TOLVYN is the financial control plane for AI infrastructure — the system of record for every API call you route through it — across OpenAI, Anthropic, Google, and DeepSeek.
Live since March 2026 · Open-source pricing database · MIT and Apache-2.0 SDKs
The problem
Engineering spent $47K on Claude last month. Which team? Which feature? Which customer? Nobody can answer the CFO's question.
Your AI SDR, support bot, and coding agent each generate thousands of requests daily. None of them appear on one ledger.
When auditors ask for AI usage records, you have provider invoices and Slack screenshots. That's not a financial record.
How it works
Register your provider key with TOLVYN once, point your SDK's base URL at the proxy, and swap your provider key for a TOLVYN key. No SDK refactor.
What makes TOLVYN different
Every AI request TOLVYN meters is hash-chained and HMAC-signed. Cryptographic proof, not logs. Your CFO or auditor can re-verify the hash chain themselves with the script shipped in every evidence package — the HMAC is keyed on a secret TOLVYN holds, so it detects tampering by anyone without that key.
Track AI cost per your customer. Know your AI COGS per customer. The unit economics SaaS founders are missing.
Upload your OpenAI / Anthropic / Google invoice. TOLVYN shows the gap. Shadow AI surfaces in minutes.
Per-agent budgets. Cost anomaly alerts — one call over 10× the recent average. Kill switch. Every AI agent on one ledger.
Nightly analysis identifies migration opportunities. Switch to a cheaper model where your actual usage patterns fit.
When OpenAI changes prices, customers using affected models get an alert projecting the monthly impact from their last 30 days of usage.
Five-minute integration
# Before
from openai import OpenAI
client = OpenAI(api_key="sk-...")
# After
from tolvyn import OpenAI
client = OpenAI(
tolvyn_api_key="tlv_live_...",
openai_api_key="sk-...", # REQUIRED for fail-open; no key, no fallback
team="engineering",
service="chatbot-api"
)
# Everything else identical// Before
import OpenAI from 'openai'
const client = new OpenAI({ apiKey: 'sk-...' })
// After
import { OpenAI } from 'tolvyn'
const client = new OpenAI({
tolvynApiKey: 'tlv_live_...',
openAIApiKey: 'sk-...', // REQUIRED for fail-open; no key, no fallback
team: 'engineering',
service: 'chatbot-api'
})# Before
curl https://api.openai.com/v1/chat/completions \
-H "Authorization: Bearer sk-..."
# After (change the URL, keep everything else)
curl https://proxy.tolvyn.io/v1/proxy/openai/v1/chat/completions \
-H "Authorization: Bearer tlv_live_..." \
-H "X-Tolvyn-Team: engineering" \
-H "X-Tolvyn-Service: chatbot-api"pip install tolvyn · npm install tolvyn · go get github.com/tolvyn/tolvyn-go
Works with OpenAI, Anthropic, Google, and DeepSeek. If TOLVYN is unreachable, SDK mode automatically retries against the provider directly.
Header-compatible. We read your existing helicone-property-* headers automatically.
Anonymized cost benchmarks aggregated from real production AI traffic. A figure is published only when at least three organisations contribute to it, so the index shows what clears that floor — and nothing else.
View the data →