We can't find the internet
Attempting to reconnect
Something went wrong!
Hang in there while we get back on track
Free MIT core · Aurora Meter Pro from $99/mo per app
Aurora Meter
Laravel Cashier + OpenMeter, for Phoenix. Real-time usage metering, plan entitlements, and Stripe billing. Count, gate, and bill on the BEAM: an ETS hot path that flushes to Postgres and adds up across every node in your cluster.
# count usage: an ETS increment, flushed to Postgres
AuroraMeter.track(org, :ai_generations)
# gate + run + meter, atomically
AuroraMeter.with_quota(org, :ai_generations, fn ->
generate_report()
end)
Three jobs, one library
Every SaaS meters, gates, and bills. On Phoenix you hand-roll all of it.
Meter
Increments hit a lock-free ETS counter: microseconds, no database on the hot path. A flusher persists the totals to Postgres on an interval you set, and every node adds its own deltas, so the stored row is the cluster total and reads converge within a broadcast interval. A broadcaster fans live values over PubSub.
Entitle
Declare plans in a small compile-time DSL, then gate features by plan. `with_quota/4` checks and reserves in one step, so a hard limit holds under concurrency on a node. Across a cluster it is checked against the local view and can overshoot by what the other nodes admitted in one broadcast interval.
Bill
The free core holds a prepaid credit ledger and ships a billing seam; Aurora Meter Pro closes the loop: Stripe checkout, webhook sync, prepaid top-ups, and metered usage reported to Stripe Billing Meters with a per-item idempotency key.
Part 1 · The free core, MIT
Everything you need to count and gate. Free, forever.
The MIT core is the whole metering and entitlements engine: counters on an ETS hot path that flush to Postgres and sum across nodes, the plan DSL and a quota gate that checks and reserves in one step, a prepaid credit ledger with holds and settlement, drop-in LiveView usage components, daily history and telemetry. No email gate, no trial, no expiry.
Live counters
Increments land in ETS and a flusher writes the totals to Postgres on an interval you set, five seconds by default, so the hot path never waits on your database. Anything not yet in an acknowledged flush batch is lost if the node dies, and a database outage makes that window longer than the interval.
Plan DSL + with_quota
Hard caps, metered overage and feature flags, checked and reserved in one step.
Prepaid credit ledger
Grant, hold, settle and release, in integer micro-dollars under a row lock, with an append-only entry for every write. Promotional credit is spent first and expires on its own schedule. Free and MIT, like the rest of the core.
LiveView components
Drop-in usage meters that read the live counters over PubSub.
Postgres, clustered
Every node meters locally and flushes its deltas, so the Postgres row is the cluster total. Reads converge within one broadcast interval, and a hard limit is checked against the local view, so a burst across nodes can overshoot a cap by what the other nodes admitted in that interval.
History and telemetry
Daily history per tenant and per feature, and a telemetry event for every increment, flush, reservation and ledger write, so your own metrics pipeline sees what the meter sees.
Runs on
| Requirement | Free core | Pro |
|---|---|---|
| Elixir | 1.15 or later | 1.15 or later |
| Erlang/OTP | 25 or later | 26 or later |
| PostgreSQL | 13 or later | 13 or later |
| Phoenix LiveView | 1.0, for the usage components | 1.0, for the dashboards |
| Oban | 2.17, for the optional workers | 2.17, for the workers |
| Igniter | 0.8, for the installer only | 0.8, for the installer only |
LiveView, Oban and Igniter are optional: nothing needs them to meter, gate or hold credit. Pro's Erlang/OTP floor is higher than the core's because the Stripe HTTP stack it resolves requires it.
Measured, not estimated
- 3,418,407 increments per second
- AuroraMeter.Counter.incr/4, eight processes on distinct counter keys, median of five runs. A micro benchmark: there is no database in the path. On the same machine an end-to-end durable write measures about 1,800 a second, three orders of magnitude apart, so read which one you need before you use either.
- core docs/evidence/v1/phase-08/08c-results.md, 2026-09-16
- 42.5 seconds to a first meter
- Median of five clean runs, range 41.6 to 44.6 seconds: a new Phoenix application, the dependency added, a cold fetch, the installer, the migration, and a proof that one counter reads back and that the hundredth reservation against a hard limit of 100 is refused. Scripted execution on one machine. Not a human median, not a usability result, not a customer measurement.
- docs/evidence/v1/phase-09/09e-quickstart.json, 2026-09-17
Both on WSL Ubuntu 24.04, AMD Ryzen 9 7900X, Elixir 1.20.1, OTP 29, PostgreSQL 16.13. Not generalised to any other machine or toolchain.
Guarantees
What it promises, and what takes that away.
A promise written without the condition that makes it true is not a promise. So every row here carries both, in the same words as the package's own documentation, and the list below it is what Aurora Meter deliberately does not claim.
One flush batch has at most one durable effect
I01
The flush receipt's primary key is written inside the same transaction as the counter and history deltas, so a batch redelivered after a lost response applies its deltas a single time.
The limit. Counters live in ETS until a batch is acknowledged, so everything not yet in one is lost if the node stops, and a database outage makes that window as long as the outage. Deleting flush receipts while a node could still retry, or writing the counter tables by hand, takes the rest away.
Credit is conserved, per wallet
I10, I11
Every wallet write serialises on that wallet's balance row lock and sufficiency is checked under the same lock, so the balance equals the sum of every transaction amount after every step. A refusal writes nothing and does not roll your own transaction back.
The limit. Conservation is per wallet: there is no cross-wallet total the ledger defends. Writing the ledger tables directly goes round the lock.
Scheduled work is at least once, with effects that are safe to repeat
I16
No scheduled entry point assumes it runs a single time. A flush with nothing pending writes nothing, a due grant expires once however often the sweeper runs, and a grant is idempotent per reference.
The limit. Your host owns the scheduler. Aurora Meter ships no job runner, and it cannot make a job of yours idempotent.
A recorded event is one persisted fact per identity
I06, I07
One persisted fact per tenant and event id, one projection effect and one export intent. A conflicting payload under an identity already used is reported rather than quietly accepted.
The limit. Only for events recorded with an identity. A counter increment is not an event: it carries no identity, it is not deduplicated, and it is not what usage reporting reads.
Usage export is at least once, with a key the provider deduplicates on
I15
Each send carries an identifier Stripe deduplicates on, so a retry of the same window does not bill those units twice. A send whose answer never arrives leaves the item uncertain rather than being resent past the provider's deduplication horizon.
The limit. Aurora Meter never reports that a provider invoice was finalised or paid. It reports what it sent and what it heard back, and raises an item for every difference so a person can resolve it.
Not claimed
- Host job execution
- Scheduled work is at least once. Nothing assumes a job runs a single time, and your host owns the scheduler.
- Strict global quotas
- A hard cap is enforced against the local view, so a burst across nodes can overshoot it by what the other nodes admitted within one broadcast interval. There is no cluster-wide lock on a quota decision.
- Buffered counts
- A count is durable once its flush batch is acknowledged, and not before. The window is everything not yet in one, and it grows for as long as the database is unreachable. Mark the feature durable when losing a count is not acceptable.
- Stripe invoice finalisation
- Aurora Meter reports what it sent to Stripe and what it heard back. Whether an invoice is finalised or paid is Stripe's answer, not ours.
- Unknown provider outcomes
- An item whose answer never arrived is surfaced for a person rather than replayed blindly, and reconciliation raises a difference rather than resolving it.
- The package README The README shipped in the version on offer, with its own guarantees section.
- HexDocs The full correctness index: every invariant with its prerequisites, its known limits and the test that holds it.
- Upgrading to 1.0 What 1.0 starts refusing, why the work happens on the transition release before you install it, the schema steps that have an order, the wallet migration that has no rollback, and the four things that do not change at all.
The showcase
Live usage, rendered where your team already looks.
Aurora Meter ships LiveView components that read the live ETS counters and update over PubSub. Drop them into an admin dashboard and every plan's usage is visible in real time. This is a mock of that surface.
Usage this period
Scale plan · resets in 12 days
Get started
Installed in one command. Metering in two calls.
With Igniter in your deps the installer writes the config, adds Aurora Meter to your
supervision tree after the repo and PubSub, creates a starter plans module and
generates the migration. Pass
--dry-run
to see the diff without writing anything. Without Igniter, the task prints the steps
instead of doing them.
# mix.exs: {:aurora_meter, "~> 1.0"},
{:igniter, "~> 0.8", only: [:dev]}
$ mix deps.get
$ mix igniter.install aurora_meter
$ mix ecto.migrate
# count usage
AuroraMeter.track(org, :ai_calls)
# gate + run + meter in one atomic step
AuroraMeter.with_quota(org, :ai_calls, fn ->
generate_report()
end)
Go deeper
The thinking behind the library.
Long-form writing on usage-based billing in Phoenix, from the ETS hot path to invoicing overage through Stripe. For the hands-on walkthroughs that build a product line by line, the guides are the place to start.
Start here
Usage-Based Billing in Elixir and Phoenix: A Complete Guide
A complete guide to usage-based billing in Elixir and Phoenix: metering events on the ETS hot path, reporting usage to Stripe, and enforcing plan quotas.
Read the guideHow the hot path works
Phoenix Usage Metering in Real Time with ETS
Build real-time Phoenix usage metering with ETS: count API calls on an in-memory hot path, enforce plan quotas, and flush to Postgres for billing.
Read the guideGating by plan
Plan-Based Feature Gating and Entitlements in Phoenix
Phoenix feature gating done right: model plan entitlements as data, enforce in contexts and LiveView, cache with ETS, and handle overrides.
Read the guideGetting paid for overage
Stripe Metered Billing in Elixir: Report Usage to Stripe Billing Meters
Stripe metered billing in Elixir: create Billing Meters, report usage idempotently from Oban, scale with the meter event stream, and reconcile the totals.
Read the guideVersus a hosted meter
OpenMeter Alternative for Elixir: Metering Usage on the BEAM
Need an OpenMeter alternative for Elixir? An honest comparison of both options, plus a BEAM-native metering pipeline built with ETS, Oban, Postgres and Stripe.
Read the guideComing from Laravel
Laravel Cashier for Phoenix: Stripe Subscriptions and Usage Billing in Elixir
Laravel Cashier for Phoenix: build Stripe subscriptions, webhooks, trials, plan gating and usage billing in Elixir, with real code and an honest comparison.
Read the guidePart 2 · Aurora Meter Pro
From live counters to invoices, in one place.
The free core counts, gates and holds prepaid credit. Pro is the revenue and operations layer on top: hosted dashboards for usage, balance and spend, Stripe billing sync, metered-usage reporting, prepaid top-ups and auto top-up, a per-request audit log, historical rollups with CSV export, quota alerts, and reconciliation. Here is what ships.
Usage & billing
Scale plan · billing period Jul 1 to Jul 31
Active subscriptions
128
▲ +6
MRR
$19,072
▲ +11%
Metered revenue
$4,210
▲ +23%
Events reported
1.24M
this period
Metered usage by month
AI generations (thousands)
Included allowances
Live meter events
streaming| Time | Customer | Feature | Qty | Stripe |
|---|---|---|---|---|
| 12:04:31 | Acme Corp | ai_generations | 3 | reported ✓ |
| 12:04:29 | Globex | api_calls | 128 | reported ✓ |
| 12:04:27 | Initech | exports | 1 | reported ✓ |
| 12:04:24 | Umbrella | ai_generations | 7 | reported ✓ |
| 12:04:21 | Hooli | api_calls | 64 | reported ✓ |
Stripe billing sync
Checkout and a self-contained webhook verifier keep subscriptions in lockstep. Metered overage lands on the invoice automatically.
Quota alerts
Threshold alerts are deduplicated per customer, per feature, per period, so a customer hovering at 80 percent gets one email rather than forty. A handler that fails is retried, which can produce a second alert for the same crossing.
Acme Corp at 90% of AI generations
9,000 / 10,000 this period
Globex exceeded the API calls hard cap
250,000 / 250,000, new calls blocked
Initech reached 75% of exports
38 / 50 this period
Historical rollups & export
Month and day rollups persist for reporting, with one-click CSV export for finance.
| Jul 2026 | 120,000 units |
| Jun 2026 | 94,000 units |
| May 2026 | 78,000 units |
| Apr 2026 | 61,000 units |
Reconciliation
A background pass compares local counters with what Stripe holds and raises an item for every difference, so nothing diverges unnoticed. A person resolves the item; the pass reports, it does not rewrite your ledger.
Last run 2 minutes ago · 128 subscriptions · 1.24M events compared
Pricing
Start free. Upgrade when Stripe is charging.
One feature set, licensed per application. Every Pro plan starts with a 14-day free trial.
- Aurora Meter Free · MIT
- $0 forever · count, gate and bill
- Aurora Meter Pro per app
- $99/mo or $990/yr per app
- Aurora Meter Pro Unlimited any number of apps
- $2490/yr · pays for itself at 3 apps
14-day refund on your first invoice, no questions asked · cancel anytime · Licence terms · Terms