Run Aider on SolCard Inference

Aider talks to any OpenAI-compatible API: set OPENAI_API_BASE to https://api.solcardinference.ai/v1, set OPENAI_API_KEY to your SolCard Inference key, and start Aider with --model openai/solcard-inference-fast. The openai/ prefix tells Aider to use the OpenAI-compatible route. We ran exactly this on November 1, 2026, and Aider fixed a bug and committed it.

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. See how billing works.
• Aider. We tested v0.86.2.

The commands

On macOS or Linux:

export OPENAI_API_BASE=https://api.solcardinference.ai/v1
export OPENAI_API_KEY=sk-si-...
aider --model openai/solcard-inference-fast

On Windows, Aider’s docs use setx for the two variables, then a new shell:

setx OPENAI_API_BASE https://api.solcardinference.ai/v1
setx OPENAI_API_KEY sk-si-...

When it starts, Aider prints the model and edit format:

Model: openai/solcard-inference-fast with whole edit format

If you already use OPENAI_API_KEY for OpenAI, pass ours on the command line instead with --openai-api-base and --openai-api-key, so the two don’t collide.

The unknown-model warning

Aider keeps a list of models it knows. Ours aren’t on it, so at startup it warns that it doesn’t know the context window size or the costs, and falls back to defaults. Aider’s warnings page says you can usually ignore it.

To skip it, start Aider with:

aider --model openai/solcard-inference-fast --no-show-model-warnings

That’s how we ran it. Aider’s model settings docs also describe a .aider.model.metadata.json file, in your home directory or repo root, that tells Aider a model’s limits and prices. For Fast it would look like this; we haven’t tested it:

{
  "openai/solcard-inference-fast": {
    "max_input_tokens": 256000,
    "max_output_tokens": 32768,
    "input_cost_per_token": 0.00000125,
    "output_cost_per_token": 0.0000025,
    "litellm_provider": "openai",
    "mode": "chat"
  }
}

Which model to pick

• openai/solcard-inference-fast for everyday edits. Start here.
• openai/solcard-inference-large for changes that need planning, or a bug Fast can’t find. It always reasons and costs about three times as much.
• openai/solcard-inference-base for a second opinion.

Every model writes at most 32,768 tokens per request. Fast, Large or Base compares them.

Check it works

Ask something small in the chat:

> What does this repo do, in one sentence?

After each reply Aider reports tokens, for example Tokens: 739 sent, 43 received. It shows no cost, because it doesn’t know our prices. Your usage page does.

To check the key on its own, list the models:

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

You should see solcard-inference-fast, solcard-inference-large and solcard-inference-base.

What an edit costs

USD per 1M tokens, from the pricing page:

In whole edit format, Aider writes back the entire file for each edit, so output grows with the size of the files it changes. Say a request sends 12,000 tokens (the repo map, two files and your chat) and gets back a 2,000-token file on Fast:

• 12,000 input × $1.25 per million = $0.015
• 2,000 output × $2.50 per million = $0.005
• About $0.02 an edit

Any part of the prompt served from cache bills at a tenth of the input rate, so repeat turns on the same files cost less. On Large the same edit is about $0.06, plus its reasoning.

Troubleshooting

These are the API’s responses; Aider may word them differently.

• 404 Cannot POST /chat/completions: OPENAI_API_BASE is missing /v1.
• 401 invalid_api_key: the key is wrong or revoked, or another OPENAI_API_KEY is set in your shell. Run echo $OPENAI_API_KEY | cut -c1-6 and check it prints sk-si-.
• 400 model_not_found: check the id after openai/. Aider may also print “Did you mean one of these?” for a typo.
• 400 unsupported_content: you added an image. Our models take text only.

The errors page lists every code.

FAQ

Why do I need the openai/ prefix? Aider’s docs say to prefix the model name with openai/ for any OpenAI-compatible endpoint.

Do I have to type the flags every time? No. Aider reads most flags from environment variables too, such as AIDER_OPENAI_API_BASE and AIDER_SHOW_MODEL_WARNINGS; see its options reference.

Are my prompts or code stored? No. Prompts and responses aren’t stored or logged, on any model. See keeps nothing and what makes an LLM API private.