litellm/litellm
Mateo Wang 761c280a6e
fix(deepseek): use native /anthropic/v1/messages endpoint and sanitize tools (#28200)
* fix(deepseek): route messages api through anthropic config

Add a DeepSeek-specific Anthropic Messages config so deepseek/... models use the native messages endpoint and preserve thinking blocks. Strip Anthropic custom tool type markers that DeepSeek rejects while keeping hosted tool types intact.

* fix(deepseek): normalize anthropic messages api base

Handle OpenAI-style DeepSeek api_base values ending in /v1 or /v1/messages by stripping those suffixes before adding the /anthropic messages path.

* chore(deepseek): format messages transformation

* chore(deepseek): add test package markers

* fix(deepseek): tighten anthropic url path check and fall back to DEEPSEEK_API_BASE

Author: mateo-berri <277851410+mateo-berri@users.noreply.github.com>

* fix(tests): normalize smart quotes in realtime guardrail refusal check

gpt-realtime nondeterministically returns refusals with Unicode curly
apostrophes (e.g. 'I’m sorry, but I can’t assist with that.'), but the
safe_markers tuple in test_text_message_blocked_by_guardrail_no_ai_response
only contains straight ASCII apostrophes. The substring match then fails
even though the response is a clear refusal, flipping CI red.

Normalize the AI text to ASCII quotes before the marker check so both
straight and curly variants count as safe outcomes.

* fix(deepseek): drop redundant anthropic v1/messages endswith check

* fix(deepseek): strip /beta suffix in anthropic messages URL normalization

Co-authored-by: Yassin Kortam <yassin@berri.ai>

---------

Co-authored-by: Felipe Rodrigues Gare Carnielli <felipe.gare@hotmail.com>
Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: Yassin Kortam <yassin@berri.ai>
2026-05-18 18:14:13 -07:00
..
a2a_protocol feat(guardrails): LLM-as-a-Judge guardrail (#26360) 2026-04-24 17:15:32 -07:00
anthropic_interface style: run black formatter on files from main merge 2026-04-17 13:02:59 -07:00
assistants
batch_completion
batches fix(vertex-ai): fix zero cost/usage on completed Vertex AI batch jobs (#27912) 2026-05-15 04:47:02 -07:00
caching fix(caching): replay openai/responses bridge cache hits as chat streams (#28158) 2026-05-18 16:27:06 -07:00
completion_extras fix(caching): replay openai/responses bridge cache hits as chat streams (#28158) 2026-05-18 16:27:06 -07:00
compression Prompt Compression - add it to the proxy (#25729) 2026-04-20 15:08:00 -07:00
containers style: run black formatter on files from main merge 2026-04-17 13:02:59 -07:00
endpoints/speech/speech_to_completion_bridge
evals style: run black formatter on files from main merge 2026-04-17 13:02:59 -07:00
experimental_mcp_client feat(mcp): add OBO MCP Auth (#27421) 2026-05-07 15:35:21 -07:00
files thread trusted params through vertex file content 2026-05-01 18:24:22 -07:00
fine_tuning address greptile review feedback (greploop iteration 1) 2026-03-27 20:04:41 +05:30
google_genai feat(proxy): LiteLLM headers on Google native generateContent routes (#25500) 2026-04-29 12:34:14 -07:00
images Feat(dashscope): add image generation support for qwen-image-2.0 and qwen-image-2.0-pro (#25672) 2026-04-22 20:03:46 -07:00
integrations feat(prometheus): add user_email and user_alias to user budget metrics (#28155) 2026-05-18 16:28:14 -07:00
interactions style: run black formatter on files from main merge 2026-04-17 13:02:59 -07:00
litellm_core_utils Litellm oss staging (#28161) 2026-05-18 16:27:44 -07:00
llms fix(deepseek): use native /anthropic/v1/messages endpoint and sanitize tools (#28200) 2026-05-18 18:14:13 -07:00
ocr chore: reject bare str at file-input sinks to prevent local-file read (#27762) 2026-05-12 16:40:07 -07:00
passthrough fix(passthrough): log when streaming spend-tracking flush fails to schedule 2026-04-30 02:39:29 +00:00
proxy Litellm oss staging (#28161) 2026-05-18 16:27:44 -07:00
proxy_auth
rag style: black formatting 2026-04-25 14:47:54 -07:00
realtime_api
rerank_api Fix review 2026-04-30 09:10:24 +05:30
responses fix(responses): preserve cache_control in Responses API -> Chat Completion transformation (#27727) 2026-05-13 12:17:06 -07:00
router_strategy feat: add weighted-routing failover (#27980) 2026-05-15 17:28:54 +00:00
router_utils fix(responses): register cooldowns on failure + fail fast on stale encrypted_content (#27820) 2026-05-13 09:03:13 -07:00
search style: run black formatter on files from main merge 2026-04-17 13:02:59 -07:00
secret_managers Implement normalize_nonempty_secret_str function to trim whitespace from secrets and treat empty values as unset. Update proxy_server to use this function for Grafana credentials. Enhance tests to validate the new normalization behavior. 2026-05-04 18:17:31 +00:00
skills chore(proxy): scope skills and container resources 2026-04-30 18:23:58 -07:00
types feat(prometheus): add user_email and user_alias to user budget metrics (#28155) 2026-05-18 16:28:14 -07:00
vector_store_files
vector_stores chore(vector stores): tighten managed store access 2026-04-30 15:04:25 -07:00
videos style: run black formatter on files from main merge 2026-04-17 13:02:59 -07:00
__init__.py feat(prometheus): add user_email and user_alias to user budget metrics (#28155) 2026-05-18 16:28:14 -07:00
_internal_context.py fix(proxy): harden request parameter handling 2026-04-16 01:38:12 +00:00
_lazy_imports.py style: run black formatter on files from main merge 2026-04-17 13:02:59 -07:00
_lazy_imports_registry.py fix(utils): import get_secret at runtime (#28014) 2026-05-15 14:01:18 -07:00
_logging.py fix(bedrock-mantle): use /anthropic/v1/messages path for Mantle endpo… (#27976) 2026-05-15 13:31:59 -07:00
_redis.py Litellm oss staging (#28161) 2026-05-18 16:27:44 -07:00
_redis_credential_provider.py feat: add ability to auth to azure with token (#27556) 2026-05-09 22:34:09 +00:00
_service_logger.py
_uuid.py
_version.py
anthropic_beta_headers_config.json fix(anthropic,bedrock,databricks): four reasoning_effort follow-ups 2026-05-03 10:03:53 -07:00
anthropic_beta_headers_manager.py
blog_posts.json
budget_manager.py docs(budget_manager): add docstring to BudgetManager.reset_cost (#27867) 2026-05-13 13:28:22 -07:00
constants.py feat(mcp): support MCP access group names in URL-based namespacing (#27726) 2026-05-13 20:20:38 -07:00
cost.json
cost_calculator.py Litellm oss staging (#28161) 2026-05-18 16:27:44 -07:00
exceptions.py Litellm oss staging (#28161) 2026-05-18 16:27:44 -07:00
main.py Litellm oss staging (#28161) 2026-05-18 16:27:44 -07:00
model_prices_and_context_window_backup.json feat(model_catalog): add Azure AI Foundry GPT-5.4 model metadata (#28030) 2026-05-16 15:08:10 -07:00
mypy.ini
policy_templates_backup.json
provider_endpoints_support_backup.json Feature/add audio support for scaleway (#26110) 2026-04-20 14:49:41 -07:00
py.typed
router.py Litellm oss staging (#28161) 2026-05-18 16:27:44 -07:00
scheduler.py
setup_wizard.py style: run black formatter on files from main merge 2026-04-17 13:02:59 -07:00
timeout.py docs: add class docstring to _LoopWrapper (#27870) 2026-05-13 13:54:00 -07:00
utils.py fix(deepseek): use native /anthropic/v1/messages endpoint and sanitize tools (#28200) 2026-05-18 18:14:13 -07:00