Guide
Using Usefully in Cursor and Continue
Coding assistants that accept an OpenAI-compatible endpoint can use any Usefully model. Setup for Cursor and Continue, and tips for choosing models for code.
· 3 min read · Usefully team
AI coding assistants are some of the heaviest model users there are: every chat, edit and explanation is a request with a large context. Pointing them at Usefully gives you a choice of models from Anthropic, OpenAI and Google on one key, automatic failover when a provider has a bad day, and a clear record of what your coding sessions cost.
Cursor
Cursor lets you supply your own OpenAI API key and override the OpenAI base URL in its model settings. Menu names move between versions, but the steps are:
- 1Open Cursor's settings and go to Models.
- 2Enter your Usefully key as the OpenAI API key.
- 3Turn on Override OpenAI Base URL and set it to
https://api.usefully.info/v1. - 4Add a custom model by its Usefully ID, for example
anthropic/claude-sonnet-5-5, and enable it. - 5Pick that model in the chat panel and send a test message.
Cursor's own built-in features may continue to use Cursor's models; the custom model is used where you select it.
Continue
Continue (for VS Code and JetBrains) configures models in a config file. Use the openai provider with Usefully's base URL:
models:
- name: Claude Sonnet (Usefully)
provider: openai
model: anthropic/claude-sonnet-5-5
apiBase: https://api.usefully.info/v1
apiKey: <your Usefully key>
roles: [chat, edit, apply]
- name: Fast model (Usefully)
provider: openai
model: anthropic/claude-haiku-5-5
apiBase: https://api.usefully.info/v1
apiKey: <your Usefully key>
roles: [chat, autocomplete]Continue's config format has changed over time; if your version uses config.json, the same three fields apply: provider: "openai", apiBase and apiKey.
Other tools
Many tools built on the OpenAI SDK read the OPENAI_BASE_URL and OPENAI_API_KEY environment variables. Set them to Usefully's base URL and your key, and those tools switch over with no configuration screen at all.
export OPENAI_BASE_URL="https://api.usefully.info/v1"
export OPENAI_API_KEY="<your Usefully key>"Choosing models for code
- Chat and multi-file edits: a strong model with a large context window. Check context sizes on the models page; large repositories fill a window quickly.
- Autocomplete and quick questions: a smaller, faster model. Latency matters more than depth for completions.
- Try two. Coding styles differ between model families. Spend a day with each on real work before settling.
Keeping costs visible
- Use a dedicated key per developer for editor use, named after the tool (
cursor-sam). Spend by key then shows what each setup costs. - Set a daily spend limit on those keys. A runaway agent loop in an editor stops at the limit instead of draining the shared balance.
- Check Logs after a heavy session. Coding assistants send large inputs, and input tokens are usually where the cost is.
Troubleshooting
- 401: the key is wrong, revoked, or the tool is still sending a different provider's key.
- 404 or model not found: the model ID must be the full Usefully ID, such as
anthropic/claude-sonnet-5-5. - Errors about a provider not being available: that model's provider may not be connected; the error suggests another model to try.
Why route an editor through a gateway
The main benefit is choice without extra accounts: you can try a model from a different provider in your editor by changing one setting, with no new sign-up or billing setup. The second is visibility. Editor usage is hard to estimate in advance, and per-request costs in Logs show what a day of coding with each model actually costs.
Full model IDs are on the models page, and the quickstart covers SDK setup. Stuck? support@usefully.info.