scriptoriumadmin/litellm
1
0
Fork 0
mirror of https://github.com/BerriAI/litellm.git synced 2026-10-06 02:48:13 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 103
49673 commits 3198 branches 1436 tags 4.5 GiB
Commit graph

12082 commits

Author SHA1 Message Date
Joshua Valluru
70ef8b24b6 fix(mcp): reject bare schemes in raw authorization 2026-09-16 10:44:00 -07:00
Joshua Valluru
6c517bfc49 fix(mcp): reject scheme-only API key authorization payloads 2026-09-16 09:00:17 -07:00
Joshua Valluru
fdb8e3533b fix(mcp): validate credentials in existing request paths 2026-09-16 08:07:10 -07:00
Joshua Valluru
87190604f6 chore: integrate current main for MCP security compatibility 2026-09-16 07:54:13 -07:00
joshua-berri
9cd787386e
Merge pull request #41314 from BerriAI/litellm_fix_mcp_jwt_oauth_persistence 
fix(mcp): authorize JWT OAuth credential persistence
 2026-09-16 06:47:40 -07:00
Joshua Valluru
ee676d59f2 fix(mcp): preserve browser OAuth for unrelated bearer tokens 2026-09-15 23:08:14 -07:00
Joshua Valluru
e035682ed1 refactor(auth): separate JWT identity and OAuth authorization 2026-09-15 22:15:57 -07:00
kerry-berri
5960881640
Merge pull request #41337 from BerriAI/litellm_fix_responses_stream_absent_usage_recount 
fix(responses): recount tokens when a streamed response completes without usage
 2026-09-15 21:19:42 -07:00
kerry
ff878e7df0 refactor(responses): copy the terminal event instead of mutating stubbed chunks 
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
 2026-09-16 03:54:04 +00:00
kerry
7cc07d437a fix(responses): build the billed terminal response immutably and guard the cache dump 
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
 2026-09-16 03:47:03 +00:00
Joshua Valluru
258176de76 fix(mcp): validate rendered static credential payloads 2026-09-15 20:32:25 -07:00
kerry-berri
4b84fa9230
Merge pull request #41336 from BerriAI/litellm_fix_anthropic_stream_absent_usage 
fix(anthropic): tolerate message_delta events without usage when streaming
 2026-09-15 20:23:01 -07:00
kerry
7121e64db4 fix(responses): type the dict terminal response so the estimated usage is billed 
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
 2026-09-16 03:13:53 +00:00
Yassin Kortam
d108cdc431
Merge pull request #38254 from BerriAI/litellm_fix_xai_responses_instructions 2026-09-15 19:55:52 -07:00
Joshua Valluru
be506936bd fix(mcp): preserve explicit caller authorization credentials 2026-09-15 19:52:40 -07:00
yuneng-jiang
65d607d161
Merge pull request #41346 from BerriAI/litellm_e2e_provider_cache 
feat(e2e): reuse exact provider responses for 24 hours
 2026-09-15 19:45:56 -07:00
Yuneng Jiang
3dda798c8a
ci(e2e): consolidate cache contracts in filtered CircleCI job 2026-09-15 19:28:50 -07:00
yucheng-berri
c14ac1dbe9
Merge pull request #41329 from BerriAI/litellm_singulr_v2_contract 2026-09-15 19:27:30 -07:00
Joshua Valluru
84e14789d1 fix(mcp): preserve usable alternate header credentials 2026-09-15 19:18:48 -07:00
kerry-berri
399420bb25
Merge pull request #41339 from BerriAI/litellm_fix_fireworks_cost_components 
fix(fireworks-ai): bill cache-write, reasoning and audio tokens via the shared cost calculator
 2026-09-15 19:16:24 -07:00
Joshua Valluru
c7e4160ee6 fix(mcp): enforce OAuth write policy across signed callbacks 2026-09-15 19:15:34 -07:00
Yucheng He
146085669c fix(guardrails): run Singulr logging_only through the base hook, key MCP scans off the proxy call type and type the payloads 
Removes the Singulr async_logging_hook and logging_hook overrides so logging_only runs through CustomGuardrail.async_logging_hook: the response scope reaches Singulr as an assistant message instead of a raw ModelResponse dump, a vendor timeout is recorded as guardrail_failed_to_respond, a request-scope block ends the scan, and the sync success callback thread makes no Singulr call.

Decides MCP versus LLM by the proxy logging object's call_type (then the call_type or server-only markers in request_data), never by name, arguments or mcp_tool_name keys a client can put in a chat body. REST /mcp-rest/tools/call pre-scans reach Singulr as mcp_request and a non-mapping arguments value is forwarded as tool_arguments instead of raising.

should_block is a strict bool defaulting to false so a null verdict is an invalid response that block_on_error decides; payload fields drop Any for Sequence, Mapping and AssistantMessage types; metadata carries only the keys present; docstrings and section comments removed per the repo comment policy.
 2026-09-15 19:14:49 -07:00
aniket-kardile
8cce2b196a feat(guardrails): singulr v2 API contract with logging_only, pre_mcp_call and post_mcp_call 
Squash of BerriAI/litellm#37464 (head da298ca7) by @aniket-kardile, adopted onto main: v2 gateway payload contract with request, response, mcp_request and mcp_response scopes, typed payload models, proxy user, org and team metadata forwarded to Singulr, and the logging_only, pre_mcp_call and post_mcp_call modes.
 2026-09-15 19:14:49 -07:00
kerry
133f1e8ef5 test(fireworks-ai): drop the explanatory comment on the cache-read constant 
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
 2026-09-16 02:06:42 +00:00
kerry
7680d3de86 fix(responses): tolerate dict terminal responses when estimating usage 
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
 2026-09-16 02:05:44 +00:00
kerry-berri
54e1b9e3de
Merge pull request #41338 from BerriAI/litellm_fix_gemini_model_version 
fix(gemini): propagate the provider's modelVersion to the response model
 2026-09-15 19:04:50 -07:00
Joshua Valluru
97211bc356 fix(mcp): authorize per-user OAuth credential writes 2026-09-15 19:00:47 -07:00
yucheng-berri
41eb2dbfeb
Merge pull request #41128 from BerriAI/litellm_llm_judge_pre_call 
feat(guardrails): support pre_call and during_call modes for llm_as_a_judge
 2026-09-15 18:58:21 -07:00
tin-berri
106365fbac
Merge pull request #41341 from BerriAI/litellm_autorouter_pricing_fields 
fix(proxy): preserve Anthropic pricing modifiers in router savings
 2026-09-15 18:56:15 -07:00
kerry
7ed20406d7 test(responses): narrow the ValueError assertion to satisfy PT011 
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
 2026-09-16 01:47:45 +00:00
Yassin Kortam
67cb0bc089
Merge pull request #41313 from BerriAI/litellm_model_activity_response_time 
feat(ui): show average response time per model in usage model activity
 2026-09-15 18:43:48 -07:00
kerry
091a38cce6 test(anthropic): import Final for the annotated locals 
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
 2026-09-16 01:43:08 +00:00
kerry
c4c96180e3 fix(gemini): strip version suffix from modelVersion and keep it on blocked streams 
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
 2026-09-16 01:42:36 +00:00
kerry
8352045f32 style(fireworks-ai): apply repository conventions to the cost component change 
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
 2026-09-16 01:39:24 +00:00
Joshua Valluru
ece1de73b4 Merge remote-tracking branch 'origin/main' into litellm_fix_mcp_jwt_oauth_persistence 2026-09-15 18:36:02 -07:00
Devin AI
8d625d0400 fix(xai): keep 'instructions' on the xAI Responses API so system messages survive web_search bridging 
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
 2026-09-16 01:35:13 +00:00
kerry
10aef224e7 style(anthropic): apply repository conventions to the missing-usage change 
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
 2026-09-16 01:34:28 +00:00
kerry
d1aa37b0a8 merge main into litellm_fix_fireworks_cost_components 
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
 2026-09-16 01:28:29 +00:00
Joshua Valluru
8a3add3c6a fix(mcp): reject scheme-only Basic credentials 2026-09-15 18:27:25 -07:00
kerry-berri
171888b716
Merge pull request #41298 from BerriAI/litellm_drop_remaining_vendor_fact_pins 
test: drop remaining tests that pin cost-map vendor facts
 2026-09-15 18:25:14 -07:00
kerry
a6d2332f7a test(responses): drive the usage-estimate failure path without patching litellm 
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
 2026-09-16 01:24:35 +00:00
kerry
5a9fe56aff fix(responses): count custom-tool and MCP argument deltas in the streamed usage fallback 
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
 2026-09-16 01:24:05 +00:00
kerry
44a6d19889 test(gemini): drop review narration from the wrapper test docstring 
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
 2026-09-16 01:14:37 +00:00
kerry
1484fd7600 fix(responses): keep the usage estimate best-effort when token counting raises 
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
 2026-09-16 01:13:26 +00:00
Joshua Valluru
13553473aa fix(mcp): reject missing upstream authentication credentials 2026-09-15 18:06:00 -07:00
yucheng
f66ffc387e fix(guardrails): only honor the judge call-origin stamp on logging_only in llm_as_a_judge 
On pre_call, during_call and post_call the hook data is the client request body, so a
client-supplied litellm_params.metadata.internal_call_origin must not skip enforcement.
Type the during_call helper's request payload as dict[str, object]

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
 2026-09-16 01:05:53 +00:00
kerry
801f6a92b4 test(gemini): assert the served modelVersion reaches the assembled stream through CustomStreamWrapper 
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
 2026-09-16 01:05:53 +00:00
yucheng
dd4829398a fix(guardrails): log the configured mode when logging_only is mixed with an enforcing llm_as_a_judge mode 
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
 2026-09-16 01:05:53 +00:00
yucheng
10506ab904 fix(guardrails): judge the whole latest user turn, run every during_call guardrail, log combined modes as configured 
Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
 2026-09-16 01:05:53 +00:00
yucheng
1b8b17cc8a fix(guardrails): judge only the latest request turn in pre_call and during_call llm_as_a_judge 
The request-side prompt told the judge to focus on the most recent user turn but the text under review was every extracted request message joined together, so a multi-turn request with an off-topic earlier turn and an on-topic latest turn scored 50 and was blocked. Request-side judging now evaluates the last extracted request text (after the configured message scoping) and passes the full role-labelled conversation only as context. Response-side judging still evaluates all extracted response text

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
 2026-09-16 01:05:53 +00:00