OPEN SOURCE · SELF-HOSTABLE · FREE

Know what every agent costs before your invoice does

Ambral measures the real economics of AI-powered work — what your agents consume, what they cost, and why — across every model, tool, and sandbox.

Get started free View on GitHub Read the docs

ambral.dev · system of record

LIVE

Models tracked

…

Providers

…

Events ingested

…

Pricing verified

…

Works with every provider in your stack

OpenAIAnthropicGoogleMistralCohereE2BComposioBrowserbaseMem0

How it works

From AI usage to economic intelligence

track_costs.php

<?php
// Send token counts — Ambral prices it server-side.
$ch = curl_init("https://ambral.dev/api/ingest");
curl_setopt_array($ch, [
    CURLOPT_HTTPHEADER => ["Content-Type: application/json", "x-api-key: ab_..."],
    CURLOPT_POSTFIELDS => json_encode([
        "idempotency_key" => "run-42-call-07",
        "agentId"         => "research-agent",
        "provider"        => "anthropic",
        "model"           => "claude-sonnet-4.5",
        "inputTokens"     => 2400,
        "outputTokens"    => 800,
    ]),
]);
$response = curl_exec($ch);
curl_close($ch);

PHP · Python · Node.js · TypeScript · curl · any language

Instrument in one line

One POST per LLM call, tool use, or compute action. Send token counts — Ambral prices it server-side.

Costs are computed, not reported

Versioned pricing keeps every historical cost accurate. Unknown models are accepted as unpriced and backfilled — never dropped.

Set budget guardrails

Daily, weekly, or monthly limits per agent or globally. Webhook alerts to Slack or Discord before you overspend.

Pricing intelligence

AI pricing changes. Ambral knows.

Provider rates change constantly. Ambral maintains versioned pricing across every model and provider, so your historical costs stay accurate. It is the economic truth behind your AI usage — and every cost is explained, not just reported.

Versioned pricing history — costs re-resolve against the right rate
Change detection — new model rates are tracked and approved
Cost explainability — input tokens × $2.50 + output tokens × $10.00
Unknown models are accepted as unpriced and backfilled, never dropped

View live model pricing

Provider · Model

Anthropic · claude-sonnet-4.5

Price change

Input, before

$3.00

Input, now

$2.50

Effective

Sep 1, 2026

Cost, explained

1,200,000 input × $2.50
+ 42,000 output × $10.00
= $3.42

Coming next: outcome economics — not just what an AI task costs, but what successful work costs, per PR, per deploy, per outcome.

The roadmap

From AI activity to economic intelligence

Today, Ambral answers what your AI costs. Next, why it costs that much. Then, what that spending accomplished.

AI action→Event→Provider + model + usage→Pricing intelligence→Real cost→Agent / task / workflow→Optimization→Outcome

Today

Know what every agent costs

Model, agent, task, and workflow costs — measured and explained in real time.

Know why it costs that much

Pricing intelligence, model comparisons, forecasting, and optimization recommendations.

Future

Know what the work accomplished

Outcome economics — cost per completed task, successful workflow, pull request, and deployment, quality-adjusted.

Capabilities

Built for teams that ship agents

Real-time dashboard

Today’s spend, weekly trends, per-provider breakdowns. No refresh needed.

Per-agent accounting

Budget alerts

Threshold-based alerts per agent or globally. Webhook delivery to Slack, Discord, email.

Cost explorer

Filter by date range, provider, agent, model. Export time-series data.

Multi-provider

OpenAI, Anthropic, Google, Mistral, Cohere, E2B, Composio — or any custom provider.

Self-hostable

Next.js + MySQL. Deploy to any VPS, Docker, or your own server. MIT-licensed, no vendor lock-in.

Pricing

Know what every agent costs.

Know why. Know what it accomplished.

The core is MIT-licensed and free to self-host forever. Paid plans unlock scale, history, intelligence, and collaboration in the managed Cloud.

Community

For developers experimenting with Ambral

$0forever

Get started free

100K events / month
30-day retention
3 agents · 1 workspace
Real-time dashboard
Basic cost tracking & budgets
3 alerts
All providers · API access
Self-host the MIT core
Community support

Pro

For developers running AI in production

$49/ month

Then $5 per additional 1M events

Start Pro

2M events / month
1-year retention
Unlimited agents & projects
Unlimited budgets & alerts
Slack / Discord / email / webhooks
Batch ingestion
Forecasting & anomaly detection
Managed pricing intelligence
Historical pricing & export
10 team members
Priority support

Team

For teams operating multiple production agents

$149/ month

Then $4 per additional 1M events

Start Team

10M events / month
2-year retention
Unlimited team members
Advanced RBAC & audit logs
Advanced budgets & forecasting
Provider/model cost comparison
Cost optimization recommendations
Scheduled reports & exports
Priority support

Enterprise

For AI economics at scale

Custom

Custom event volume & retention
SSO/SAML · SCIM · advanced RBAC
Audit & compliance
Data residency & private deployment
SLA & dedicated support
Custom pricing sources & integrations
Cost allocation & chargeback
Enterprise reporting

Managed pricing intelligence keeps your registry accurate as providers change their rates — that’s the reason to pay, not the token count.

Built for real agent workloads

One dashboard for every dollar your agents spend

5 min

To first dashboard

Clone, configure, deploy. See costs immediately.

100%

Open source core

MIT-licensed. Self-host forever. No vendor lock-in.

Any provider

LLMs, compute, tools

OpenAI, Anthropic, E2B, Browserbase — one view.

Start tracking agent costs in 5 minutes

MIT-licensed core. No credit card. No vendor lock-in. Self-host forever or let us run it for you.

Get started free View API docs