Use MiniMax M3 with Roo Code — velqa.dev
Roo Code is a fork of Cline with advanced mode profiles (Code, Architect, Ask, Debug). It is configured the same way for an OpenAI-compatible provider.
Prerequisites
- VS Code with the Roo Code extension installed
- A velqa.dev API key (see the quickstart)
Configuration
- Open Roo Code in the VS Code sidebar
- Click the Settings icon (gear)
- Under API Provider, select OpenAI Compatible
- Fill in the fields:
| Field | Value |
|---|---|
| Base URL | https://api.velqa.dev/v1 |
| API Key | sk-... |
| Model ID | minimax-m3 |
| Context Window Size | 131072 |
| Max Output Tokens | 16384 |
- Click Save
Context window and output (important)
Fill in Context Window Size and Max Output Tokens in the OpenAI Compatible provider's advanced settings. They tell Roo Code when to compact the context. Without them, a long agentic session lets the context grow unbounded and eventually triggers Trop de tokens demandes. Reessaie plus tard., while burning through your quota faster.
Values per model:
| Model ID | Context Window Size | Max Output Tokens |
|---|---|---|
minimax-m3 | 131072 | 16384 |
hy3 | 262144 | 16384 |
glm-4.7-flash | 200000 | 16384 |
deepseek-v4-flash | 65536 | 8192 |
Available models
| Model ID | Roo Code use case |
|---|---|
minimax-m3 | Code mode — generation and refactoring |
glm-4.7-flash | Ask mode — quick questions, documentation |
deepseek-v4-flash | High-volume, economical |
Tool calling compatibility
Roo Code requires OpenAI native tool calling. MiniMax M3 supports this format. Do not configure an XML fallback — it is not supported by Roo Code.
Test
Run a simple task in Ask mode: What is in this folder?. A response with no error confirms that the velqa.dev gateway is responding correctly.
Troubleshooting
- `401 Unauthorized`: incorrect or missing API key
- `Model not found`: check the exact identifier at velqa.dev/models
- No tool calling: check that the Model ID matches exactly a model that supports tools, such as
minimax-m3