Skip to content

Jan

Jan runs models on your own machine, and remote ones beside them. Add the gateway as a custom provider and every remote chat is checked against your policies, charged to your team and recorded in Request logs, with GPT, Claude and Gemini models in one list.

What it is

Jan is a desktop chat app. Each person runs their own copy; it calls remote providers directly from that machine. Its custom providers speak the OpenAI or the Anthropic format.

The two settings

Open Settings › Model Providers, choose Add Provider, and in Add Custom Provider:

Field Value
API format OpenAI-compatible (the default)
Provider name For example "Vidai"
Base URL https://<your-gateway>/v1
API key Your Vidai key (starts vk_), never a provider's key

Choose Create. With Anthropic-compatible instead, the Base URL is https://<your-gateway> (no /v1); the gateway translates Messages calls to GPT and Gemini models too.

The gateway's address is under Connect your app on the Keys page and in System › Settings › Gateway. You get your key from Keys › Issue a key; its secret is shown once.

A minimal call

  1. After Create, Jan lists the models your key may call under the provider's Models.
  2. Start a new chat and pick a model under your provider, for example gpt-4o-mini.
  3. Send "Reply with exactly: ok". A short answer means the address and key are right.

Streaming

Jan streams answers. A stream from another provider's model arrives in the format you chose; Vidai records the tokens and the cost either way.

Tool calls

Jan can't tell what a custom provider's models can do. To let a model use tools (MCP servers included), open the provider's Models, edit the model and turn on Tools under Capabilities. Tool calls then work on every provider the gateway translates to.

Reading what the gateway did

Jan doesn't show response headers. Look the call up in Request logs: filter on Key, Model or Path (/v1/chat/completions or /v1/messages). Cost is never in the answer.

Errors

Jan shows the gateway's error message in the chat. Common ones: 401 invalid_api_key, 403 model_not_allowed, 403 content_guardrail_blocked (a guardrail stopped the message), 404 on every call (the Base URL is missing /v1), 429 (a limit on your key or team). The full list is in Headers and errors.

Model names

Jan fetches GET /v1/models when you save, which the gateway answers with exactly the models your key may call, named as your organisation approved them under Models in the Control room. Add a name by hand with + if one is missing.

What goes through the gateway

Jan feature Through the gateway
Chats, streaming, tools, images (once Vision is on for the model), with a gateway model selected Yes, to OpenAI, Azure OpenAI, Anthropic, Gemini and Vertex AI models
Local models and Jan's own local API server No: they stay on your machine
Other providers in Jan, with their own keys No

Check it works in a minute

Send "Reply with exactly: ok" to a gateway model, then open Request logs a few seconds later and filter on your key.

Good to know

  • Jan is one person's app, so each person uses their own key, issued in Keys › Issue a key; their calls show under them and are charged to their team.
  • Use one key per provider. Jan's fallback keys (API keys › Advanced) move to the next key on 401, 403 or 429, so a refusal such as model_not_allowed would be retried on another key, and the calls would show under several keys.
  • Jan's own docs: Custom endpoints.