⏱️ Lectura: 11 min
OpenChamber now lets you run the same coding task on up to five AI models at once, and lets you keep the best result or merge what each model did best. The project is open source, runs on the OpenCode SDK, and moves seamlessly between the desktop app, the browser, your phone, and your code editor.
📑 En este artículo
The core idea behind OpenChamber is that the agent doesn’t stop when you close the app: you give it a goal (Session Goal) and it keeps iterating turn after turn until it’s done. On top of that, it adds a flow that goes from a GitHub issue to a merged pull request without leaving the tool.
TL;DR
- OpenChamber is an open source agentic development environment, with a native app for macOS, Windows, and Linux.
- Multi-run and Fusion: runs the same task on up to five AI models at once and merges or compares results.
- Runs on the OpenCode SDK, which is installed with
curl -fsSL https://opencode.ai/install | bash. - Session Goals keeps the agent iterating toward a goal turn after turn, even with the app closed.
- Supports the full GitHub issue-to-pull-request flow, including rerunning failed checks without leaving the app.
- Remote access uses Private Relay: QR code pairing, end-to-end encryption, no open ports.
- It doesn’t collect or send code, prompts, diffs, or session content: everything stays on your local machine.
- It has a native mobile app in beta and browser/PWA access with a password gate for exposing it securely.
Introduction
Agentic development environments (tools where an AI agent doesn’t just autocomplete code but plans, edits files, runs commands, and opens pull requests) multiplied between 2025 and 2026. OpenChamber falls into that category with a different approach: instead of betting on a single model, it runs the same task on several at once and compares or merges the results.
The project doesn’t replace the agent engine; it runs on the OpenCode SDK, which its creators describe as the best open source agent experience available today. Early adopters sum it up like this on social media: “Opencode + OhMyOpencode + Openchamber. This is some cooked stack. VSCode looks like legacy notepad++ now.”
What happened
OpenChamber added, in a single release, six features that used to live scattered across different tools: Session Goals, Multi-run and Fusion, Changes Walkthrough, Preview, the issue-to-pull-request flow, and scheduled cron work. Each one tackles a specific problem in daily work with AI agents.
- Session Goals: you set a goal and the agent keeps working toward it turn after turn, even with the app closed.
- Multi-run and Fusion: run a task on up to five different models, keep the best result or merge the strongest parts of each.
- Changes Walkthrough: a large diff gets grouped into ordered steps that explain how the whole change fits together.
- Preview: you point to an element in your running app and send the agent all the context behind that element.
- Issue to pull request: start from a GitHub issue or PR, rerun the checks that failed, and merge without leaving OpenChamber.
- Scheduled work: run a prompt on a cron schedule, combined with Session Goals to aim for a specific outcome.
Context and history
According to early users, OpenChamber started out as a small project: “I saw OpenChamber back when it was a small project, I think maybe a couple months ago, and now it is unrecognizable.”, says Harsha Kotcherlakota (hkay-dev). That accelerated evolution is typical of where agentic tools stand right now: in just over a year they went from autocompleting lines to handling entire work sessions, with memory across turns and terminal access.
The decision to build on OpenCode instead of a proprietary engine is key to understanding the project. OpenChamber isn’t competing to be the best agent; it’s competing to be the best orchestration layer around existing agents: running several in parallel, ranking them by result, and keeping them alive across sessions.
Technical details and performance
OpenChamber ships as a native desktop app for macOS, Windows, and Linux, with support for multiple windows and projects in parallel. It includes “Open In” shortcuts to open the active project in Finder, the Terminal, or your editor, and “Project Actions” to spin up dev servers, SSH forwards, and local URLs without leaving the app.
The same session stays available from the browser (with PWA support), from your phone or tablet, and from a native mobile app that’s still in beta. To expose a local instance to the internet without leaving it open to anyone, OpenChamber adds an interface password gate, designed to make browser access safe to share.
How a new device connects
Remote pairing uses Private Relay: you scan a single-use QR code and the new device connects without opening ports or exposing a public server. The connection stays end-to-end encrypted and can be revoked at any time.
~ ❯ openchamber serve --ui-password ******
┌ OpenChamber Started
│
◆ port 3000 (PID 34189)
● local: http://127.0.0.1:3000/
● tunnel: https://oc-3a9f.trycloudflare.com
│
└ daemon running
~ ❯ openchamber connect-url --relay
┌ OpenChamber relay pairing link
│
◆ openchamber://connect?v=2&p=eyJ2IjoyLCJwYWlyaW5n…
● Relay: wss://relay.openchamber.dev/ws
● Fingerprint: 7F3A-D21C
│
└ relay pairing link generatedThe first command starts the local daemon with a password for the web interface; the second generates the Private Relay pairing link, with its fingerprint to verify that the device connecting is the right one.
flowchart TD
A["Editor, browser, or phone"] --> B["openchamber serve"]
B --> C["OpenCode SDK"]
C --> D["Model 1"]
C --> E["Model 2"]
C --> F["Model N"]
D --> G["Result fusion"]
E --> G
F --> G
G --> H["Pull request on GitHub"]The diagram sums up the path of a typical task: the client (whether the desktop app, the browser, or the editor) sends the task to the openchamber serve daemon, which distributes it among the configured models through the OpenCode SDK and gathers the results before opening the pull request, if applicable.
Privacy: what stays on your machine
According to the project’s documentation, project names, paths, prompts, code, diffs, and session content are not collected or sent to OpenChamber’s servers. Being open source, that privacy model can be audited by reading the code instead of trusting a privacy policy.
📌 Note: OpenChamber is open source and its code is available on GitHub, so the privacy model it promises (no code or sessions collected) can be verified by reading the repository instead of trusting a terms policy.
Getting started
OpenChamber runs on the OpenCode SDK, so that’s the first step. The official installation uses a curl script built for macOS and Linux:
macOS and Linux
curl -fsSL https://opencode.ai/install | bashWindows
On Windows, the most direct route is to install OpenChamber’s native desktop app from its official site; if you prefer the command line, run the same script inside WSL (Windows Subsystem for Linux) to have the OpenCode SDK available from a Linux terminal.
With the SDK installed, the next step is to start the OpenChamber daemon with an interface password and, if you’re going to use it from another device, generate the pairing link:
openchamber serve --ui-password YOUR_PASSWORD
openchamber connect-url --relayThe first command exposes the local interface (and a tunnel if needed); the second gives you the QR code or link to pair your phone or another laptop over Private Relay. From there, you define a Session Goal (for example, “migrate the REST endpoints to GraphQL without breaking the tests”) and let the agent iterate, with the option to run it in parallel on two or three models with Multi-run to compare results.
To confirm the daemon is running and which port it’s on, the same output from openchamber serve shows it: look for the local: http://127.0.0.1:PORT/ line in the terminal.
Which environment fits which case
| Environment | When to use it | Advantage | Limitation |
|---|---|---|---|
| Desktop app | Daily work, several projects open at once | “Open In” shortcuts to Finder, Terminal, and editor; Project Actions to spin up local servers | You need to install and keep the binary updated |
| Browser / PWA / mobile | Checking or launching tasks from your phone or a machine that isn’t yours | Access from any device, with background notifications | Needs the password gate or a tunnel to expose it securely |
| Editor extension | When you already live inside your code editor and don’t want to switch windows | Open files straight from a tool’s output, with right-click actions | Built for one editor at a time, without the full desktop app’s Agent Manager |
Impact and analysis
The bet on Multi-run and Fusion addresses a real problem with working with a single model: each model has different biases and blind spots for the same task. Running the task in parallel and comparing (or merging) results hands the developer a decision they used to make blind, with a single attempt and a single model, with no point of comparison.
The second pillar, Session Goals, tackles another problem: today most code agents sit there waiting for a user response before continuing. Turning that into a persistent goal, one that keeps running turn after turn even if you close the app, brings the tool closer to an asynchronous collaborator than an interactive autocomplete.
💡 Tip: If you’re going to leave a Session Goal running for hours unsupervised, pair it with the issue-to-pull-request flow: that way the result ends up as a reviewable PR instead of loose changes in your working tree.
The privacy model (everything local, nothing collected, remote access optional and revocable) is, in itself, a response to a growing concern among development teams: sending proprietary code to a third party for an agent to process. The project being open source means that promise can be audited in the code, not just read on a marketing page.
What’s next
The native mobile app is still in beta, so it’s likely to gain features (and bug fixes) in upcoming versions. The official FAQ confirms the project relies on the OpenCode SDK “today”, which leaves the door open for it to add other agent engines in the future if the ecosystem pushes it in that direction.
For now, OpenChamber’s differentiator isn’t training or fine-tuning its own model, but the orchestration layer: multi-model, multi-device, with work persisting across sessions. It’s a pattern other code agent tools will likely end up copying if the formula works in practice.
📖 Summary on Telegram: View summary
Try it yourself: run curl -fsSL https://opencode.ai/install | bash and then add OpenChamber from openchamber.dev to see Multi-run working with your own models today.
Frequently Asked Questions
Is OpenChamber open source?
Yes. The project is open source and its code is available on GitHub, which makes it possible to audit both the features and the privacy model it promises.
What agent engine does it use under the hood?
OpenChamber runs on the OpenCode SDK, which its creators chose because, as they describe it, it’s the best open source agent experience available today.
Is it free?
Yes, OpenChamber itself is free. The actual cost depends on the usage of the models you run through OpenCode, each with its own per-token price.
Can I use it remotely, from my phone?
Yes. You can open it from the browser, use tunnels when needed, and protect access with an interface password or Private Relay’s encrypted pairing.
What happens if I close the app while the agent is working?
With a Session Goal defined, the agent keeps iterating turn after turn toward that goal even while the app is closed; you pick the session back up later from any paired device.
Does it run on Windows?
Yes, there’s a native desktop app for Windows, in addition to macOS and Linux; for the command-line installation of the OpenCode SDK on Windows, the practical route is to use WSL.
References
- OpenChamber: the project’s official site, with the full list of features and the privacy policy.
- OpenCode: official installer for the agent SDK OpenChamber runs on.
- Wikipedia: Integrated development environment: general background on what a development environment is and how it evolved to incorporate agents.
📱 Like this content? Follow @programacion on Telegram for daily tech content in Spanish: quick summaries, fresh content every day.
Imagen destacada: Foto de Daniil Komov en Unsplash
0 Comments