Web search, page reading and research in Chat
Models search the web, read pages, run code and write cited research reports, using the provider's own tools or Janus's search for self-hosted models.
What it is
In Chat, a model is offered the tools it can really use. Claude, Gemini 3 and newer, OpenAI and xAI models search the web and run code on their provider's servers; xAI models can also search X. Any chat model can read a web page you link to. Self-hosted models such as vLLM, llama.cpp or Ollama can search through a service you set up: SearXNG, Brave Search, Exa or Tavily. Deep research has the model plan, search for several rounds and write a cited report, and Compare sends one message to up to three models side by side.
Why you want it
A model that can't look anything up answers from old training data. People expect current answers with sources, including from the models you host yourself, and the searches should be metered and governed like everything else.
Choose SearXNG, Brave Search, Exa or Tavily for models whose provider has no search of its own.
Actual Janus interface · Demo identities and synthetic usageHow it works
- Provider-run tools are on by default; administrators can turn each one off for everyone, and people can turn them off per chat
- Janus never runs code itself; code runs in the provider's own sandbox
- Read web pages with any model: Claude and Gemini use their provider's reader, other models use Janus's, which opens only public internet addresses, fetches up to 5 MB and reads the first 30,000 characters
- Web search for self-hosted models through SearXNG, Brave Search, Exa or Tavily, with a Test search button; keys are stored encrypted and never shown again; the Helm chart can install a private SearXNG
- Sources are listed under the reply, and search results and pages are marked to the model as information, not instructions
- Deep research: up to 20 search rounds and a cited report saved on the reply as a Markdown file
- Compare: up to two more models answer the same message with their own time and cost; continue from whichever answer you prefer
- Through the API, /v1/chat/completions accepts the generic web_search, x_search, web_fetch and code_execution tools for models whose provider runs them
POST /v1/chat/completions
{ "model": "claude-sonnet-5",
"tools": [{ "type": "web_search" }, { "type": "web_fetch" }],
"messages": [ … ] }
# activity comes back as janus_activity, citations as annotationsIn the product
Screenshot above. Actual Janus interface with demo identities and synthetic usage.
See it on your own network.
The Community edition is free for up to 25 people. The 30-day Business trial unlocks every Business feature.