[ NEURALARC_INTERNAL ]
Self-Hosted Server & Agent Control Plane
What it is
Control your own machine — and trigger coding agents against your repos — from your phone.

[ THE_PROBLEM ]
Why this existed
The gap between noticing a problem and being at a machine where you can fix it wastes hours a week — and there is no safe way to trigger a build from a phone.
[ WHAT_WE_BUILT ]
What we built
System stats, process management, a browser-based terminal over websockets, a file browser, network info and cron management, plus an agent orchestration cockpit. Register a project by local path or by searching your GitHub repos and cloning one; pick the agent and model; optionally define validation steps that must pass. Queue a task and a background worker checks out the default branch, fast-forwards from origin, cuts a branch, runs the agent in the repo, runs the validation steps, then commits, pushes and opens a pull request — with live agent output streaming to the task view. One-command setup from a fresh clone: virtualenv, dependencies, frontend build, config and a token.
- System stats, process management and cron management
- Browser-based terminal over websockets
- File browser and network information
- Project registration by local path or by searching and cloning a GitHub repository
- Per-project agent and model selection
- Optional validation steps that must pass before completion
- Background worker: branch from an up-to-date default, run the agent, validate, commit, push, open a pull request
- Live agent output streamed to the task view
- One-command setup from a fresh clone
[ HOW_IT_IS_USED ]
How a company uses it
A technical founder or small team turns idle time into shipped work — read a bug report on the train, queue the fix, review the pull request when you land. The validation gate means nothing merges that does not build.
Built with
[ COMMON_QUESTIONS ]
Questions clients ask
Is it safe to expose a control panel like this?
Only with care. It ships bound with token authentication and an outbound tunnel option rather than an open port, and the token is an administrator credential. Anyone deploying this needs to treat it as remote code execution on their machine, because that is exactly what it is.
What do the validation steps do?
They are the gate between "the agent produced something" and "a pull request exists". Lint, test and build commands defined per project must pass, so a change that does not compile never becomes a PR waiting for your attention.
Why open a pull request instead of committing to main?
Because review is the point. The agent doing the work does not remove the need to look at it — it removes the need to be at a keyboard while it happens.
[ RELATED_WORK ]
Similar builds
NeuralArc product — Tellnova
Background Coding Agent Runtime
Runs coding agents in isolated Git worktrees so delegated work never touches your active workspace.
Read case studyNeuralArc product — Tellnova Studio
Chat-to-App Builder with Live Preview
Describe an app in chat, watch it build and run in the browser.
Read case studyNeuralArc product — NeuralCode
CLI Coding Assistant
A terminal-first coding assistant with persistent context and memory.
Read case studyIs this close to your problem?
Most engagements start with a version of something on this page. Tell us what is different about yours and we will tell you what it changes.
Start a conversation