LiteLLM adds shared decision endpoints and a Playground tab for Jev, Clef and more
On 10 October 2026, LiteLLM said users could call Jev, Clef, Microsoft-Decision-1 and other decision models through its proxy, then try them in the Playground’s Decisions tab.
The unified routes shipped in LiteLLM v1.104.2, released on 8 October. POST /v1/decisions accepts OpenAI’s format (input plus a list of questions); POST /v1/systemone accepts state plus named questions. Both use the configured provider list and the proxy’s normal keys, budgets, spend tracking and guardrails. The initial v1.104.2 list includes OpenAI, Perplexity, TypeSafe, OpenRouter, Cloudflare and self-hosted Strands Decider. Image input works only with OpenAI.
There are two version caveats. The tweet names Microsoft-Decision-1, but Microsoft Foundry is absent from v1.104.2’s initial list; LiteLLM’s current docs say Microsoft Foundry and Databricks support landed on main after v1.106.0-dev.3. The current docs also describe the Decisions Playground tab, while noting that it is not in v1.104.2 or v1.105.0. Check the build and provider matrix before choosing an endpoint.
This adds a shared proxy and testing surface for decision models. It does not introduce new model weights or change Jev’s catalog scores.