Pricing.
Same surface across CLI, API, SDK, MCP, and Agent SDK. Three honest numbers: requests per minute, requests per month, SF tokens per month. No tracked-day gates, no “weighted” multipliers buried in fine print.
Build against the read surface before you pay.
- rpm
- 60
- req/mo
- 1K
- sf tokens
- 100K
- · Every public read endpoint
- · CLI, SDK, MCP, and Agent SDK
- · Community support
For small products, research workflows, and real API traffic.
- rpm
- 300
- req/mo
- 50K
- sf tokens
- 2M
- · AI-assisted endpoints included
- · Stripe-managed billing
- · Email support
For automated agents, higher concurrency, and heavier LLM use.
- rpm
- 1,000
- req/mo
- 500K
- sf tokens
- 20M
- · LLM proxy access for agents
- · Higher request and SF-token ceilings
- · Priority email support
For desks that need private data work or custom limits.
- rpm
- —
- req/mo
- —
- sf tokens
- —
- · Custom RPM and monthly ceilings
- · Private feeds and integration work
- · Contracted support and security review
Burst ceiling. Returns 429 with Retry-After: 60. Same window for every endpoint.
Every API call counts 1. LLM endpoints do not get a hidden multiplier in this counter; SF-paid model compute is metered separately.
One SF token = one micro-dollar ($0.000001) of LLM compute we paid for on your behalf. Lower-cost models stretch further. Read endpoints cost zero. BYOK paths through the SDK do not count.
FAQ.
What happens at the wall?
429 with an X-SF-Upgrade-Url header. Block reasons: rpm, hard_cap, or sf_tokens. Subscribe and you're back online instantly. Counters don't reset on upgrade.
What about the Agent SDK's OpenRouter calls?
BYOK — bring your own OpenRouter key. You pay OpenRouter directly. Those tokens don't count against your SF-tokens budget. Only LLM endpoints that route through SF infrastructure consume SF tokens.
What about MCP?
MCP tools that hit SF endpoints count against your monthly requests. Tools that hit Kalshi or Polymarket directly don't — we're not in the path, so we don't meter. Authenticate the MCP server with your sf_live_* key.
How is an SF token defined?
One SF token equals one micro-dollar ($0.000001) of LLM compute. We read OpenRouter's reported usage.total_cost for the call, multiply by 1M, and that is your SF-token charge. When a provider does not return cost, we estimate from the server-side model pricing table and keep the request in your audit trail.
Can I downgrade later?
Yes. Use the Stripe Customer Portal from your dashboard. Downgrade takes effect at the end of the current billing period.
Crawlers and training bots?
Crawler UAs (Googlebot, ClaudeBot, GPTBot, Perplexity, etc.) get a separate generous bucket. AI-training bots are welcome — they never count against any user's quota.
Custom pipelines.
Everything above ships publicly. If you need indicators tuned to your strategy, private feeds with lower latency, dedicated cross-venue routing, or direct integration into your data stack — we build it.
For quant research desks, market makers, and prop teams operating at scale.
Talk to the team →