CuriousDevs is an AI infrastructure company. We ship the products that AI teams desperately need — starting with LLM cost attribution and enterprise AI agent governance.
Open source friendly · Design partners welcome · No sales call required
# Install TokenFin SDK
$ npm install @curiousdevs/tokenfin
# Wrap any LLM call — one line
const res = await track(openai.chat({...}), {
team: 'ml-infra', feature: 'search'
})
✓ Cost attributed: $0.0034 → ml-infra/search
✓ Monthly projection: $2,847
✓ Anomaly alert: spend +34% vs last week
$ █
The problem we're solving
Every AI team we talk to has the same four problems. We built TokenFin and AgentOS to solve all of them.
Your OpenAI invoice says $12,400. It doesn't say which team spent $7,000 of that — or on which feature. Finance asks. Engineering shrugs. Someone opens a spreadsheet. Nobody wins.
You picked a model in week 1. Now it's month 6, your product scaled, and nobody benchmarked whether you're still on the right model. Latency crept up. Quality crept down. Costs crept sideways.
Your AI agents run in production with no policy guardrails, no decision logs, and no kill switch. That's fine — until the EU AI Act auditor arrives or a model hallucinates at scale.
The EU AI Act General-Purpose AI provisions apply from August 2025. The High-Risk AI rules kick in August 2026. Most engineering teams don't know where to start. Panic is not a compliance strategy.
Sound familiar? That's why we built this.
See how TokenFin fixes this →Our Products
We build the AI infrastructure that should already exist. TokenFin solves the LLM cost problem. AgentOS solves the agent governance problem.
TokenFin · Private Beta
LLM cost attribution and FinOps for teams spending $10K+/month on AI APIs. Attribute spend by team, feature, and user. Catch waste before the CFO does — and prove ROI before budget reviews.
AgentOS · Coming 2026
Identity, permissions, observability, governance, and security for AI agents — unified in one platform. Because 79% of enterprises have deployed agents, but only 11% trust them enough for production.
CuriousLabs · Open Source
SDKs, frameworks, and experiments — open sourced for the AI engineering community. MIT licensed. No strings. Because the best way to build trust is to build in public.
Platform
From raw token counts to board-level ROI — TokenFin connects every dollar to the decision that spent it.
Every single LLM call — attributed to the exact feature, user, team, and model that generated it. No more guessing from a flat API invoice.
Burn-rate alerts fire before your bill does. Set team-level and feature-level thresholds.
Hard caps and soft warnings per team or feature. Never get a surprise bill again.
Every call, every decision, every policy event — searchable and export-ready for your audit team.
Stop relying on provider leaderboards. Compare GPT-4o, Claude 3.5 Sonnet, Mistral Large, and Gemini 1.5 Pro on your actual prompts — cost-per-token, latency p95, and quality scores side-by-side.
How it works
One npm install. Works in Node, Python (pip), or any runtime. No infrastructure changes required.
npm install @tokenfin/sdkAdd a single track() wrapper around any OpenAI, Anthropic, or Bedrock call. That's it.
tf.track("chat-feature", () => openai.chat(...))Log in to tokenfin.curiousdevs.com and see real-time attribution data, model comparisons, and cost trends.
tokenfin.curiousdevs.com/dashboardSet alerts, apply budget guardrails, swap models with confidence, and cut your AI bill by 20–40%.
avg_savings = "20-40% in week one"Integrations
One SDK. Zero rip-and-replace. TokenFin wraps your existing providers — instrument in minutes, not months.
+ HuggingFace, Cohere, Azure OpenAI, Vertex AI — and more added every sprint
Why teams choose us
Not someday. Not after a six-week implementation. TokenFin delivers signal on day one — or we haven't done our job.
Teams using TokenFin find and eliminate LLM waste within the first week. Real attribution reveals which features drive cost — so you optimize what matters, not what's visible.
One npm install. One SDK call. Your dashboard starts populating immediately — no data pipeline, no schema changes, no DevOps ticket. You have signal before the meeting starts.
Wrap any LLM call — OpenAI, Anthropic, Bedrock, Mistral — with a single track() call. No refactoring. No vendor lock-in. Works with whatever you're already running.
Stop guessing at spreadsheets. TokenFin gives you chargeback reports by team, feature, and model — in seconds. Your CFO gets the data they need. You get budget approval.
AgentOS generates conformity documentation, risk registers, and audit trails automatically. Don't scramble in Q4 — ship compliant from the start.
The average AI team wastes over $200K/year in unattributed, un-optimized LLM spend. TokenFin makes that visible. Once visible, it gets fixed — fast.
Pricing
Start free. No credit card. Upgrade when you need more.
Starter
Free forever
Perfect for solo developers and small experiments.
Pro
per month
For growing teams that need attribution, alerts, and multi-model analytics.
Enterprise
Contact us
For large teams needing SSO, SLAs, VPC deployment, and custom retention.
All plans include: AES-256 encryption · HTTPS/TLS 1.3 · GDPR + DPDP Act compliant · 99.9% uptime
FAQ
These are actual questions from developers and engineering managers evaluating TokenFin.
Have a question not listed here?
Ask us directly →Newsletter
Monthly deep-dives on LLM FinOps, model benchmarks, governance updates, and what we're building. No spam. Unsubscribe anytime.
We respect your privacy. Unsubscribe at any time.
Join teams already saving 20–40% on their LLM bills. Free to start. No credit card required.
Free plan forever · No credit card · Setup in <15 min