Infercom in Claude and ChatGPT. EU-sovereign, one click.
Infercom Router is a connector for Claude Desktop, Cowork, claude.ai and ChatGPT. Your assistant delegates the heavy lifting to open models on Infercom in German data centres: cheaper, with capabilities like transcription and 192k context, and with a record of which model ran where. The router picks the model; you see tokens and costs in real time.
Register for free How it works
The right model, automatically
Coding, analysis, summaries, images, audio: rules plus classification choose between MiniMax, gpt-oss, Gemma and Whisper. EU models are the default.
Transparent down to the token
Every answer names the model, the reasoning, tokens and cost. The dashboard shows usage and credit.
Heavy lifting moves to Europe
Long documents, transcriptions and bulk processing run on Infercom in Germany with zero data retention. Your assistant only receives the result. Whatever you type directly into Claude or ChatGPT is still seen by their provider.
Set up in two minutes
- Register and sign in. Infercom employees use the service free of charge.
- Top up credit (from €25) or store your own Infercom key.
- Add the connector in your assistant; per-assistant guide below.
- Go: “Use Infercom and summarise this document.”
How the router works
Your assistant stays the cockpit and sees your input. The router handles model choice, execution on Infercom and billing, and returns the result with its reasoning. Content is stored neither by us nor by Infercom.
Setup per assistant
Claude Desktop, Cowork, claude.ai, mobile (Pro, Max, Team, Enterprise; Free with one connector)
- Customize → Connectors → Add custom connector.
- Enter the
https://mcp.pioneerdesk.pro/infercom/mcp:https://mcp.pioneerdesk.pro/infercom/mcp. No other fields. - Claude opens our sign-in. Register or sign in, done.
- In chat: “Use Infercom and summarise this document.” Claude calls the tool ask.
Team/Enterprise: an owner adds the connector in organisation settings, members enable it themselves. Claude Code: claude mcp add --transport http infercom https://mcp.pioneerdesk.pro/infercom/mcp
ChatGPT (Plus, Pro, Business, Enterprise, Edu; web app)
- Settings → Apps → Advanced settings → enable Developer mode. In Business/Enterprise an admin has to allow it.
- Apps → Add connector →
https://mcp.pioneerdesk.pro/infercom/mcphttps://mcp.pioneerdesk.pro/infercom/mcp, authentication OAuth. - Complete the sign-in and enable the connector in chat.
- Use it in chat or Deep Research. For Deep Research, search and fetch are available.
Developers: via the Responses API as a tool of type mcp with the same URL.
Gemini
The Gemini app has no custom connectors yet. The router works there through the Gemini CLI, which supports MCP servers:
- Add to ~/.gemini/settings.json under mcpServers:
{"infercom": {"httpUrl": "https://mcp.pioneerdesk.pro/infercom/mcp"}} - Start Gemini CLI; /mcp lists the server and sign-in opens in the browser.
- Phrase tasks as you would in Claude.
Once Google enables connectors in the Gemini app, the same endpoint works unchanged.
Chat in the portal: sovereign end to end
If you want to work without any US provider, chat directly in the portal. The controlling model runs on Infercom in Germany and the history is stored encrypted with us (30 days, deletable at any time, “don’t save” per conversation).
Understands your files
Upload PDF, Word, Excel, PowerPoint, Markdown, code, images, audio and video. Scanned PDFs are read, images described, audio transcribed, all on Infercom.
Knowledge base per conversation
Large files are split into sections and embedded. For every question the matching passages go to the model, with file name and section as the source.
Read the web and research live
Links in your message are fetched by our own server. With “Research” we search live (own meta search, optionally Google, only the query leaves) and hand the best pages to the model with source and date. Fetched pages are kept versioned in a shared cache.
Verifact on demand
Verified German law with official references, trust score and date as the evidence base. Per conversation, 0.10 € per evidenced answer, free for Infercom staff.
Quality levels
Standard: one model, chosen by the router. Verified: gpt-oss-120b checks every statement against the evidence and flags what is unsupported. Ensemble: two models answer, a third checks and merges. Cost per step visible, all on EU models.
Sovereign and transparent
History encrypted with us, 30 days, deletable any time or not stored at all. Every answer names models, tokens and cost. No US provider sees your question, files or answer.
Pricing
Credit customers pay the prices below per million tokens, net. For example, €25 of credit buys about 46.175.824 output tokens on gemma-4-31B-it. With your own Infercom key, Infercom bills you directly and the connector is free.
| Model | Region | Capabilities | Context | Input / 1M | Output / 1M |
|---|---|---|---|---|---|
gemma-4-31B-it | EU | vision | 131.072 | 0.26 € | 0.46 € |
gpt-oss-120b | EU | reasoning, tools | 131.072 | 0.29 € | 0.77 € |
MiniMax-M2.7 | EU | tools | 196.608 | 0.78 € | 3.12 € |
Meta-Llama-3.3-70B-Instruct | Global | text | 131.072 | 0.78 € | 1.56 € |
DeepSeek-V3.1 | Global | reasoning, tools | 131.072 | 3.90 € | 5.85 € |
DeepSeek-V3.2 | Global | reasoning | 32.768 | 3.90 € | 5.85 € |
E5-Mistral-7B-Instruct | EU | embedding | 4k | 0.17 € | – |
Whisper-Large-v3 | EU | audio | – | 0.13 € / hour | – |
All 8 models of the Infercom catalogue (synced daily with api.infercom.ai). EU = data centres in Germany, the router's default. Global = infrastructure in the US/Japan, only on explicit request. Prices net of VAT; Whisper per audio hour, E5 input tokens only. Top-up from €25 gross, invoice by e-mail immediately.
Optional: a Verifact answer in chat costs 0.10 € per question, net, only while the switch is on and Verifact returns an evidenced answer. Free for Infercom staff.