mirror of
https://github.com/BerriAI/litellm.git
synced 2026-11-15 09:12:18 +00:00
563dfd5075
* fix(interactions): bind Deep Research agents to their deployment and bill the agent A request that sends an agent with a model group now sends the agent to Google and is refused with a 400 unless the group's deployment is that agent. Agent interactions are logged and priced under the agent, including streamed and background jobs, with a Vertex cost row for deep-research-preview-04-2026. Background cost polling now waits 7800 seconds by default so a 120-minute job still settles. * fix(interactions): refuse extra_body that changes the agent or model A nested agent or model in extra_body was merged into the request after the agent binding check, so it could run a different agent than the one billed, inject an agent onto a model deployment, or add a second selector. Refuse those before any provider or OAuth call, while identical agent repeats and unrelated extra_body fields still pass through. --------- Co-authored-by: Joshua Valluru <326636767+joshua-berri@users.noreply.github.com> |
||
|---|---|---|
| .. | ||
| __init__.py | ||
| test_endpoints.py | ||
| test_gemini_agents_endpoints.py | ||
| test_google_api_endpoints.py | ||
| test_google_endpoint_routing.py | ||
| test_google_gemini_proxy_request.py | ||
| test_interactions_agent_param.py | ||
| test_managed_agents_model_param.py | ||