Model or dataset
Stack-Cairn/LiveAgent avatar
Stack-Cairn/LiveAgent

LiveAgent is three languages and three sponsors who sell the routes it calls

A fully functional AI Agent desktop client that supports Webui access and can be creatively customized and expanded!

2,262 stars249 forksTypeScriptMIT

At a glance

What is it?
A local-first desktop agent with an optional remote gateway, written across three languages and shipped through continuous integration as signed installers on a rolling beta channel. The most interesting parts are the sponsor disclosures, which name the relay services that provided the development resources and describe the same kind of provider the app is built to route to.
Who is it for?
Use this if you want an agent that can actually edit files and run commands on your machine rather than one confined to a chat box, and if you are willing to read what it is about to do before it does it. Two things to weigh before you install it.
Can I use it commercially?
Yes. MIT is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
Is it still maintained?
Yes. The repository received new commits within the last day.
What is it written in?
Mainly TypeScript, according to GitHub's language statistics.

Answers come from the project's GitHub data, last synced on November 15, 2026, and from our analysis. They are not legal advice.

From daily snapshots

Star history

  • Stack-Cairn/LiveAgent2,262 ★ · +241
Star history
Repository2026-09-112026-11-15
Stack-Cairn/LiveAgent2k2.3k

241 Stars in the last 65 days, measured from the daily GitHub snapshot. Compare with other projects

Editorial analysis

Three languages, and the Rust side is a single crate

The repository is genuinely polyglot, and the split follows the deployment story rather than convenience.

The interface surfaces are TypeScript: one package for the shared interface components, one for the desktop application, one for the gateway's web interface. The desktop backend is Rust, inside a desktop-shell framework's own crate directory. The remote gateway is Go, and it is the only piece built as a standalone server binary.

So the Rust manifest lists exactly one member. That is unusual for a workspace file, and it says the Rust code was not decomposed into a reusable library and a shell around it. There is one crate, the shell, which is the right call if nothing else is meant to consume that code.

The Go choice is the more interesting one. A gateway whose job is streaming state to a browser over a socket protocol is a small amount of concurrency with a hard latency budget, and Go is a defensible way to get that without a runtime dependency tree.

The three-language split is consistent with how the project documents itself, which treats the desktop app and the gateway as two separate things you may or may not deploy.

The sponsors sell exactly the routes the app calls

The sponsor section is unusually specific, and it is the part of the page worth reading carefully rather than skipping past.

Three relay service providers are listed, each with a registration link and a paragraph describing the service: relaying for coding assistants and for models including domestic Chinese ones.

Two of them go further and say how they were involved. One states that the vast majority of the model resources used to develop the software were provided by them. Another states that the remaining model resources came from them. The third is thanked for support and offers pay-as-you-go billing.

So the funding was credits rather than cash, which is a common and unremarkable arrangement for a tool built on paid APIs.

What makes it worth naming is the coincidence with the feature list. The application advertises multi-model routing across three provider protocols with custom base URL support for third-party compatible services. The sponsors are providers of exactly that service. The page states both plainly, with no attempt to obscure either, and a reader is left to draw their own conclusion about how much of the routing design was shaped by the sponsors' capabilities.

A vendored dependency is linted and typechecked as if it were theirs

Several build scripts target a package under a different organisation's scope, and the container build explains why.

The comment is about a virtualisation library: it is vendored into the tree, it ships TypeScript source with its export map pointing at source and type declaration directories rather than built output, and so the build stage needs the whole package rather than just an installable artefact.

That is the explanation for being in the workspace at all. A published package with build output would be installed from a registry and never appear in your scripts. A package that ships source has to be built from source, and once you are building it, you own its failures.

Which is what the scripts show. There are separate commands to typecheck it and to run its tests, alongside the equivalents for the project's own packages. A third-party library is being maintained inside the workspace with the same rigour as first-party code, which is a significant ongoing commitment for something most projects would simply consume.

It also means an update to that dependency is a source change in your tree rather than a version bump in a manifest.

Three tiers of checking, and the release notes are generated by a model

The check scripts are worth reading because they reveal what this project considers worth automating.

One script is invoked at three strictness levels, fast, all, and strict. That is a project that has decided different changes deserve different amounts of checking, which is the right instinct and harder to implement than a single gate.

Alongside it a test runner covers six named files. Two are about the interface boundary between the three TypeScript surfaces, which is the kind of test you write when the packages drift. One checks that formatting only runs on files that changed. One checks diff hygiene.

And two are about releasing: one tests the version handling, and one tests the script that generates release notes with a language model. That last one is unusual and worth flagging on its own. The changelog is written by a model, and the script that does it has unit tests, so the failure mode is a badly formatted entry rather than a silent no-op.

It also means the release notes are generated by running a model against commit data, which is a decision with cost implications that the repository does not discuss anywhere visible.

The gateway container runs as a system account with no home directory

The container build is three stages and hardens the runtime more carefully than most.

The first stage builds the web interface, the second builds the gateway, and the third is the runtime. All three declare the build platform explicitly rather than the target, which is the cross-compilation pattern: the builder runs natively and produces artefacts for another platform.

The Go build disables the C library, so the result is static, and strips the binary. There is a comment above the platform arguments explaining that they must be left without defaults, because a default silently shadows the values the build system injects. That is a specific, learned kind of note.

The runtime stage installs one package, a certificate authority bundle. Then it creates a system account with a fixed numeric identifier, no home directory, and a login shell that refuses login, and prepares a state directory owned by that account with restrictive permissions.

For a service that executes commands on your behalf, that is the right baseline. A gateway that can run shell commands should not be running as a user with a home directory full of your keys and shell history.

The macOS build picks its architecture from the build machine

The build file is mostly variables, and the interesting ones are the release ones.

There is a conditional on the host architecture that selects between an Intel target and an Apple Silicon target, with a variable for each so either can be overridden. Windows and Linux get a single target each, and Linux carries a list of bundle formats.

Then the release surface: a notarisation identity profile, a path to the desktop configuration for the platform, a path to a separate release configuration, a script that builds the disk image settings, the background image used for the disk image volume, the volume icon path, and a script that verifies the finished disk image.

The version is handled by conditional flag composition: the release configuration is always passed, and a version-specific configuration is appended only if that variable is set. So the same target produces a developer build and a release build with one toggle.

Windows is the opposite shape: one target, three output formats. A setup wizard for most users, a package for silent enterprise installation, and a portable archive for no install at all.

And the development defaults are visible in the same file, including a gateway token and a listen address. Those are development values, which is fine, and they are exactly the values someone would ship if they read only the development target.

Editorial conclusion

Use this if you want an agent that can actually edit files and run commands on your machine rather than one confined to a chat box, and if you are willing to read what it is about to do before it does it. Two things to weigh before you install it. It can expose local services to the public internet with one click and it runs shell commands unattended, so the permission surface is much wider than a chat assistant's. And the project is funded by API relay providers whose business is reselling access to the same models, which is disclosed plainly on the page and is worth weighing rather than skipping.

Frequently asked questions

What is LiveAgent?

It is a local-first AI agent desktop client that combines model reasoning with local system tools, so the agent can read and write files, run shell commands and manage scheduled tasks. The desktop app needs no server at all; an optional gateway is what lets you control the same local agent from a browser.

Is LiveAgent legit?

It is a real MIT licensed project with signed installers produced by continuous integration, publishing on a rolling beta channel with incrementing counters. Its development was funded by credits from model API relay services rather than cash, which the sponsor section states plainly, and those are the same kind of provider its custom base URL support targets.

Does LiveAgent send my data to a server?

The stated design is local-first. Local tool execution, persistent memory in the browser's own storage and scheduled tasks all run on your machine, and the gateway is optional and documented as only necessary for browser access rather than for the agent itself to work.

What are LiveAgent's system requirements?

macOS on either Intel or Apple Silicon; Windows on x64, which requires the web view runtime that ships with Windows 11; and Linux on x86_64, which requires a webkit toolkit at version 4.1, which the page ties to Ubuntu 22.04 or Debian 12 or newer.

How is LiveAgent packaged for Linux and Windows?

Linux ships as an AppImage that must be made executable before running, a package for the Debian and Ubuntu family installed with the package tool, or an RPM for the Fedora and openSUSE families. Windows ships as a setup wizard for most users, a package for silent enterprise installation, or a portable archive that runs without installing anything.

Official sources

  1. Issues
  2. License: MIT
  3. README
  4. Releases
  5. Stack-Cairn/LiveAgent on GitHub
Add this badge to your README

If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.

Add this badge to your README

markdown
[![Hysen Labs](https://hysenlabs.com/badge/stack-cairn-liveagent.svg)](https://hysenlabs.com/projects/stack-cairn-liveagent)

Star history chart for your README

markdown
[![Star history](https://hysenlabs.com/chart/stack-cairn-liveagent.svg)](https://hysenlabs.com/projects/stack-cairn-liveagent)

Was this useful?