fix: preserve provider on LiteLLM-routed models

[lucasgomide]

LiteLLM-routed models kept LLM.provider stuck on the openai default because construction never passed the computed provider into init. Downstream consumers, including the model-provider allowlist, therefore treated groq, cohere, mistral, and similar calls as openai. BaseLLM now infers the provider from a provider/model prefix when none is supplied.

---

Note

Medium Risk
Changes default provider inference and multimodal/instructor wiring for all LiteLLM-backed models; incorrect behavior could mis-route allowlists or break vision/file payloads, though scope is narrow and well covered by new tests.

Overview
LiteLLM-routed models (groq/…, cohere/…, etc.) no longer default provider to openai when it is omitted at construction. BaseLLM now sets provider from the model’s prefix/model string via _extract_provider, so allowlists and other consumers see the real vendor while routing still goes through LiteLLM.

Multimodal file handling is split from provider identity: sync/async message file processing uses _multimodal_formatter_name() instead of self.provider. On the LLM LiteLLM path, provider stays (e.g. anthropic) but formatting uses the openai content-block schema LiteLLM expects on the wire.

InternalInstructor builds the instructor model id without double-prefixing when the model string is already qualified (e.g. groq/llama-3.3-70b).

Tests cover prefix → provider on LiteLLM, multimodal formatter behavior, and instructor qualification.

^{Reviewed by Cursor Bugbot for commit 948b22b. Bugbot is set up for automated code reviews on this repo. Configure here.}

[coderabbitai]

Note

Reviews paused

It looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the reviews.auto_review.auto_pause_after_reviewed_commits setting.

Use the following commands to manage reviews:

• @coderabbitai resume to resume automatic reviews.
• @coderabbitai review to trigger a single review.

Use the checkboxes below for quick actions:

• ▶️ Resume reviews
• 🔍 Trigger review

📝 Walkthrough

Walkthrough

The change extracts providers from prefixed model identifiers for LiteLLM routing and prevents duplicate prefixes when creating Instructor clients. Tests cover OpenAI, Anthropic, Groq, Together, and Groq-qualified models.

Changes

Provider routing

Layer / File(s)	Summary
Provider extraction and fallback validation lib/crewai/src/crewai/llms/base_llm.py, lib/crewai/tests/test_llm.py	BaseLLM extracts the provider from string model identifiers. Non-string models retain the "openai" fallback. Tests verify provider and model preservation across LiteLLM fallback cases.
Instructor qualified model handling lib/crewai/src/crewai/utilities/internal_instructor.py, lib/crewai/tests/utilities/test_converter.py	InternalInstructor preserves an existing provider-qualified model and prepends the provider only when it is absent. Tests cover groq/llama-3.3-70b forwarding.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Title check	✅ Passed	The title clearly summarizes the main change: preserving provider identity for LiteLLM-routed models.
Description check	✅ Passed	The description directly explains provider inference, instructor qualification, multimodal handling, and related test coverage.

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch luzk/fix-llm-provider

---

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.}

[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 277-280: Make provider qualification idempotent across BaseLLM and
the downstream object path: preserve the original qualified model while setting
provider, and ensure the internal instructor path does not prepend provider when
self.llm.model already starts with it. Update the LiteLLM routing flow
accordingly and add a regression test covering a model such as
groq/llama-3.3-70b to verify it is passed only once.

🪄 Autofix

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: 1d754196-b76d-4390-9e6d-c19b0d4457eb

📥 Commits

Reviewing files that changed from the base of the PR and between 18c52c4 and 5a2b3c2.

📒 Files selected for processing (2)

• lib/crewai/src/crewai/llms/base_llm.py
• lib/crewai/tests/test_llm.py

[cursor]

Cursor Bugbot has reviewed your changes using default effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit e655682. Configure here.}