[netbrah]

Fixes #23385

What

Add OpenAIContentGenerator that implements the existing ContentGenerator interface, enabling gemini-cli to work with any OpenAI-compatible API endpoint — OpenAI, Anthropic Claude (via proxy), local models (Ollama, vLLM), and corporate LLM gateways.

Why

Multi-provider support is the most impactful way to grow gemini-cli's community. The agent loop, MCP integration, tool system, and TUI are excellent — but they're currently locked to the Gemini API. This PR unlocks them for every OpenAI-compatible model while preserving all Gemini-native functionality unchanged.

This is battle-tested: we've been running this architecture against Claude Opus 4, GPT-4o, and corporate LLM proxies on large codebases (100K+ files) for weeks.

How

File	Change
packages/core/src/core/openaiContentGenerator.ts	New 740-line file: Content/Part ↔ OpenAI message translation, SSE streaming, tool call accumulation, context budget enforcement, schema cleaning
packages/core/src/core/contentGenerator.ts	Add OPENAI_COMPATIBLE auth type, env var detection, factory wiring
packages/core/src/core/tokenLimits.ts	Pattern-based matching for Claude, GPT, o1/o3, Codex models
packages/cli/src/ui/utils/updateCheck.ts	Skip update banner when running via proxy
packages/core/src/core/openaiContentGenerator.test.ts	11 tests: non-streaming, streaming, tool calls, schema cleaning, model defaults, conversation translation
packages/core/src/core/tokenLimits.test.ts	Extended with 7 new model limit tests

Activation (zero-config for existing users):

export OPENAI_COMPATIBLE_ENDPOINT="https://api.openai.com"
export OPENAI_COMPATIBLE_API_KEY="sk-..."
export OPENAI_COMPATIBLE_MODEL="gpt-4o"  # optional

Test Plan

• npx vitest run packages/core/src/core/openaiContentGenerator.test.ts — 11 tests pass
• npx vitest run packages/core/src/core/tokenLimits.test.ts — 6 tests pass (including 7 new model cases)
• npx tsc --noEmit -p packages/core/tsconfig.json — clean
• Pre-commit hooks (ESLint + Prettier) pass
• Zero changes to existing Gemini flow — all existing tests unaffected

Made with Cursor

[gemini-code-assist]

Summary of Changes

Hello, I'm Gemini Code Assist¹! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!

This pull request introduces significant enhancements to the gemini-cli by integrating support for OpenAI-compatible API endpoints. This change broadens the tool's applicability, allowing users to leverage a wider array of large language models and services. The implementation includes a new content generator, intelligent context management, and seamless translation between different API formats, ensuring that the core functionalities of gemini-cli are accessible across diverse LLM ecosystems.

Highlights

• OpenAI-Compatible Provider Support: Added a new OpenAIContentGenerator that implements the existing ContentGenerator interface, allowing gemini-cli to work with any OpenAI-compatible API endpoint, including OpenAI, Anthropic Claude (via proxy), local models (Ollama, vLLM), and corporate LLM gateways.
• Enhanced Multi-Provider Functionality: Unlocks the agent loop, MCP integration, tool system, and TUI for every OpenAI-compatible model, significantly growing the gemini-cli community by extending its capabilities beyond the Gemini API while preserving all Gemini-native functionality.
• Context Budget Enforcement and Tool Handling: Implemented pre-send context budget enforcement by estimating token counts and trimming tool outputs, along with robust translation for content, tool calls, and function responses between Gemini and OpenAI formats, including SSE streaming accumulation.
• Dynamic Token Limits and Update Check Logic: Updated token limit detection to include pattern-based matching for various OpenAI-compatible models (e.g., Claude, GPT, Codex) and modified the update check mechanism to skip when running via an OpenAI-compatible proxy.

Using Gemini Code Assist

The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.

Invoking Gemini

You can request assistance from Gemini at any point by creating a comment using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands on the current page.

Feature	Command	Description
Code Review	/gemini review	Performs a code review for the current pull request in its current state.
Pull Request Summary	/gemini summary	Provides a summary of the current pull request in its current state.
Comment	@gemini-code-assist	Responds in comments when explicitly tagged, both in pull request comments and review comments.
Help	/gemini help	Displays a list of available commands.

Customization

To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.

Limitations & Feedback

Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for GitHub and other Google products, sign up here.

Footnotes

1. Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩

[gemini-code-assist]

Code Review

This pull request introduces support for OpenAI-compatible providers by adding a new OpenAIContentGenerator. The changes are extensive, including the new generator, modifications to the content generator factory, and updates to token limit calculations. The implementation is thorough and includes comprehensive tests.

My review identified a potential issue in the new OpenAIContentGenerator. There's a risk of silently dropping tool responses when multiple responses lack an ID due to how they are deduplicated. This issue is flagged as high severity and should be addressed to ensure correctness and robustness.

[gemini-code-assist]

There's a potential for data loss here when handling functionResponse parts. If multiple functionResponse parts in the contents array are missing an id, they will all be assigned an empty string '' as their frId. Consequently, they will overwrite each other in the seenToolIds map, and only the last one will be processed. This could lead to silently dropping tool responses, causing incorrect behavior in the agent loop.

[netbrah]

Fixed in 38cebcd — functionResponse parts without an id now get unique index-based fallback keys (_fr_0, _fr_1, ...) instead of collapsing into empty string. No data loss.

[gemini-cli]

Hi there! Thank you for your interest in contributing to Gemini CLI.

To ensure we maintain high code quality and focus on our prioritized roadmap, we have updated our contribution policy (see Discussion #17383).

We only guarantee review and consideration of pull requests for issues that are explicitly labeled as 'help wanted'. All other community pull requests are subject to closure after 14 days if they do not align with our current focus areas. For this reason, we strongly recommend that contributors only submit pull requests against issues explicitly labeled as 'help-wanted'.

This pull request is being closed as it has been open for 14 days without a 'help wanted' designation. We encourage you to find and contribute to existing 'help wanted' issues in our backlog! Thank you for your understanding and for being part of our community!