Your AI, every tool you own
— one voice away.
Connect Claude, Gemini or a local model. Conduit pipes your AI through every MCP, every skill, your whole memory — one voice you can talk to from any device. Solo, or as a single shared brain for your whole company.
Free: 20 requests included · Solo 14 €/mo · Max 29 €/mo · Teams: 4-week paid setup
Connected
Every tool you already use — wired into one brain.
Pick your AI: Claude Code, Gemini via Antigravity, or a local model through Ollama. Conduit speaks the MCP protocol, so anything that exposes one becomes a verb your AI can do, by voice, from any device.
Conduit for Teams
Your AI. In your data room.
You’ll get your own AI instance in your own data room — hosted in Germany and built for GDPR (DPA/AVV available on request). Wired into your stack (DATEV, HubSpot, M365, …) with a shared company memory. Every employee accesses the same brain through Conduit — what a colleague feeds in today (processes, notes, customer info), the whole team uses tomorrow. The knowledge stays with you, and grows with every use.
Not another Cowork. Tools like Claude Cowork give one person a Claude-only agent in a US cloud, working on files on their own desktop. Conduit gives your whole team a shared brain — in your data room, hosted in Germany, wired into your stack, on the AI provider of your choice. The memory is the company’s, not one laptop’s — and it stays when people leave.
Your AI instance will run in a Frankfurt or Berlin data center — or, on Enterprise, inside your own VPC or on-prem. With a local model, no US-cloud and no data crossing borders.
No more “ask the person who knows”. The knowledge sits in Conduit. A new hire on Monday is as productive as a two-year veteran by Friday.
We build the MCP layer to your DATEV, HubSpot, Lexware, M365. Every employee gets the same access, the same way, by voice or by chat.
How a pilot works.
- 1We map your stack and define the first 3 MCPs to wire (typical: DATEV, HubSpot, M365).
- 24-week paid setup: AI instance in your data room, MCPs, skills, memory, per-team scopes.
- 3Monthly retainer for updates + new MCPs. You bring the Anthropic or Google API key. We never resell tokens.
Tell us about your team.
How it works
Three pieces. No reseller in the middle.
Bridge on your Mac
A tiny daemon spawns your chosen AI's local CLI per request: claude or gemini. A local Ollama model is reached over HTTP instead. SQLite log, JWT auth, warm-pool process. Runs on launchd, restarts itself.
Cloudflare Tunnel
Outbound-only WSS — no open ports, no inbound firewall holes. Auto-provisioned per account; TLS is terminated at Cloudflare's edge and message contents are never stored on our servers.
Conduit clients
The web app at app.tryconduit.de, installable to the home screen on iPhone and Android and to the dock on Mac and Windows. Live re-attach over WS so a 4G drop never breaks the answer mid-stream.
Why Conduit
What no other tool gives you.
| Conduit | Claude Cowork | Claude.ai / ChatGPT | Cursor | |
|---|---|---|---|---|
| Voice in | ✓ | — | ✓ | — |
| Mobile (iOS · Android) | ✓ | — | ✓ | — |
| Access your repos / shell / files | ✓ | Your desktop | — | Editor only |
| Your own MCP servers | ✓ | Connectors only | — | — |
| Your own skills + memory | ✓ | Per-user | — | — |
| Same brain for the whole team | Teams, in development | — | — | — |
| Provider-agnostic (Claude / Gemini / local) | ✓ | Claude only | Locked-in | Locked-in |
| Runs in your data room (DE / on-prem) | Teams, in development | — | — | — |
| AI processing stays on your machine (local model) | ✓ | US cloud | — | — |
Pricing
Pay for the bridge. Tokens stay yours.
Bring your own Claude or Gemini account, or run a local model through Ollama. We never resell tokens. We charge for the tunnel, the apps, and the convenience of having your AI everywhere.
- Full UI & history access
- Auto-tunnel setup
- 20 real requests to try it
- Unlimited messages
- Hosted voice in & out · ~2 h speech/day
- Claude, Gemini or a local model
- Everything in Solo
- 4× voice budget · ~8 h speech/day
- Priority support, answered by us
- AI instance in your data room
- Custom MCPs (DATEV, HubSpot, M365)
- Per-team scopes & audit log (planned)
- Requires API key (Anthropic or Google)
- Enterprise On-Prem from €4,990
FAQ.
Which AI does Conduit use?+
Whichever you bring. The bridge runs Anthropic's claude CLI, Google's gemini CLI (Antigravity), or a local model through Ollama. You pick it in Settings. Claude and Gemini work with tools, so they read files, run commands and reach your MCPs. The local Ollama route is a chat only: it has no tool access, so no file work and no commands. We never resell tokens; you connect your own account or API key. Switch providers anytime without losing your chats or memory.
What does the Free tier actually do?+
Free includes 20 real requests so you can put Conduit to work before you decide. They are one-off, not a monthly allowance, and there is no expiry and no credit card. After that the account stays free and read-only: the full UI, your chat history and memory, your bridge status, and live output while a task runs. To keep sending, you move to Solo or Max.
How is the bridge installed?+
When you install the Conduit app on your Mac for the first time, it auto-provisions a private Cloudflare tunnel for your account and starts the bridge as a launchd service. No DNS or Cloudflare setup on your side — just install and sign in.
Does my data leave my machine?+
Chat history lives in local SQLite on your bridge (your Mac) and we never store it. To generate answers, your prompts do go to whichever AI provider you connect, Anthropic or Google. Pick a local model through Ollama and nothing leaves your machine at all: the model runs on your own hardware, and the trade-off is that it replies in text only, with no file access and no commands. Between your phone and your Mac we only see encrypted WebSocket traffic, TLS-terminated at Cloudflare. Teams will be hosted in Germany, or in your own VPC on Enterprise.
How can my whole team share one brain?+
Conduit for Teams is in development — we're onboarding design partners now. It will run one shared AI instance in your data room — hosted in Germany and built for GDPR (DPA/AVV available on request). Shared memory, skills, and MCPs (DATEV, HubSpot, M365, …). Each employee accesses the same brain; what one person feeds in today, the whole team uses tomorrow. Per-team scopes, audit log, and on-prem option are planned.
How is this different from Claude Cowork?+
Cowork gives one person a Claude-only agent that works on files on their own desktop, in Anthropic's US cloud. Conduit is built to give a whole team one shared brain — running in your own data room (Germany or on-prem), wired into your stack via your own MCPs, on whichever AI provider you bring. With Cowork the memory lives on a laptop and walks out the door when someone leaves; with Conduit the knowledge is the company's and compounds. They're complementary for a solo desktop task — but for a team that wants its own AI in its own data room, that's Conduit.
What if Anthropic or Google build their own mobile apps?+
They probably will, for their own model. Conduit is provider-agnostic by design. You can swap Claude for Gemini or for a local model in Settings without losing memory, MCPs or chats. We're the layer that stays.