Use SolCard Inference in Cursor with a custom base URL

To use SolCard Inference in Cursor, paste your sk-si- key into the OpenAI API key field under Cursor Settings > Models, turn on Override OpenAI Base URL with https://api.solcardinference.ai/v1, and add our model IDs as custom models. Cursor’s docs list OpenAI, Anthropic, Google, Azure and AWS Bedrock as key providers, with no separate entry for an OpenAI-compatible API, so you borrow the OpenAI slot. We haven’t run Cursor against our API ourselves; every Cursor step below comes from Cursor’s docs or posts on its forum, mostly from Cursor staff.

Before you start, read the privacy section below. Cursor’s own servers stay in the path.

What you need

• A SolCard Inference API key. Make one on the dashboard’s API keys page. Keys start with sk-si- and are shown once.
• Some balance or a plan. One prepaid USD balance pays for every key and model; see how billing works.
• A Cursor plan that allows your own keys. In a forum reply from October 2026, Cursor staff said using your own API key works on Teams plans, not only Enterprise. On Enterprise, an admin can restrict models and keys.

The settings

The steps, from that same staff reply:

1. Open Cursor Settings and go to Models.
2. Scroll to the bottom and expand API Keys.
3. Paste your sk-si- key into the OpenAI API key field and turn on Use OpenAI API Key.
4. Turn on Override OpenAI Base URL and enter https://api.solcardinference.ai/v1.
5. Add a custom model whose name exactly matches one of our model IDs.
6. Restart Cursor, open a new chat and pick the model.

The model names to add:

solcard-inference-fast
solcard-inference-large
solcard-inference-base

Pasting the key and turning on the toggle are two separate steps. In an August 2026 reply, Cursor staff said “Override Base URL only applies while the key is active.” If the key toggle is off, the base URL override doesn’t apply.

Enter the base URL exactly as above, ending in /v1. Cursor staff say the override URL usually ends in /v1, and ours does.

Agent mode or Ask mode

Cursor’s Agent mode doesn’t send the same request format as Ask mode. In a March 2026 reply, Cursor staff wrote that “in Agent mode, Cursor sends requests in the OpenAI Responses API format,” while “Ask mode uses the standard Chat Completions format.”

That matters less here than for most custom endpoints, because our API serves both OpenAI Chat Completions and OpenAI Responses at https://api.solcardinference.ai/v1. Since we haven’t run Cursor against it, here’s the fallback anyway: if Agent mode errors and Ask mode works, the request format is the likely cause. In a follow-up in the same thread, the same staff member suggests checking which path Cursor actually calls.

The override covers more than your models

Override OpenAI Base URL isn’t scoped to one model. A forum guide from September 2026, written by someone who runs an API gateway and not by Cursor staff, says that while it’s on, “OpenAI-family requests currently go through that custom endpoint.” So requests for Cursor’s own OpenAI models can end up sent to us, and our API answers a model ID it doesn’t know with model_not_found.

The fix is to switch the override off when you want Cursor’s built-in models back, and on again for ours. It’s a manual toggle every time.

What still goes through Cursor

Your requests don’t go straight from your editor to us. Cursor’s API keys docs say “all requests are routed through Cursor’s servers for final prompt building.” In a May 2026 reply, Cursor staff called a full bypass “an architectural limitation,” adding that “even with a custom endpoint the request still goes through our backend.”

Cursor’s Zero Data Retention policy doesn’t cover this setup. The same docs page says: “Cursor’s Zero Data Retention policy does not apply when you use your own API keys.” It also says your key “is not persisted after the request completes.”

What that means for you:

• On our side, prompts and responses are never logged or stored, on any model. Keeps nothing follows one request through.
• On Cursor’s side, its zero-retention promise doesn’t apply to these requests, so read Cursor’s own terms before you send code you can’t share. Can you paste client code into an AI tool? has a checklist for that call.

If keeping a third party out of the path is the point, an agent that calls our API directly is a better fit: Cline in VS Code, or Zed as an editor.

What it won’t do

• No Tab completion on our models. Cursor’s docs say “Custom API keys only work with chat models” and that Tab completion keeps using Cursor’s built-in models.
• Text only. Our models take no images, so a pasted screenshot is rejected with unsupported_content.
• 32,768 output tokens per request, on every model. A long rewrite can need more than one turn.
• On Teams and Enterprise, Cursor may still charge. Cursor’s docs say eligible requests made with your own key carry its Token Rate of $0.25 per million tokens, on top of what we charge. Individual plans don’t draw from Cursor’s included usage for these requests.

Which model and what a turn costs

Start on solcard-inference-fast and switch to solcard-inference-large in the model picker when a change needs planning; Large always reasons and costs three times as much per input token. Fast, Large or Base compares the three.

Cursor builds the prompt on its servers, so how much of it hits our cache is up to how Cursor assembles it. Take an agent turn that sends 60,000 tokens and writes 2,000. On Fast, it’s $0.080 if nothing is cached and about $0.024 if 50,000 of those tokens are cache hits. On Large, the same turn is $0.238 uncached or about $0.069 with those cache hits, plus its reasoning tokens.

Check it works

Check the key and the base URL from a terminal first:

curl https://api.solcardinference.ai/v1/models \
  -H "Authorization: Bearer sk-si-..."

You should get a list with solcard-inference-fast, solcard-inference-large and solcard-inference-base. Then open a new chat in Ask mode, pick solcard-inference-fast and ask something small.

Troubleshooting

These are the API’s error codes, from our errors page; Cursor may show them in its own words.

• 401 invalid_api_key: the key is missing, revoked or unknown. Check that Use OpenAI API Key is on.
• 400 model_not_found: the model name doesn’t match ours exactly, or a Cursor model went through the override. Turn the override off for Cursor’s models.
• 400 unsupported_content: an image reached the model. Remove it.
• 402 insufficient_balance: credits and plan can’t cover the request. Top up and retry.
• Works in Ask, fails in Agent: see the request format note above.

FAQ

Which Cursor plans can use their own key? Cursor’s docs describe how own-key requests are billed on individual plans and on Teams and Enterprise, and Cursor staff confirmed Teams works. Check your plan’s settings before you top up.

Do my prompts reach SolCard Inference directly? No. Cursor’s docs say requests are routed through Cursor’s servers for final prompt building, and its Zero Data Retention policy doesn’t apply with your own keys. We keep nothing on our side.

Can I use our Anthropic Messages endpoint in Cursor instead? Cursor’s API keys docs list Anthropic as a supported key provider but don’t document a base URL override for it. Use the OpenAI override above.

Does Tab autocomplete use our models? No. Cursor’s docs say Tab completion keeps using Cursor’s built-in models.