fix: skip response_format for Deepseek (provider doesn't support it)

[tcconnally]

Problem

When using CrewAI with Deepseek, the API returns:

ERROR: 'response_format type is unavailable now'

Deepseek's API does not support OpenAI's response_format parameter. While llm.py has a supports_response_schema check (via LiteLLM), the native OpenAI completion handler in openai/completion.py adds response_format unconditionally.

Fix

Two changes:

1. base_llm.py: Added PROVIDERS_WITHOUT_RESPONSE_FORMAT constant
2. openai/completion.py: Gate the response_format parameter on the provider check

Before

if self.response_format is not None:
    params["response_format"] = ...

After

if (
    self.response_format is not None
    and self._extract_provider(self.model)
    not in {"deepseek"}
):
    params["response_format"] = ...

Closes #5990

Summary by CodeRabbit

• Bug Fixes
  • Improved compatibility with LLM providers that don’t support response_format (notably DeepSeek). The system now applies response_format only when the selected provider supports it, avoiding errors and improving stability when using alternative model providers.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: b8554eda-76cc-427b-a159-589bef30bb8e

📥 Commits

Reviewing files that changed from the base of the PR and between f836370 and 738051d.

📒 Files selected for processing (1)

• lib/crewai/src/crewai/llms/providers/openai/completion.py

🚧 Files skipped from review as they are similar to previous changes (1)

• lib/crewai/src/crewai/llms/providers/openai/completion.py

---

📝 Walkthrough

Walkthrough

A new module-level constant PROVIDERS_WITHOUT_RESPONSE_FORMAT = {"deepseek"} is added to base_llm.py. OpenAICompletion._prepare_completion_params imports this constant and wraps the params["response_format"] assignment in a conditional that skips it when the extracted provider is in that set.

Changes

DeepSeek response_format guard

Layer / File(s)	Summary
response_format provider constant and conditional guard lib/crewai/src/crewai/llms/base_llm.py, lib/crewai/src/crewai/llms/providers/openai/completion.py	PROVIDERS_WITHOUT_RESPONSE_FORMAT set is declared with {"deepseek"}; _prepare_completion_params imports and checks this set so DeepSeek requests omit the response_format parameter.

Estimated code review effort

🎯 1 (Trivial) | ⏱️ ~3 minutes

Poem

🐇 A rabbit hops past DeepSeek's gate,
Knowing response_format won't cooperate.
A tidy set, a simple check,
Keeps 400 errors off the deck.
No more crashes, smooth as clover —
The bug, at last, is finally over! 🌿

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title 'fix: skip response_format for Deepseek (provider doesn't support it)' clearly and concisely describes the main change: conditionally skipping the response_format parameter for Deepseek due to API incompatibility.
Linked Issues check	✅ Passed	The pull request fully addresses issue #5990 by implementing conditional response_format handling for Deepseek, preventing the 400 API error and allowing CrewAI to work with Deepseek's LLM service.
Out of Scope Changes check	✅ Passed	All changes are directly related to the stated objective: adding a provider blocklist constant and modifying the OpenAI completion handler to respect it for Deepseek compatibility.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@lib/crewai/src/crewai/llms/base_llm.py`:
- Around line 129-131: The completion.py file uses a hardcoded {"deepseek"}
literal to check if a provider supports response_format, instead of importing
and using the PROVIDERS_WITHOUT_RESPONSE_FORMAT constant already defined in
base_llm.py. To fix this, add an import statement for
PROVIDERS_WITHOUT_RESPONSE_FORMAT from the base_llm module into completion.py,
then replace the hardcoded {"deepseek"} set literal in the provider check logic
with the imported PROVIDERS_WITHOUT_RESPONSE_FORMAT constant to eliminate
duplication and centralize the provider blocklist.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 708376cb-d6b0-4a1c-9a9b-8a3875b4b02f

📥 Commits

Reviewing files that changed from the base of the PR and between a5cc6f6 and f836370.

📒 Files selected for processing (2)

• lib/crewai/src/crewai/llms/base_llm.py
• lib/crewai/src/crewai/llms/providers/openai/completion.py