Contents
Paste this into Claude Code or Codex:
Add SolCard Inference to my models: solcardinference.ai/install
That’s the setup. Your agent reads the page, follows the steps for itself, and adds SolCard Inference’s models next to the ones you already have.
What happens
solcardinference.ai/install isn’t a script. It’s instructions written for an agent to read, and it sends each agent to its own step-by-step runbook. Your agent asks you once before it changes anything, then:
- Downloads a small open-source proxy, CLIProxyAPI, checks its checksum, and runs it as a background service that only your computer can reach.
- Asks you to approve SolCard Inference in your browser. Sign up or sign in, check the code matches, and click Approve. The agent collects your key from that approval; you never paste it.
- Points itself at the proxy. Requests for your own models go on to Anthropic or OpenAI with your own login, unchanged. Requests for ours go to SolCard Inference.
- Checks both work before it tells you it’s done.
Everything it adds lives in ~/.solcard-inference, the agent’s own settings, and one line in your shell profile. Your other settings stay as they are.
- Claude Code keeps your Claude models, on your own Claude plan, and lists SolCard Inference’s models below them in
/model. - Codex (the CLI, or Codex in the ChatGPT app) keeps your ChatGPT models, on your own ChatGPT plan, and lists ours next to them in the model picker.
The runbooks cover macOS, Linux and Windows. The Claude desktop app’s Code tab and Cowork only run Claude’s own models, so set it up from Claude Code in a terminal.
Restart, then pick a model
- Claude Code: type
/exit, open a new terminal, and runclaude./modelnow lists the SolCard Inference models under yours. - Codex: quit, run
codex app-server daemon restart, and startcodexagain. In the ChatGPT app, quit and reopen it. The model picker lists the SolCard Inference models.
Switch whenever you like. Your own models still bill your own plan; ours bill your SolCard Inference balance.
Which of our models
solcard-inference-fastfor the everyday turns: reading files, small edits, running tests.solcard-inference-largefor the hard ones: planning a change, a bug you can’t find. It always reasons, and costs about three times as much as Fast.
Fast, Large or Base has the details.
What a session costs
A coding agent sends the whole conversation every turn, so a long session sends a lot of input. Most of it is cached, and cached input costs a tenth of the normal rate:
| Model | Input | Cached input | Output |
|---|---|---|---|
| Fast | $1.25 | $0.125 | $2.50 |
| Large | $3.75 | $0.375 | $6.25 |
USD per 1M tokens.
Say a turn on Fast resends 80,000 tokens of conversation, 75,000 of them cached, and writes 1,500 tokens:
- 75,000 cached × $0.125 per million = $0.0094
- 5,000 new × $1.25 per million = $0.0063
- 1,500 output × $2.50 per million = $0.0038
- About $0.02 a turn
Your usage page shows every request: model, tokens in and out, and cost.
If you code every day, look at the plans. Starter is $20 a month for 20% more usage than the same money buys as you go, and raises your limit from 60 to 120 requests a minute. Agents fire requests fast, so the higher limits matter. API keys use your credits first, then your plan; see how billing works.
Switching back
Ask the agent:
Remove SolCard Inference from my models.
It takes our models out of its settings, stops the proxy, deletes its files, and leaves your own models as they were. Your SolCard Inference key still exists; delete it on the dashboard’s API keys page if you’re done with it.