Conduit Download
  1. Conduit
  2. White-label
Early access

Stop demoing your AI product inside someone else's app

Conduit is a white-label AI desktop client: an open-source, bring-your-own-key assistant you can put your own name, logo, colours, and model policy on. Rebranding it is a config file, not a fork.

Become a design partner See what you get
  • Runtime branding ships today
  • Your customer's data never touches you
  • Open source, AGPL-3.0

The stock app, rebranded in Settings as a fictional “Northwind AI” — name, tagline, and all eighteen colours, applied at runtime.

The problem

You built an MCP server, or an AI product with a real tool surface. To demo it, you ask a prospect to install Claude Desktop or ChatGPT, sign in, and add your server. When it works, your product shows up as a line item in a competitor's assistant — under their brand, on their model, inside their relationship with your customer.

One click if you packaged it as an extension, a hand-edited config file if you did not. How to demo an MCP server to a customer walks through what each route costs you.

What a white-label build gives you

Hand over one app. Yours.

It opens with your name in the title bar. The model is whichever one shows your product best. The first screen says what you want it to say.

Your brand, not theirs

Product name, tagline, logo, window title, and around eighteen colours per mode, light and dark. Your tools stop being a plugin and start being the product.

Never a data processor

Conversations stay in a database on the customer's machine and model calls go straight to the provider. Your infrastructure is never in the path — and neither is ours.

So a proof of concept doesn't start with a data processing agreement.

Any model policy — even air-gapped

Seventeen providers, hosted and local. A frontier model for the flagship demo, a cheap fast one for a trial, or Ollama and LM Studio for an on-premise customer who cannot call an external API at all.

How it works

A config problem, not a fork

Design tokens live in a shared package, and the same brand.md drives two modes. Improvements to the core reach your build without a merge conflict.

  1. Describe your brand

    One brand.md file: identity, logo, and a palette for light and dark. Hex colours only — a brand file can't express a network fetch.

  2. Apply it at runtime

    Import it in Settings → Branding on a stock install. Name, tagline, logo, window title, and colours change in minutes, and reversibly. Ships today.

  3. Or build your own installer

    Build-time branding reaches what a running app can't: app icon, installer name, bundle identifier, bundled fonts, and your own update channel. Early access, from source.

brand.mdone file, two modes
[identity]
appName     = "Northwind"
displayName = "Northwind AI"
tagline     = "Message Northwind..."

[logo]
file = "logo.png"

[palette.dark]
bg       = "#0F1115"
card     = "#161A21"
ink      = "#E8EAED"
hue      = "#E4572E"
hueSolid = "#B8441F"
# …eighteen colours, for light and dark

Above: the same palette entered in Settings → Branding, previewed live across the whole window before it's saved — and one click to undo.

Available today

What you get today

Everything below ships in the current release, and all of it carries your brand once you rebrand an install in Settings → Branding.

  • Provider-agnostic core — 17 providers, local and hosted
  • MCP connector runtime — local stdio and remote streamable HTTP, with consent
  • Local-first storage, BYOK, no proxying
  • Runtime branding — Settings → Branding (name, logo, colours)
  • Cross-platform release pipeline — unsigned installers on GitHub
  • Artifacts, including sandboxed HTML, Mermaid, and KaTeX
  • Documents — chat with your own text, Markdown, CSV, Word, and PDF files
  • Image generation — OpenAI, Gemini, OpenRouter
  • One design in dark and light, with a main-colour setting
  • Interface in eight languages
  • MCP prompts and resources in the composer

Packaging your own installer is in early access. What isn't available yet is in the questions below.

Design partners

Help shape the packaging layer while it's cheap to change

The packaging layer is being built now. That's the useful moment to be involved. A design partner gets:

  • Direct influence over what the packaging layer supports
  • Your connectors as the reference case the tooling is built against
  • Early builds as each piece lands
  • A direct line to the person building it, not a support queue

Good fits

Companies with an MCP server or tool surface they already demo; teams selling AI into regulated or air-gapped environments; anyone whose sales motion starts with "first, install Claude Desktop."

What we ask in return is the honest version of your use case: what you'd ship, what would block you, and what would make it not worth doing.

Start a thread

On GitHub Discussions: say what you're building and how you demo it today. If it's a fit we'll take it to a call; if it isn't, we'll say so.

White-label questions

What is a white-label AI client?

A chat application someone else builds and maintains, which you ship as your own product. You supply the name, the logo, the colours, the model policy, and the tools it opens with; the vendor supplies the client itself — provider adapters, connector runtime, artifact sandbox, release pipeline. Most white-label AI offerings are hosted chat widgets for websites. Conduit is a native desktop app that is provider-agnostic and local-first, so a white-label build is genuinely yours rather than someone else's account with your logo on it.

Can I ship an AI chat client under my own brand?

Partly, today. On a stock Conduit install you can change the product name, the tagline, the logo inside the app, the window title, and around eighteen accent and surface colours from Settings → Branding — that is runtime branding, and it ships now. Producing your own installer, with your own app icon, installer name, and bundle identifier, is build-time branding: it exists, it is built from source, and it is early access for design partners. Shipping a build that arrives already connected to your MCP servers is in development. You cannot download a pre-built branded installer from GitHub today.

What isn't available yet?

Packaging your own installer from brand.md, with your app icon, installer name, update channel and bundled fonts, is built from source and in early access for design partners. Shipping a build that arrives already connected to your MCP servers is in development, as are OS code-signed and notarized builds; provisioning credentials to your users' installs is designed but not built. There is no pre-built branded installer to download. If you need a production OEM build soon, start a thread on GitHub Discussions before you plan around it.

Does the AGPL let me white-label Conduit?

Yes, with an obligation attached. Conduit is AGPL-3.0-only. A rebranded build is a modified copy, so distributing it — an installer, a download link, anything that leaves your own machines — carries the duty to make the corresponding source available to whoever received the binary. That duty is not contingent on how much you changed: a name, a palette, an icon, and an update endpoint is a modification like any other. Section 13 applies the same duty if you run a modified version as a network service rather than shipping it. Trademark is a separate body of law from the licence, and the AGPL grants nothing about use of the name Conduit itself. This is a description of the licence, not legal advice.

Do I need my own update server for a branded build?

Yes — for a packaged build it is mandatory, not a nicety. A distributed build carries its own update endpoint and its own signing key, full stop. Inheriting ours would point your customers at an update manifest that was never meant for them, and bake in a public key that cannot verify releases you sign. It is the one setting in the branding format where falling back to the built-in value would be unsafe rather than merely wrong-looking, so a packaged build without it is meant to fail to build.

What cannot be rebranded?

Three things, deliberately. The on-disk data directory keeps Conduit as its organisation qualifier regardless of the product name, because changing it would silently orphan the data of anyone upgrading from a stock install. The updater user-agent string and the identity Conduit presents to MCP connector servers also stay as they are. Those are wire identity rather than display identity: everything a person sees can carry your brand, while the strings other systems key on do not move.

Is there a commercial licence instead of the AGPL?

Not a published one. Conduit has a single copyright holder and contributions arrive under a contributor licence agreement, which is what makes separate commercial terms possible in principle — the ability was preserved on purpose. But no commercial licence is on offer today, and nobody should plan a launch around terms that do not exist yet. If the AGPL does not suit your distribution model, that is a conversation to start early rather than a box to tick late.

Related

Other ways teams use Conduit

For developers

An MCP client that asks first

Local and remote servers, OAuth, registry install, and graded consent.

Conduit as your MCP client For sensitive work

An AI assistant that doesn't phone home

No telemetry, local storage, and fully offline with a local model.

Keep it private For air-gapped customers

Local models, with a real app around them

Ollama, LM Studio, or your own server — no key, no bill, no internet.

Run models locally

Claude Desktop details on this page — one-click .mcpb extensions, remote custom connectors, and the platforms it runs on — were checked against Anthropic's installation, local MCP server, and custom connector documentation, checked 7 September 2026. If something has changed since, their documentation is right and this page is wrong — tell us. Meanwhile the client itself is open source and free to use: read the documentation, see what stays on the machine, or browse the source.