Private open-weight models in your editor.

Model names for IDE use

Use these exact strings in the model field of each tool's configuration. For general coding assistance, the 7B instruct model is a reasonable starting point.

Model	Notes
qwen/qwen2.5-7b-instruct	General purpose. Good balance of speed and capability.
qwen/qwen2.5-14b-instruct	Stronger reasoning.
qwen/qwen2.5-1.5b-instruct	Fast and lightweight. Suited to autocomplete-style tasks.
mistralai/mistral-7b-instruct-v0.3	Alternative instruct model.

Cursor

Add Marigold as a custom model

Cursor supports any OpenAI-compatible provider via its model settings. Open Settings → Models and add a new model entry.

Settings → Models → Add model

API Provider:  OpenAI-compatible
Base URL:      https://api.marigold.run/v1
API Key:       your-marigold-api-key
Model name:    qwen/qwen2.5-7b-instruct

Alternatively, set these in your Cursor ~/.cursor/settings.json and restart.

~/.cursor/settings.json

{
  "openai.apiKey":  "your-marigold-api-key",
  "openai.baseURL": "https://api.marigold.run/v1"
}

Continue

Add Marigold to config.json

Continue stores its model configuration in ~/.continue/config.json on macOS and Linux, or %USERPROFILE%\.continue\config.json on Windows. Add one or more entries to the models array.

~/.continue/config.json

{
  "models": [
    {
      "title":    "Qwen 7B (Marigold)",
      "provider": "openai",
      "model":    "qwen/qwen2.5-7b-instruct",
      "apiBase":  "https://api.marigold.run/v1",
      "apiKey":   "your-marigold-api-key"
    },
    {
      "title":    "Qwen 14B (Marigold)",
      "provider": "openai",
      "model":    "qwen/qwen2.5-14b-instruct",
      "apiBase":  "https://api.marigold.run/v1",
      "apiKey":   "your-marigold-api-key"
    }
  ]
}

Multiple models can coexist in the array. Continue lets you switch between them from the chat panel. The title field is what appears in the dropdown.

Cline

Set the API provider to OpenAI Compatible

In VS Code, open the Cline extension settings panel. Set the API provider to OpenAI Compatible and fill in the three fields below.

Cline extension settings

API Provider:  OpenAI Compatible
Base URL:      https://api.marigold.run/v1
API Key:       your-marigold-api-key
Model ID:      qwen/qwen2.5-7b-instruct

The model ID field accepts the full Marigold model name including the namespace prefix. Cline stores this in VS Code user or workspace settings.

Aider

Pass the base URL and key on the command line or in config

Aider requires the openai/ prefix on the model name when using an OpenAI-compatible provider. Note this is Aider's own provider prefix, not the model namespace -- use openai/qwen2.5-7b-instruct, not qwen/qwen2.5-7b-instruct.

Command line

aider \
  --openai-api-base https://api.marigold.run/v1 \
  --openai-api-key  your-marigold-api-key \
  --model           openai/qwen2.5-7b-instruct

.aider.conf.yml (project root or home directory)

openai-api-base: https://api.marigold.run/v1
openai-api-key:  your-marigold-api-key
model:           openai/qwen2.5-7b-instruct

Environment variables

export OPENAI_API_BASE=https://api.marigold.run/v1
export OPENAI_API_KEY=your-marigold-api-key

aider --model openai/qwen2.5-7b-instruct

Direct SDK use

Building your own tool or script

Any SDK or library that wraps the OpenAI API accepts a custom base URL. Pass base_url (or the equivalent parameter) and your Marigold key. The model name is the full registry name with namespace prefix.

Python -- openai SDK

from openai import OpenAI

client = OpenAI(
    base_url="https://api.marigold.run/v1",
    api_key="your-marigold-api-key"
)

response = client.chat.completions.create(
    model="qwen/qwen2.5-7b-instruct",
    messages=[{"role": "user", "content": "..."}]
)

Python -- LangChain

from langchain_openai import ChatOpenAI

llm = ChatOpenAI(
    model="qwen/qwen2.5-7b-instruct",
    openai_api_base="https://api.marigold.run/v1",
    openai_api_key="your-marigold-api-key"
)

TypeScript / Node

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://api.marigold.run/v1",
  apiKey:  "your-marigold-api-key",
});

const response = await client.chat.completions.create({
  model:    "qwen/qwen2.5-7b-instruct",
  messages: [{ role: "user", content: "..." }],
});

Common issues