Wally
Use Wally's fast open-weight models in Dify: chat models with tool calling for agents and workflows, reasoning, and image input on DeepSeek V4.1 Flash, DeepSeek V4.1 Flash (Fast) and Kimi K3.
Wally serves an OpenAI-compatible API at . This plugin points Dify at it, so you don't need to configure a generic OpenAI-compatible provider by hand.
Setup
- Sign in to the RunAnywhere console and create a Wally API key ().
- In Dify, install Wally from the Marketplace.
- Go to Settings → Model Provider → Wally, paste the API key and save. Dify checks the key with one short request.
Usage
The models below appear in every model picker: LLM nodes, chatflows and agents.
- Agents and tools: all eight models support tool calling, including parallel and streamed tool calls.
- Reasoning: Dify shows the model's reasoning as a collapsible thought. Set Reasoning effort in the model parameters to choose how much the model thinks, or leave it empty for the model default.
- GLM-5.3 Flash, GLM-5.3 (Fast) and MiniMax M3 always think; the lowest level is on GLM-5.3 Flash and on the other two.
- DeepSeek V4.1 Flash, DeepSeek V4.1 Flash (Fast) and Kimi K3 think by default; turns thinking off.
- DeepSeek V4 Pro (Fast) and DeepSeek V4 Flash 0731 don't think unless you pick a level.
- Structured output: set Response Format to , or to and paste a JSON schema.
- Other models: to use a Wally model that isn't listed yet, choose Add Model, enter its ID exactly as returns it, and set its context size, tool calling and image input options.
Prices are on runwally.com/pricing. API reference: docs.runanywhere.ai/wally-cloud/api.
Credentials
One Wally API key. Dify stores it encrypted and sends it as a Bearer token on each request.
Connection requirements
Your Dify instance needs outbound HTTPS access to (port 443). No other endpoint is called.
Source and support