[ NEURALARC_PRODUCT_—_TELLNOVA ]

Background Coding Agent Runtime

2026Developer PlatformsIn-depth

What it is

Runs coding agents in isolated Git worktrees so delegated work never touches your active workspace.

Visit the live product
Screenshot of Background Coding Agent Runtime
Background Coding Agent Runtime — live product

[ THE_PROBLEM ]

Why this existed

Coding agents are useful and disruptive at the same time: useful because they can do the mechanical work, disruptive because they want the same working directory you are using. Running more than one is chaos.

[ WHAT_WE_BUILT ]

What we built

A desktop application and a self-hosted web mode sharing one runner host. Sessions dispatch to pluggable runners against a canonical clone with a per-attempt worktree, so several agents can work on the same repository simultaneously without colliding. Local JSON files are the source of truth for a running session; authenticated clients mirror bounded snapshots to the cloud for cross-machine history. GitHub authentication, an embedded terminal, a repo picker with per-repo environment config, model selection across providers, image attachments, an outbound tunnel for remote access, and auto-update. The self-hosted server binds to loopback and generates a bearer token treated as an administrator credential. Hermetic integration tests cover the host, Git operations, the server, the shipping flow and the tunnel; desktop builds ship for macOS (Intel and Apple Silicon), Windows and Linux (x64 and ARM64) with published checksums.

  • Desktop application and self-hosted web mode sharing one runner host
  • Sessions dispatched to pluggable runners against a canonical clone
  • Per-attempt Git worktrees so parallel agents never collide
  • Local JSON files as the source of truth for running sessions
  • Bounded session snapshots mirrored to the cloud for cross-machine history
  • GitHub authentication, embedded terminal and repo picker with per-repo environment config
  • Model selection across providers and image attachments
  • Outbound tunnel for remote access
  • Self-hosted server bound to loopback with a bearer token treated as an admin credential
  • Hermetic integration tests over host, Git, server, shipping and tunnel
  • Signed desktop builds for macOS, Windows and Linux with published checksums

[ HOW_IT_IS_USED ]

How a company uses it

An engineering team parallelises the work that does not need a human in the loop — dependency bumps, test backfill, mechanical refactors — while developers keep working. Worktree isolation is what makes running five agents at once safe rather than chaotic.

Built with

ElectronTanStack StartConvexBetter Authnode-ptyelectron-builder

[ COMMON_QUESTIONS ]

Questions clients ask

Why worktrees instead of separate clones?

Worktrees share one object store, so spinning up an isolated working directory is nearly instant and costs little disk, even on a large repository. Separate clones would mean a full copy per attempt. That efficiency is what makes running several agents at once practical rather than theoretical.

Do my repositories or credentials leave my machine?

Code stays local — the runner operates on a local clone. Only bounded session snapshots sync to the cloud, and only when you are signed in, to give cross-machine history. The self-hosted server binds to loopback by default and the bearer token it generates is an administrator credential, which the documentation is explicit about.

What kind of work is actually suitable for a background agent?

Work with a clear definition of done and a cheap way to verify it — dependency upgrades, test backfill, mechanical refactors, migrating a pattern across files. Work requiring product judgement is not a good fit, and pretending otherwise is how teams lose trust in the tool.

Is 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