TRAE
TRAE can connect to Fusion's Chat Completions or Responses API through a custom model, without CC Switch. Both protocols support Fusion Max 2.0's five reasoning effort levels; use high as recommended. This guide covers the TRAE China desktop client. The international edition, enterprise edition, and older versions may have different controls and availability.
Connection parameters
Use a FUMO Dashboard key for an account registered and funded online.
| Parameter | Value |
|---|---|
| Base URL | https://capi.fumolab.ai/api/fusion/v1 |
| Chat Completions (full URL) | https://capi.fumolab.ai/api/fusion/v1/chat/completions |
| Responses (full URL) | https://capi.fumolab.ai/api/fusion/v1/responses |
| Available model IDs | fusion-max-2.0, fusion-code-1.0 |
| API format | OpenAI Chat Completions / Responses |
| Recommended model | fusion-max-2.0 |
| Input window / output limit | 373000 / 131072 tokens |
| Max 2.0 reasoning effort | Supports low, medium, high, xhigh, max; recommended: high |
Setup
1. Install the latest version and prepare your accounts
Install the latest desktop client from the official TRAE download page↗, and sign in to your TRAE account. Skipping sign-in at first launch does not enable AI features; sign in before using them.
Create a FUMO API key using Getting started, or see Enterprise onboarding for enterprise accounts. Your FUMO API key authenticates model requests; it does not replace TRAE account sign-in.
2. Add a custom model
- In the IDE or Agent window, open Settings → Models from the account menu and click Add model. You can also select Add model directly from the conversation's model menu.
- Select Custom model.
- Choose OpenAI Chat Completions or OpenAI Responses API as the API format.
- Turn on Full URL and enter the matching full Custom request URL. For Chat Completions, use:
https://capi.fumolab.ai/api/fusion/v1/chat/completions
- Set Model ID to:
fusion-max-2.0
- Optionally set Display name to
FUMO · Fusion Max 2.0. Enter the FUMO API key for the selected access channel.
Use the full URL to avoid having the client append the path again. If you turn Full URL off, enter only https://capi.fumolab.ai/api/fusion/v1 and let TRAE append the path for the selected API format.
To use Responses, create a separate custom model configuration:
- Choose OpenAI Responses API as the API format.
- Turn on Full URL and enter this complete request URL:
https://capi.fumolab.ai/api/fusion/v1/responses
- Use the Fusion model ID and API key for the same access channel, then complete the context, image input, and save steps below.
The API format and request URL cannot be changed when editing a saved model; create a new configuration to switch protocols. Do not leave OpenAI Chat Completions selected and change only the URL to /responses. With Full URL off, enter only https://capi.fumolab.ai/api/fusion/v1 for either format and let TRAE append the matching path. FUMO does not support Anthropic Messages; do not select that format or use a /messages endpoint.
3. Configure advanced settings
Use the following context, capability, and effort settings for Fusion Max 2.0. Select the API format before configuring effort, and keep the protocol and URL consistent.
| Setting | Recommended value |
|---|---|
| Model family | Default |
| Context window: input | 373000 |
| Context window: output | 131072 |
| Image input | Enabled |
| Thinking mode: reasoning effort parameter | Keep the client default (Chat Completions: reasoning_effort) |
| Thinking mode: supported effort values | low, medium, high, xhigh, max |
| Tool calling rounds | Keep the client default |
| Temperature / Top P / Top K | Leave blank to use service defaults |
Both FUMO protocols support high. In API requests, Chat Completions uses the top-level reasoning_effort field, while Responses uses nested reasoning.effort. This is a difference in request format, not a restriction to default effort in Responses. Match the client configuration to the selected API format. See the Responses API for the request fields.
Fusion is a separate model service. Choose Default for the model family; presets for GPT, Claude, or other families are unnecessary. 373000 is the recommended client input window, rather than the model's maximum context. Actual output may also be shorter than the configured per-response limit.
Enter the five Supported effort values one at a time, pressing Enter after each value so that each becomes a separate tag. This declares the available levels; select high in the conversation after saving. The built-in Default option does not need to be added manually; it uses FUMO's server default of medium. Fusion Max 2.0 does not support disabling thinking, so do not select Off.
4. Save and select the model
- Click Add model. TRAE calls the endpoint to check connectivity. If it fails, troubleshoot using the returned status code and error details.
- Confirm the custom model is enabled in model management.
- Return to the conversation and select your Fusion model under Custom models in the model menu. Confirm that the composer shows your Fusion model's name. Auto Mode uses only TRAE's built-in models.
- Open the model menu again, hover over your Fusion model, and set Reasoning effort to High (high) in the model information panel. This level is recommended for both protocols; declaring the five supported values does not automatically select
high. - First verify an ordinary conversation, such as “Reply with only the result of 17 × 23.” Then test file reading and tool calling in a test project.
Adding a model and completing an Agent workflow are separate checks. Begin with a local test project, then try more complex tasks.
5. Optional: add Fusion Code 1.0
Repeat the setup with the same endpoint and change the model ID to:
fusion-code-1.0
Confirm your key is authorized for this model. Code 1.0 does not support adjustable reasoning effort, so leave the effort parameter name and values unconfigured. Use the same endpoint and input / output settings.
Troubleshooting
Model settings are missing or messages cannot be sent
Sign in to your TRAE account and update to the latest desktop client. This guide follows the China edition's custom model workflow. If enterprise administrator policies restrict the feature, contact your administrator.
Adding a model returns 404
Check the API format, Full URL, and endpoint together. With Full URL on, use the matching complete URL: https://capi.fumolab.ai/api/fusion/v1/chat/completions or https://capi.fumolab.ai/api/fusion/v1/responses. With it off, enter only https://capi.fumolab.ai/api/fusion/v1 and let the client append the path. Avoid duplicate paths or a protocol mismatch.
The endpoint returns 401 or 403
Check the complete key, its access channel, and model authorization. See First-request troubleshooting.
TRAE still uses a built-in model
Select the Fusion model in the current conversation's model menu to leave Auto Mode, and confirm the configuration is enabled in model management.
High and other effort options are missing
First confirm that the API format matches the endpoint. Edit the custom model, keep the effort parameter configuration for that format, and add all five English values, pressing Enter after each one. Save, then hover over your Fusion model in the model menu to view its effort options. Display labels may be translated; enter the original English values in the configuration, and select High (high) as recommended.
