fix: AgentRunResult.usage API compatibility on pydantic-ai 2.x

[vitali87]

Based on #964 by @subhamruhela13.

Summary

A fresh pip install code-graph-rag resolves pydantic-ai 2.x, where AgentRunResult.usage is a property; the interactive chat then crashes on its first response with TypeError: 'RunUsage' object is not callable at codebase_rag/main.py:655. The dev lock held pydantic-ai 1.102.0 (where usage is still a method), which is why the suite stayed green while every fresh install was broken.

Changes

• response.usage() -> response.usage (cherry-picked from fix: AgentRunResult.usage API compatibility #964, authorship preserved)
• pydantic-ai>=2.0.0 floor (the property landed in 2.0.0; verified GoogleCloudProvider, AgentRetries, and the deferred-tools API all exist there) and uv.lock upgraded to 2.24.0
• Vertex model creation moved from GoogleProvider(project=..., location=..., credentials=...) (removed in 2.x) to GoogleCloudProvider, the same construction path 1.102.0 delegated to behind a deprecation warning
• Dropped the griffe>=1.0,<2 workaround: pydantic-ai 2.x declares griffelib, which installs the same top-level griffe/ package, so the old pin made both packages write one directory
• Regenerated the docs dependency section

Review fixes applied

• CodeRabbit on fix: AgentRunResult.usage API compatibility #964: the three usage.return_value mocks in test_model_switching.py now assign RunUsage property-style, so the loop consumes realistic token data
• Added a RED->GREEN test pinning the Vertex branch to GoogleCloudProvider(project/location/credentials); without it the suite passes even if the Vertex path is reverted to the 1.x constructor

Test plan

• uv run pytest -m "not integration and not e2e and not slow": 6459 passed; test_vector_store.py + test_unixcoder_unit.py pass with the semantic/milvus extras installed
• Real gesture: cgr start chat against local Ollama completes with real RunUsage token accounting instead of the TypeError
• ty check clean with the repo's pre-commit scoping

Summary by CodeRabbit

• New Features

  • Improved Google Vertex AI integration with clearer project, region, and credential configuration.
  • Updated model usage handling for more reliable agent responses and usage reporting.

• Bug Fixes

  • Fixed compatibility with the latest supported AI framework version.
  • Corrected response usage processing to prevent runtime errors.

• Documentation

  • Simplified installation guidance by removing an obsolete setup requirement.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: cb5bc261-a6c4-4554-85c7-6d76feef9de3

📥 Commits

Reviewing files that changed from the base of the PR and between 43451bb and 3536615.

⛔ Files ignored due to path filters (1)

• uv.lock is excluded by !**/*.lock

📒 Files selected for processing (6)

• codebase_rag/main.py
• codebase_rag/providers/base.py
• codebase_rag/tests/test_model_switching.py
• codebase_rag/tests/test_provider_classes.py
• docs/getting-started/installation.md
• pyproject.toml

💤 Files with no reviewable changes (1)

• docs/getting-started/installation.md

---

📝 Walkthrough

Walkthrough

The PR updates pydantic-ai compatibility, changes response usage access to a property, and switches Google Vertex model creation to GoogleCloudProvider. Tests cover response usage mocks and Vertex provider construction.

Changes

Provider and usage compatibility

Layer / File(s)	Summary
Usage API upgrade pyproject.toml, codebase_rag/main.py, codebase_rag/tests/test_model_switching.py, docs/getting-started/installation.md	The minimum pydantic-ai version is 2.0.0. Response usage is read as a property, and test responses provide RunUsage directly. The installation guide removes griffe from the key dependencies list.
Vertex provider construction codebase_rag/providers/base.py, codebase_rag/tests/test_provider_classes.py	Google Vertex model creation uses GoogleCloudProvider with the configured project, region, and credentials. Tests verify the provider and GoogleModel calls.

Estimated code review effort: 2 (Simple) | ~10 minutes

Possibly related PRs

• vitali87/code-graph-rag#964: Updates the same response.usage() to response.usage compatibility behavior.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly summarizes the main pydantic-ai 2.x compatibility fix and uses Conventional Commits format.
Description check	✅ Passed	The description clearly explains the changes, issue context, and test results, but it omits the template's Type of Change and Checklist sections.
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.

✨ 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 fix/agent-run-result-usage-property

---

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.}

[greptile-apps]

Greptile Summary

Updates pydantic-ai compatibility for current releases: agent usage accounting reads the result usage property, and Vertex model creation uses GoogleCloudProvider. The executed compatibility check against the locked pydantic-ai 2.24.0 dependency confirmed that the updated usage and Vertex construction paths complete successfully. No defects were found.

Confidence Score: 5/5

T-Rex Logs

What T-Rex did

• Ran the executable Pydantic AI migration validation script against pydantic-ai==2.24.0 and confirmed that the superseded usage raises TypeError while the migrated usage and Vertex wiring complete with VALIDATION_OK.
• Documented the before-and-after verification flow, noting the exact script source and the after-output log that shows all checks passing with exit code 0.

_{Ran code and verified through T-Rex}

_{Reviews (1): Last reviewed commit: "test: pin the Vertex branch of GooglePro..." | Re-trigger Greptile}

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
100.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud