Know exactly
what your AI costs.

No SDK. No code changes. No $20/seat observability tax.

One dashboard for every LLM provider you use. Forecast your month-end bill, catch cost spikes before they land, and track every proxy request in real time — across OpenAI, Anthropic, Gemini, and 6 more. Set up in 90 seconds. No vendor lock-in.

Runs on Cloudflare · $5 per month

OpenAI Anthropic Google AI OpenRouter Mistral Groq xAI DeepSeek Cohere Budget alerts 342 models priced 9 providers AES-256-GCM Cloudflare-native OpenAI Anthropic Google AI OpenRouter Mistral Groq xAI DeepSeek Cohere Budget alerts 342 models priced 9 providers AES-256-GCM Cloudflare-native

02 — How it works

Three steps. No fluff.

01

Connect.

Paste your provider admin keys once. Encrypted at rest with AES-256-GCM. Never logged, never displayed.

provider openai
api_key sk-admin-••••
✓ encrypted & stored
02

Track.

Usage syncs every hour. See spend by day, model, and provider on a dashboard that doesn't hurt your eyes.

Today $42.17
03

Save.

Budget alerts fire before bill shock. The optimization engine flags wasted spend and cheaper model alternatives.

Budget alert

Monthly limit crossed at $487 / $450.

03 — Features

Everything you need to not go broke.

What others charge $20/seat for, Obol does for $5 flat. No per-seat pricing. No usage tiers. No surprises.

01

9 providers, one dashboard

OpenAI, Anthropic, Google, OpenRouter, Mistral, Groq, xAI, DeepSeek, Cohere. No tab-switching between provider consoles.

02

Per-model cost breakdown

See exactly which model is eating your budget — by spend, requests, tokens, and cache hits. Not just totals.

03

Optimization engine

Automatically spots wasteful model usage and suggests cheaper alternatives with the same capabilities. Pays for itself on day one.

04

Month-end forecast

Exponentially smoothed projection with day-of-week adjustment. Know your bill before the bill — not after.

05

Cache savings tracker

See your prompt-cache hit rate per model, savings already earned, and money still on the table. No other tool shows this.

06

Spike + drift detection

MAD catches sudden spikes. CUSUM catches the slow $2/day creep that basic threshold alerts miss entirely.

07

Budget alerts

Email and Slack notifications when spend crosses your threshold — daily, weekly, or monthly. No more bill shock.

08

CSV + JSON export

Pull your usage data for accounting, compliance, or your own dashboards. Your data, no lock-in.

09

Zero-trust security

AES-256-GCM encryption on every API key. Never logged. Never displayed. PBKDF2 passwords. No tracking cookies.

04 — Preview

A dashboard you'll actually open.

useobol.pages.dev/overview

This month

$1,247

↓ 12% vs last

Today

$42

Top model

Claude Sonnet 4.5

$487

Connections

4

OpenAI Anthropic OpenRouter

05 — LLM proxy

Metered in real time.
One-line swap.

Point any of 9 providers — OpenAI, Anthropic, Gemini, Mistral, Groq, xAI, DeepSeek, Cohere, or OpenRouter — at Obol and every call lands in your dashboard before the response stream finishes. No admin keys required. Just swap the base URL.

  • Real-time cost tracking Every response includes x-obol-cost-cents. No polling, no batch sync. Streaming requests get heartbeat keep-alive for long-running o3/thinking calls.
  • No SDK required Change the base URL. That's it. Works with any OpenAI-compatible client. No wrapper, no decorator, no agent framework.
  • Virtual keys with guardrails Per-key rate limits, monthly budget caps, provider allowlists, model filters, IP restrictions, configurable timeout. Rotate without touching your app.
  • Cross-provider fallback X-Obol-Fallback: groq,openrouter — automatic failover if the primary is down. Retries with backoff, idempotency-key support for safe retries.
client.python
before
from openai import OpenAI

client = OpenAI(
    api_key="sk-proj-...",
)
with obol
from openai import OpenAI

client = OpenAI(
    base_url="https://useobol.pages.dev/v1",  # ← add this
    api_key="obol_sk_live_...",               # ← swap this
)
Powered by Keplor — open-source LLM log aggregator ~20ms overhead

04.5 — Desktop

Your spend,
in the menu bar.

A tiny native widget that lives in your menu bar. Summon with ⌘⇧O / Ctrl+Shift+O.

  • System tray Today's spend + mood face, refreshed every 2 min.
  • Budget alerts Flashes at 80% and 100% of your cap.
  • 6 MB native binary Tauri 2. No Electron. macOS, Windows, Linux.

macOS · Windows · Linux · Tauri 2 · v0.1.0

Obol
(•_•)

On track this month.

21% of $200 budget

This month

$1,247

Today

$42

Top model

Claude Sonnet 4.5

$487

Forecast

$1,680

month end

Last 14 days

updated 2 min ago 3 connections

05 — Integrations

Plays nicely with your stack.

OpenAI

Provider

Anthropic

Provider

Google AI

Provider

OpenRouter

Provider

Mistral

Provider

Groq

Provider

xAI (Grok)

Provider

DeepSeek

Provider

Cohere

Provider

Resend

Alerts

Slack

Alerts

Cloudflare

Infra

05.5 — Personality

Pick your
dashboard's tone.

The mood meter at the top of your overview is a running commentary on your spend. Three flavours: professional for the standup, casual for most days, unhinged when you need a wake-up call.

Pro feature · Switch in Settings

Professional boring
(•_•)

You're on track this month.

Current pace: $42 / $200 budget.

Casual default
(◠‿◠)

Nice, you're cooking but not burning.

21% of budget · 14 days left.

Unhinged roast me
(ಠ益ಠ)

You spent $487 on GPT-4o this week. Do you own a yacht?

Claude Haiku would've cost $31. Just saying.

06 — Pricing

Simple. Fair. Minimal.

Razorpay — cards, UPI, PayPal, netbanking.

Free

$0

forever

  • 1 provider connection
  • 7-day history
  • Basic spend chart
  • 1 budget alert
  • Desktop widget — 7-day trial
  • CSV/JSON export — 3 per 30 days

Pro

$0

per month

  • All providers
  • 90-day history
  • Month-end forecast (SES + day-of-week)
  • Spike + drift detection (MAD + CUSUM)
  • Cache savings dashboard
  • Optimization engine
  • Unlimited budget alerts
  • Unlimited CSV + JSON export
  • Unlimited desktop widget access
Go Pro

Razorpay · cards, UPI, PayPal, netbanking

Cancel anytime · No refunds on partial months · Prices include applicable taxes

07 — Questions

Frequently asked.

How does Obol fetch my usage?

You connect your provider API keys (admin keys for OpenAI and Anthropic, management key for OpenRouter's activity API). We call each provider's usage endpoint hourly, with idempotent upserts so re-runs never double-count. For Google AI — which has no public usage API — upload a CSV from Google AI Studio.

How does the month-end forecast work?

Simple exponential smoothing over your month-to-date daily spend gives us a recency-weighted daily rate — early-month outliers stop dominating the projection. Once you have 14+ days of history, day-of-week multipliers kick in so Wednesday doesn't get treated like Saturday. The card shows a confidence tier (low / moderate / high) so you know when to trust the number.

What's different about Obol's anomaly alerts?

Two detectors, not one. A Modified Z-score using median + MAD catches single-day spikes and is robust against the outlier-masking effect that plain z-score has on bursty spend. Separately, CUSUM watches the 14-day window for gradual drift — the $2/day creep that point-spike detectors never see. Both run per-provider, so a +$30 OpenAI spike offset by -$20 Anthropic still fires.

Does caching actually save me money?

Often yes, and Obol shows you exactly how much. The cache dashboard tracks your prompt-cache hit rate per model, the savings you've already earned from hits, and the additional savings still on the table if you improved your hit rate to 100%. Models that don't support caching are excluded — no noise.

Is my API key safe?

Yes. All keys are encrypted at rest with AES-256-GCM. We never log or display plaintext keys.

Which providers are supported?

Nine providers: OpenAI, Anthropic, Google AI, and OpenRouter for dashboard sync + proxy. Mistral, Groq, xAI (Grok), DeepSeek, and Cohere for proxy-only with real-time cost tracking. The LLM proxy works with any OpenAI-compatible API.

Can I cancel anytime?

Yes. Pro is monthly — cancel from Settings → Billing. You keep Pro access until the end of the current billing period.

How is Obol different from Helicone, LangSmith, or Portkey?

Most LLM observability tools require an SDK wrapper, a proxy library, or a framework integration. Obol works by syncing directly with provider billing APIs — no code changes for OpenAI, Anthropic, or OpenRouter. When you do use the proxy, it's a one-line base URL swap, not an SDK. Pricing is $5/month flat, not per-seat or per-event. And Obol tracks cache savings and forecasts your month-end bill — features most tools don't offer at all.

Why Cloudflare?

Fast, global, and free to start. We run on Workers + D1 + KV, which lets us keep pricing at $5/mo while serving from 300+ edge locations worldwide.

08 — Start tracking

Stop guessing.
Start knowing.

Free forever. No credit card. No SDK integration. Connect your first provider in 90 seconds and see where your money goes.