[Verified] One key, 2,000 models: what SandBase CLI's free tier actually covers

What SandBase CLI does—and what its free tier doesn’t tell you

SandBase CLI is open-source software that bridges your AI coding tools—Claude Code, Cursor, Codex, ChatGPT, and 21 others—to 2,000+ AI models and APIs through a single API key. The pitch is clear: stop juggling different subscriptions; route everything through one interface without re-entering credentials.

The catch, which the official pricing page omits: your “starter free credits” have no published value. Once you hit that limit, you pay usage rates for whatever models you call. SandBase doesn’t publish its own platform fee on top of those rates either.

The model is genuinely simple: open-source CLI (free download, Apache-2.0 license), one-time starter credits (amount unlisted), then pay-as-you-go for model calls. No monthly subscription, no seat licenses. But the transparency gap starts immediately. A developer planning the first week needs to know how many credits they’re getting, and SandBase won’t say.

The architecture: one key, 25 clients, 2,000+ destinations

SandBase CLI works as a local MCP server—that’s the standard protocol Claude Code and Cursor use to talk to external tools. Install the CLI on your machine, authenticate once with your SandBase API key, and point your editor to it. From there, switch models inside Claude Code or Cursor without managing separate logins for each.

The 2,000+ models cover the obvious (Claude, GPT-4, Gemini) plus open-source services, specialized inference APIs, and custom endpoints. You can track which model handled each request and how many tokens it used through SandBase’s activity log. OAuth login and rollback (reverting calls if they go wrong) are built in—useful if a stray request burns credits.

The platform doesn’t watermark output or restrict commercial use. You can use it for production services, internal tools, or client work without special licensing. The open-source CLI part (Apache-2.0, on GitHub at sandbaseai/cli) means you can self-host it offline if supply-chain risk matters to you.

🔍 AMPM exclusive check

What does the vendor page leave unsaid? We checked https://sandbase.ai/pricing directly on 2026-09-22:

• Starter credit amount: The page says “Start: Free credits” with no number—no mention of how many API calls, tokens, or days those cover. It’s blank on the official site.
• Platform fee: SandBase takes a cut on top of each model’s base price. The rate is not posted anywhere. You cannot say whether the markup is 5% or 20%; they don’t disclose it.
• Pay-as-you-go detail: The page confirms usage-based billing with no subscription, but doesn’t itemize what “usage” means across model types (tokens for text, per-call for streaming, per-image for vision, etc.).

Everything below matches the official page. The gaps are omissions, not contradictions.

☀️ AMO, the budget-minded cat

From a cost angle, SandBase is transparent about one thing: you don’t pay until you use something. But that “something” lands on two blind spots AMO can’t plan around.

First, how many free credits do you actually get? Without that number, a developer can’t answer “Can I try this free for a week?” or “Will my side project stay free?” The pricing page doesn’t say. Your first discovery is empirical: call models until the counter hits zero. You might have a month of light tinkering or burn through credits in one afternoon. You find out by usage, not by reading.

Second, what’s the markup on top of the model’s own price? If Claude Opus costs US$15 per million input tokens through Anthropic’s API, does it cost the same through SandBase or US$18? The platform overhead is real—servers, infrastructure, profit margin—but it’s invisible. You have to sign up, create a workspace, and start logging calls to see line-item costs. For a budget-conscious developer comparing against calling APIs directly, SandBase is opaque until you’re inside it.

For small use cases—switching between Claude and GPT inside one editor without managing three subscription dashboards—the free starter credits probably cover a few days of light work. If you already pay for Claude and GPT subscriptions independently, the extra cost of routing through SandBase depends on that hidden platform fee. It could be 2% or material; the data isn’t public. The honest answer: SandBase is “try it free” until the mystery credit budget runs out, then pay an unknown markup on model prices. Whether that’s cheaper than APIs à la carte depends on your usage and SandBase’s fee—both unknown upfront.

🌙 PIMI, the performance-minded cat

PIMI cares whether SandBase actually solves the problem: can I really switch between Claude Code, Cursor, and ChatGPT inside one editor without re-entering credentials each time?

The architecture supports it. SandBase runs as a local MCP server, the standard protocol both Claude Code and Cursor use to communicate with external tools. Point your editor config at the SandBase endpoint, and requests flow through. A single API call to SandBase routes to whichever model you’ve selected in the UI. One key handles the whole thing.

The feature set is solid for production work. 2,000+ models covers the major services, niche inference platforms, and custom APIs. The activity log that tracks every call, tokens burned, and runtime is what production teams actually need—you can audit which model ran what, how long it took, and what it cost. OAuth and rollback prevent you from getting stuck if a call fails or runs the wrong way. The open-source CLI (self-hostable, auditable on GitHub) means you can run it offline or inspect the code for security concerns. For teams worried about supply-chain risk or third-party requests, that’s a real option.

What’s missing from the public data: performance benchmarks (does SandBase add latency?), uptime SLAs (is it production-grade?), and customer case studies. The open-source part is transparent, but the hosted SandBase platform service doesn’t publish reliability metrics or latency curves. PIMI would want to pilot it before trusting a production workflow to it—test whether the round-trip through SandBase, OAuth, and routing introduces unacceptable delays or failures.

For teams that actually context-switch between models inside their editor, SandBase is smart. A Cursor user with Claude, ChatGPT, and Gemini subscriptions can spend 10 minutes reconfiguring editor settings daily, or 10 minutes setting up SandBase once. The payoff is real, assuming the platform stays reliable and the feature set doesn’t regress.

Verdict: who should use it, who should skip it

Use SandBase CLI if you:

• Work in multiple AI editors (Claude Code, Cursor, ChatGPT) and want to switch models inside one interface without re-entering credentials.
• Want an audit trail of which model handled which request and token costs.
• Can tolerate an unknown platform fee on top of model pricing because the convenience is worth it.
• Can live with a surprise if your starter free credits are smaller than expected.
• Trust open-source software over closed platforms.

Skip SandBase CLI if you:

• Use only one AI model (Claude-only or GPT-only workflow) and don’t switch.
• Need to know exact costs before trying anything—the hidden credit amount and unlisted platform fee will block you.
• Run ultra-low-latency production services where an extra network hop through SandBase could matter.
• Require SLA and support contracts; Team/Enterprise plans exist but are custom-quoted and non-transparent.

The hidden pricing isn’t a dealbreaker—it’s a gotcha that forces you to sign up before “free” makes sense. But the feature set is solid, and for developers who genuinely switch models often, SandBase saves time and credentials.

Written: 2026-09-22 Price last verified: 2026-09-22 Tool last health-checked: 2026-09-22

🔗 Related on AMPM

• SandBase CLI pricing & free tier

📎 Sources

All prices and quotas in this article come from AMPM’s verified dataset for SandBase CLI (last verified 2026-09-22); primary sources:

• https://sandbase.ai/pricing
• https://sandbase.ai

🐾 Meet the cats: AMO and PIMI

AMPM is a Taiwan-based AI-tool pricing watchdog. Our motto: ask before you subscribe. Two cats argue every tool from two sides:

• ☀️ AMO — the budget-minded cat. Always asks: is the free tier enough, and is paying actually worth it?
• 🌙 PIMI — the performance-minded cat. Cares about whether it works well, fast, and gets your job done.

When they are done arguing, you know whether to pay. Every number is checked by us, with the verification date at the end.