Skip to main content
Corvex Token Factory connects to Kilo Code as a custom provider that uses the OpenAI Compatible API. Kilo Code can fetch the model list from the public Token Factory catalog, so you select models instead of typing their IDs.

​
Prerequisites

1

Add a custom provider

Open the Kilo Code settings (gear icon), choose Global Config, select the Providers tab, scroll to the bottom, and choose Custom provider.
2

Enter the Token Factory details

After you enter the Base URL, Kilo Code fetches available models from GET /v1/models. Under Models, choose zai-org/GLM-5.3, deepseek-ai/DeepSeek-V4-Flash-0731, or both. Leave the reasoning toggle on and the image toggle off for each model, then select Submit. If a model does not appear, add its catalog ID manually.
3

Set the model limits

Kilo Code does not read context limits from the endpoint and assumes a 256K window. Open the global config at ~/.config/kilo/kilo.jsonc. The provider you saved in the previous step is already there under provider.corvex; add limit to each model you selected:
Kilo Code reads this file when it starts, so reload the VS Code window (Developer: Reload Window) after saving. The output limit is a per-response budget and is sent as max_tokens; leave room in the context window for the request.Neither model accepts image input.
4

Select the model and verify the connection

New sessions start on Auto Free, Kilo Code’s own model router, which does not use Token Factory. Select the model button next to the mode selector under the chat box, find the Corvex Token Factory group, and choose zai-org/GLM-5.3. The button then shows the model ID and the context bar shows the configured window (393.2K).Start a task that reads a file in your repository. A completed Read tool call confirms the provider configuration and key.

​
Compatibility notes

  • Model IDs. Use the exact id from GET /v1/models. IDs are case-sensitive and use namespace/model format. Do not add a corvex/ prefix.
  • Context length. If a long task exceeds the model’s window, the gateway returns context_length_exceeded. Start a new task or shorten the context. See Errors.
  • Reasoning. Both models reason before answering. Reasoning tokens count toward the output budget.
  • Connect Cline — the same provider settings in a related extension.
  • Use the OpenAI SDK — the same endpoint from Python or TypeScript.
  • Errors — authentication, rate-limit, and validation responses.
Last modified on September 28, 2026