fix(bedrock): strip cross-region inference profile prefixes for context window

[Solaris-star]

Description

AWS Bedrock cross-region inference profile model ids look like us.anthropic.claude-sonnet-4-.... BedrockCompletion.get_context_window_size() keyed its table on bare ids and used self.model.startswith(...), so prefixed profiles never matched and fell through to the 8192 default. That makes a 200K-token Claude model behave like ~8K and triggers premature summarization.

This PR:

• strips bedrock/ and geographic prefixes (us-gov., apac., us., eu., …) before lookup
• matches longest table prefix first (so claude-3-5-sonnet beats claude-3)

Fixes #6244

Test Coverage

Local (Python 3.12 + crewai[bedrock]), direct unit construction with mocked Session:

anthropic.claude-3-5-sonnet-... → 170000
us.anthropic.claude-sonnet-4-... → 170000
eu./apac. profiles → 170000
amazon.titan-text-express → 6800

Added test_bedrock_cross_region_inference_profile_context_window.

Checklist

• Bug fix
• Tests added
• Self-reviewed

[coderabbitai]

📝 Walkthrough

Walkthrough

Changes

Bedrock context sizing

Layer / File(s)	Summary
Prefix-aware context-window lookup lib/crewai/src/crewai/llms/providers/bedrock/completion.py, lib/crewai/tests/llms/bedrock/test_bedrock.py	BedrockCompletion strips geographic inference-profile prefixes before matching context-window entries, prioritizes longer prefixes, and tests regional model identifiers against the expected 200,000-token context size.

Suggested reviewers: lorenzejay

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly matches the main change: fixing Bedrock context-window lookup for cross-region inference profile prefixes.
Description check	✅ Passed	The description directly describes the same Bedrock prefix fix and added regression test.
Linked Issues check	✅ Passed	The PR satisfies the issue's primary context-window fallback bug by stripping geographic prefixes and adding coverage.
Out of Scope Changes check	✅ Passed	The changes stay focused on Bedrock context-window prefix handling and related tests, with no obvious unrelated additions.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.

✨ 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.}

[coderabbitai]

🧹 Nitpick comments (1)

lib/crewai/src/crewai/llms/providers/bedrock/completion.py (1)

2093-2106: 🎯 Functional Correctness | 🔵 Trivial | ⚡ Quick win

Apply prefix stripping to supports_multimodal and handle case-insensitivity.

The PR objective notes that the linked issue also identifies a prefix problem in supports_multimodal for non-us. profiles. You can fully resolve that by reusing _strip_inference_profile_prefix there and removing the hardcoded us. variations.

Additionally, standardizing the parsed model ID to lowercase prevents mismatches if a user passes mixed-casing (e.g., US.Anthropic...).

♻️ Proposed refactor to share logic and handle casing

     `@staticmethod`
     def _strip_inference_profile_prefix(model_id: str) -> str:
         """Strip geographic cross-region inference profile prefixes.
 
         AWS Bedrock inference profile ids look like
         ``us.anthropic.claude-sonnet-4-...`` / ``eu.`` / ``apac.`` / ``us-gov.``.
         Context-window tables are keyed on the bare foundation-model id.
         """
-        model_id = model_id.removeprefix("bedrock/")
+        model_id = model_id.lower().removeprefix("bedrock/")
         for geo in ("us-gov.", "apac.", "us.", "eu.", "jp.", "au.", "ca."):
             if model_id.startswith(geo):
                 return model_id[len(geo) :]
         return model_id

To complete the issue fix, you can then apply this helper to supports_multimodal() outside the current diff:

    def supports_multimodal(self) -> bool:
        ...
        model_lower = self._strip_inference_profile_prefix(self.model)
        vision_models = (
            "anthropic.claude-3",
            "anthropic.claude-sonnet-4",
            "anthropic.claude-opus-4",
            "anthropic.claude-haiku-4",
            "amazon.nova-lite",
            "amazon.nova-pro",
            "amazon.nova-premier",
            # removed the us.* duplicates
        )
        return any(model_lower.startswith(m) for m in vision_models)

🤖 Prompt for 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.

In `@lib/crewai/src/crewai/llms/providers/bedrock/completion.py` around lines 2093
- 2106, Update supports_multimodal to normalize self.model through
_strip_inference_profile_prefix and lowercase it before matching. Remove the
hardcoded us. model variants from its vision_models list, retaining only bare
model prefixes so all supported geographic profiles and mixed-cased IDs are
handled consistently.

🤖 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.

Nitpick comments:
In `@lib/crewai/src/crewai/llms/providers/bedrock/completion.py`:
- Around line 2093-2106: Update supports_multimodal to normalize self.model
through _strip_inference_profile_prefix and lowercase it before matching. Remove
the hardcoded us. model variants from its vision_models list, retaining only
bare model prefixes so all supported geographic profiles and mixed-cased IDs are
handled consistently.

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 9b7a5475-70d5-4b93-a6f5-55900f5e7b2a

📥 Commits

Reviewing files that changed from the base of the PR and between 69c0308 and 4ea6ca5.

📒 Files selected for processing (2)

• lib/crewai/src/crewai/llms/providers/bedrock/completion.py
• lib/crewai/tests/llms/bedrock/test_bedrock.py