Support for Gemini API

[netgfx]

Problem Statement

The system is not able to utilise Gemini API

Proposed Solution

By supporting Gemini API the agent can be utilized locally or via other systems like CI pipelines

Alternatives Considered

No response

Affected Area

Review Agent / LLM interaction

Additional Context

No response

[stay-foolish-forever]

Hi @netgfx, thanks for raising this! We fully agree with the direction — supporting multiple LLM providers is an important goal for this project.

We're currently designing a more robust and broadly compatible multi-provider model integration solution that will go beyond single-vendor support. This new approach aims to make it easier to plug in different LLM backends (including Gemini and others) in a consistent and well-structured way.

Related changes are expected to land in the near future. We'll keep you posted on the progress.

[roperi]

@stay-foolish-forever

So which vendors are currently supported? Is it only Anthropic ones? Can you do Qwen ? if so, how do i configure it? thank you.

[lizhengfeng101]

Hi @roperi, thanks for the question!

Currently Supported Protocols

OCR currently supports two wire protocols:

1. Anthropic Messages API — for Claude models (default)
2. OpenAI Chat Completions API — for OpenAI models and any OpenAI-compatible provider

This means any provider that exposes an OpenAI-compatible /chat/completions endpoint can be used, including DeepSeek, Azure OpenAI, self-hosted models via vLLM/Ollama, etc.

Using Qwen

Qwen is not natively integrated, but it can work if you use an OpenAI-compatible endpoint (e.g., Alibaba's DashScope provides an OpenAI-compatible mode).

Here's how to configure it:

ocr config set llm.url "https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions"
ocr config set llm.auth_token "sk-your-dashscope-api-key"
ocr config set llm.model "qwen-plus"
ocr config set llm.use_anthropic false

Or via environment variables:

export OCR_LLM_URL="https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions"
export OCR_LLM_TOKEN="sk-your-dashscope-api-key"
export OCR_LLM_MODEL="qwen-plus"
export OCR_USE_ANTHROPIC=false

Note: This relies on the provider's OpenAI protocol compatibility. If you run into any issues, feel free to report them — we'd be happy to help troubleshoot.

Summary

Provider	How
Anthropic (Claude)	Built-in, default protocol
OpenAI (GPT)	Set use_anthropic = false
Qwen (DashScope)	OpenAI-compatible mode, set use_anthropic = false
DeepSeek	OpenAI-compatible endpoint, set use_anthropic = false
Any OpenAI-compatible	Point llm.url to the endpoint, set use_anthropic = false

Hope this helps!

[westfly]

[lizhengfeng101]

It looks like everything is working properly. You can change the language preference by setting the language, for example: ocr config set language English. Do you have any other questions?