The archive · Product Ideas · Product decision · 2025
Omnara's command center runs Claude Code sessions from terminal, web or phone
A wrapper that keeps the native Claude Code experience but makes it reachable everywhere: start in the terminal, answer an agent from your phone.
Omnara
What it had to solve
Ishaan and Kartik kept hitting the same wall: Claude Code could work for long stretches, then stalled the moment it needed human input — and if they were away from the desk, everything paused. Existing wrappers changed or merely mirrored the terminal experience, so they built an 'agent command center' around the native session.
How it works
Omnara began as a two-person frustration with desk-bound coding agents. Ishaan and Kartik watched Claude Code work in long stretches, then stall the moment it needed input; if they were away from the laptop, progress simply paused. Their August 2025 Show HN described the answer as an 'agent command center': the same Claude Code session, reachable from terminal, web or mobile.
The design decision was to change nothing about the agent itself. Rather than wrapping Claude Code in a new interface, Omnara's CLI parses the session file and terminal output to capture user and agent messages, mirrors them to a web dashboard and mobile app in real time, and listens on both channels so the human can answer from whichever screen is at hand. Native output, permissions, notifications and mode switching all survive, and the backend is open source.
The founders also positioned the wrapper as a general human-in-the-loop layer for any agent, not just Claude Code — for example, a node in n8n workflows that pings a person when a reply is needed. The launch was free for up to ten agent sessions a month, then paid for unlimited sessions.
The Show HN of 12 August 2025 drew 310 points and 168 comments, and the repository evolved from command center to platform: agent state committed atomically to Postgres, automatic recovery from crashes and disconnects, machines from sandbox providers or the user's own laptop, bring-your-own models and tools, and Apache-2.0 cloud or self-hosted deployment.
Why it lands
- The wedge was reachability, not intelligence: agents already worked but stalled waiting for a human who might be anywhere, so making the session portable was the unlock.
- Keeping the native Claude Code experience — output, permissions, mode switching — meant users got a superset of the terminal instead of a compromise.
- Parsing the session file and terminal output instead of fighting the SDK let the wrapper mirror state faithfully and accept input from either side of the connection.
- Positioning as a general human-in-the-loop layer for any agent widened the idea beyond one tool and made the open-source backend the long-term product.
- Open-sourcing the backend turned the launch into a trust signal and later into a durable-agent platform with Postgres state and bring-your-own machines and models.
What it did
The Show HN of 12 August 2025 drew 310 points and 168 comments; one commenter compared the thread's reception to the famous Dropbox HN thread, calling it a good sign for the project. The repository showed 2.8k stars and 217 forks when the material was collected, and its own tagline had grown from a Claude Code command center into 'the API for production-grade agents'.
What you can take
When the model works but the workflow stalls, fix the last mile: keeping the native experience while making the session reachable everywhere turned waiting for an agent into a solved problem.
Since then
After launch, the command-center idea stayed but the product broadened. The repository, collected in September 2026, describes a platform for managed agents: state commits atomically to Postgres and recovers from crashes and disconnects; execution uses sandboxes or your own machines; models come through OpenRouter, LiteLLM or Ollama with your own keys; and it runs as cloud or self-hosted under Apache 2.0, an open-source alternative to Claude Managed Agents. Developers in the launch thread compared the reception to the Dropbox thread. No funding or revenue figures appear in the sources here.
Sources
spotted an error? The archive wants to know.
Your turn
You just read one. Describe the brief you are staring at, and see who has been given the same problem.
Free account · 3 free questions · no card