SandBase

SandBase is a single access layer between your coding agents and the things they need to actually do work. Rather than holding separate keys and adapters for every provider, you run one setup command and your local tools, Codex, Claude, Cursor or Kiro, gain access to a catalogue of 1,195 models alongside 1,105 real-world APIs covering web search, scraping, data extraction and SaaS systems. The same key works whether you’re calling a capability from your own product or building a reusable agent.

The catalogue is the part worth exploring. Every card shows its price next to a working request, so a Claude Opus call, an Exa search at $0.007, or an image edit at $0.06 are all costed before you commit to anything. Beyond raw models and APIs there are ready-made agents and packaged skills you can test, clone or install, which gives you a starting point rather than a blank workflow.

For multi-step work, the Build Agent side lets you define a job, give it tools, watch a single run before anything depends on it, then publish it for other apps to call or put it on a schedule. Every execution is kept as a run with its input, output, trace and status, and Activities reports what each one cost, which matters when agents are spending money on your behalf. The CLI is open source and tracks exactly what it installed, so unregistering rolls back its own changes without touching settings you manage yourself. New accounts get $1 in credits to try things, then it’s pay-as-you-go with team and enterprise options for volume, SSO and private deployment.