feat: Add support for GLM

[mylvghb]

Model Provider Integration Request

GLM (Zhipu AI) is a leading large language model provider offering a series of models based on the GLM (General Language Model) architecture. It provides an OpenAI-compatible API format, making integration straightforward. Both domestic (China) and international endpoints are available.

Models

Model	Description
glm-4	Flagship model with complex reasoning, multi-turn dialogue, and tool calling
glm-4-air	Cost-effective version suitable for most applications
glm-4-flash	Ultra-fast response version for simple tasks
glm-4v	Multimodal model supporting image understanding
glm-3-turbo	Lightweight model for low-latency scenarios

Key Features

• OpenAI-compatible API: RESTful API with OpenAI-compatible request/response format
• Tool Calling (Function Calling): Supports structured tool/function invocation
• Streaming: Supports SSE streaming output
• Multimodal: GLM-4V supports image inputs alongside text

Resources

Domestic (China)

• Official Docs: https://open.bigmodel.cn/dev/api
• API Reference: https://open.bigmodel.cn/dev/api#glm-4
• Model List: https://open.bigmodel.cn/modelcenter/square
• Base URL: https://open.bigmodel.cn/api/paas/v4/

International (Global)

• Official Site: https://z.ai
• API Docs: https://docs.z.ai
• Base URL: https://api.z.ai/api/paas/v4/
• Authentication: API Key-based via Authorization: Bearer {api_key} header

Integration Notes

Since Zhipu AI provides an OpenAI-compatible API, the existing OpenAI provider implementation could potentially be reused with the following adjustments:

1. Allow custom base_url pointing to the domestic (open.bigmodel.cn) or international (api.z.ai) endpoint
2. Add GLM-specific model name mappings
3. Handle any minor API differences between domestic and international versions (e.g., specific parameter support, rate limits)

Checklist

• I have checked that this provider is not already supported
• I have reviewed the existing provider implementations for reference

[linear-code]

RIG-1343

[HerShin5]

+1 on GLM support. One implementation detail I’d suggest: make it a generic OpenAI-compatible provider path with an overridable base_url, rather than tying it too tightly to only one Zhipu endpoint.

That would cover both official GLM endpoints listed here (domestic + international), and also the common case where teams route GLM or other models through an internal gateway/proxy for billing, fallback, rate limits, or model aliases. I’ve used the same pattern with OpenAI-compatible gateways like https://kiroforge.cloud/v1: the client only needs configurable base URL + API key + model ID, while routing/model availability can stay server-side.

So for Rig, maybe the GLM provider could reuse most of the existing OpenAI-compatible request path, with provider-specific defaults for model names and docs, but still allow users to override the endpoint.

[oiwn]

GLM is currently version 5, their docs could be outdated.

[mylvghb]

@HerShin5 I also suggest adding OpenAI-compatible and Anthropic-compatible, so that other providers not listed in this creates can be integrated.

[mylvghb]

GLM is currently version 5, their docs could be outdated.

Got it, noted that GLM is now at version 5.2.