An 8MB executable that turns ChatGPT clients into RWKV users

What it does

RWKV-Runner is an automation layer and management GUI for the RWKV open-source language model. It stuffs model downloads, dependency resolution, inference startup, and an OpenAI-compatible API into a single lightweight executable—roughly 8 MB—so existing ChatGPT clients can talk to RWKV without code changes. The frontend and backend can also split apart, letting you run the inference server headless or deploy the WebUI separately.

The interesting bit

The project bundles features you would not expect from a mere model launcher: built-in MIDI hardware input for composition, a model conversion tool, and one-click LoRA fine-tuning on Windows. It also works in reverse—flip the API URL in settings and it becomes a generic client for OpenAI, Ollama, or GPT-Playground.

Key highlights

• OpenAI API compatibility out of the box; every ChatGPT client becomes an RWKV client automatically.
• Runs on modest hardware with pre-set VRAM tiers and a WebGPU fallback for AMD/Intel GPUs.
• Front-end/back-end separation supports headless server deployment or standalone WebUI hosting.
• Built-in extras: model conversion, download management, embeddings endpoint, and LoRA fine-tune (Windows only).
• Can proxy to commercial APIs too, acting as a unified chat/completion client for OpenAI and Ollama.

Caveats

• The README warns that Windows Defender sometimes flags the executable as malware.
• One-click LoRA fine-tuning is Windows-only; other platforms need a different workflow.
• Public deployments require manual guardrails: the default max_tokens ceiling is high enough to cause resource spikes if left unrestricted.

Verdict

Grab it if you want a low-friction way to run RWKV locally or self-host an OpenAI-compatible endpoint without wrestling with Python environments. Skip it if you are already comfortable running rwkv.cpp or web-rwkv directly and do not need the GUI sugar.