Setup, client by client

Every config below passed a real run on the API: a temp repo, the client pointed at genmux, an instruction to create a file, the file checked afterwards. Replace nb_live_... with your key; qwen3.8-max is a sensible default, any tool-calling model works.

OpenCode

OpenCode takes any OpenAI-compatible provider via the @ai-sdk/openai-compatible package. Put the block below in opencode.json (project, or ~/.config/opencode/opencode.json), export GENMUX_API_KEY, and pick a model as genmux/<id>. Each request carries ~20 tools and a 32k max_tokens — both fine: the definitions cost about 1.5K prompt tokens, and max_tokens is clamped to the model's ceiling.

opencode.json

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "genmux": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "genmux",
      "options": { "baseURL": "https://api.genmux.tech/v1", "apiKey": "{env:GENMUX_API_KEY}" },
      "models": {
        "qwen3.8-max": { "name": "Qwen3.8-Max" },
        "doubao-seed-2-pro": { "name": "Doubao Seed 2 Pro" }
      }
    }
  }
}

Run

export GENMUX_API_KEY=nb_live_...
opencode --model genmux/qwen3.8-max
# or one-shot:
opencode run --model genmux/qwen3.8-max "add a README with install steps"

Aider

Aider talks to any OpenAI-style base URL via the openai/ model prefix and uses its own edit formats, so it needs no tool calling. Three environment variables and a model flag are the whole setup. For smaller models --edit-format whole is most robust; the 'unknown model' metadata warning aider prints for our ids is harmless.

Terminal

export OPENAI_API_BASE=https://api.genmux.tech/v1
export OPENAI_API_KEY=nb_live_...
aider --model openai/qwen3.8-max
# smaller models: aider --model openai/qwen3.8-max --edit-format whole

Cline and Roo Code

Both VS Code extensions have an 'OpenAI Compatible' provider. Fill in the settings below, tick 'supports streaming', and leave 'supports images' off unless you pick a vision model (its model page says vision). Tool calls arrive as the usual cards; every write and command still goes through the extension's approval flow.

Settings → API Provider

API Provider	OpenAI Compatible
Base URL	https://api.genmux.tech/v1
API Key	your genmux key (nb_live_...)
Model ID	qwen3.8-max (or any tool-calling model)
Supports streaming	on
Supports images	off, except on vision models

Continue

Continue reads a YAML config from ~/.continue/config.yaml or a project config.yaml. Declare the model with provider: openai, our apiBase and capabilities: [tool_use] so agent mode sends tools; the roles line makes the same model available for chat, edit and apply. The CLI (npm i -g @continuedev/cli, then cn --config ./config.yaml) uses the same file.

config.yaml

name: genmux
version: 0.0.1
schema: v1
models:
  - name: Qwen3.8 Max
    provider: openai
    model: qwen3.8-max
    apiBase: https://api.genmux.tech/v1
    apiKey: nb_live_...
    roles: [chat, edit, apply]
    capabilities: [tool_use]

CLI

npm i -g @continuedev/cli
cn --config ./config.yaml