harness, OpenChamber is the UI/runtime.
Six Core Capabilities
- Session Goals — set a goal and the agent works across multiple conversation turns toward it, even when the app is closed.
- Multi-run and Fusion — run the same task on up to 5 models simultaneously; keep the best result or "fuse the strongest parts."
- Changes Walkthrough — a large diff is broken into ordered, explained steps showing how the change was built.
- Preview — click an element in your running app to send all its underlying context to the agent.
- From issue to pull request — start from a GitHub issue or PR, feed failing CI back to the agent, and merge without leaving OpenChamber.
- Scheduled work — cron-style prompts combined with Session Goals for observable outcomes.
- Harness = OpenCode SDK (installed via
curl -fsSL https://opencode.ai/install | bash). OpenChamber doesn't reinvent the agent loop; it picks the best-experience open harness. Swap harnesses without swapping UI — a fundamental contrast with the VS Code + Copilot camp, which bundles harness, runtime, UI, and model choice. - Runtime = OpenChamber — session goals, multi-model parallelism, change walkthroughs, scheduled tasks, remote access, cross-device sync, and Private Relay E2E encryption all run on the user's own machine. Project names, paths, prompts, code, diffs, and session content are never collected.
- Remote access security model — browser access can be gated by a UI password; tunnel links are rotatable and revocable; Private Relay uses no public ports, one-time QR pairing, end-to-end encryption, revocable anytime.
- Privacy as auditable code, not marketing — fully open source; "the model can't see you" is verifiable in the GitHub repo, not a ToS clause.
- VS Code camp (Cursor, Windsurf, Copilot) — binds harness, UI, runtime, and models; switching tools means starting over.
- TUI camp (Claude Code, Codex CLI, aider, Goose) — lightweight harness but pushes all session management (worktrees, commits, PRs, review) onto developers.
- Web camp (Devin, Replit Agent, v0) — full session experience on the web, but users are locked into vendor clouds.
- License ambiguity — the project is open source but the FAQ doesn't specify a license; check the repo's LICENSE file before commercial use.
- OpenCode SDK dependency — custom backends will only open up once the OpenCode SDK API stabilizes.
- Mobile apps are beta — long-running sessions on phones may be less stable than desktop.
- Private Relay — QR pairing is elegant, but stable NAT traversal on weak networks (corporate intranets, hotel Wi-Fi) needs real-world testing; some users report occasional timeouts on mobile networks.
- Multi-model cost — running 5 models per task means 5x token cost for one result, and Fusion adds its own instability.
- OpenChamber: https://openchamber.dev/
- FAQ: https://openchamber.dev/#faq
- Documentation: https://docs.openchamber.dev/
- OpenCode SDK install: https://opencode.ai/install
- Hacker News Chinese edition: https://buzzing.cc/
Cross-Platform Experience
Native desktop apps for macOS / Windows / Linux, plus PWA browser access, mobile/tablet-friendly interaction, and iOS/Android native apps (beta). The underlying layer offers VS Code context-menu integration, an Agent Manager for multi-model parallelism, and direct file opening. An MCP server ecosystem allows connection from Claude Code, Codex, Qoder, OpenClaw, Qwen Code, Gemini CLI, opencode, pi, and QwenPaw.
The Architectural Bet: Harness vs. Runtime Boundary
Why It Matters: Three Competing Forces
OpenChamber takes a fourth path: TUI-class harness + desktop/web/mobile UI + local-first runtime. Before August 10 these pieces existed separately (OpenCode as harness, Chrome extensions as UI, cron tools as schedulers) but were never unified. The thesis matches the broader shift: the next stage of AI coding toolchains is not making single tools smarter, but letting the platform layer and harness layer evolve decoupled.