Ignite
Event-driven compute

Ignite runs your code the millisecond an event arrives — a webhook, a queue message, a row change, a cron tick — then tears it down before the meter starts on idle. Nothing to provision. No cold-start tax. Functions that spark to life, do the work, and disappear.

The event backbone behind backends that never sleep

VoltlinePulsar FoodsNorthbeamCinderworkSwitchyardHalo LogisticsVoltlinePulsar FoodsNorthbeamCinderworkSwitchyardHalo Logistics
The runtime

Built for themoment something happens.

Most platforms keep a server warm just in case. Ignite keeps nothing warm and still answers in milliseconds — because the runtime, the scheduler, and the network were designed as one machine, not three bolted together.

9-millisecond cold starts

Our microVMs resume from a frozen memory snapshot instead of booting an OS. The first request after an hour of silence lands as fast as the millionth. We never pre-warm, and we never bill you to.

Triggers, not endpoints

Wire a function to a queue, an object upload, a database row change, a schedule, or an HTTP route. The event is the contract: you write the handler, we deliver the payload and the retries.

Scale to zero, then to a million

Idle costs nothing — not a cent of reserved capacity. A 3 a.m. traffic spike fans out to ten thousand isolated instances and back down before the dashboard finishes redrawing.

Exactly-once, on purpose

Every event is fenced with an idempotency key and retried with backoff until it lands. Duplicate deliveries are deduped at the edge, so your handler never has to.

State that survives the teardown

Durable execution checkpoints a function between steps. A workflow can sleep for thirty days, then resume on the exact line it paused — no orchestrator, no babysat queue, no state machine to hand-roll.

What the spark looks like at scale

9ms
Median cold start
$0
Cost while idle
14B
Invocations per day
310
Edge regions
Developer experience

From git pushto live in 800ms.

No YAML labyrinth, no provisioning step, no deploy you schedule for a quiet Tuesday. Push your code; Ignite builds it, isolates it, and routes the first event before your terminal finishes scrolling.

Deploy on push

Connect a repo and every commit becomes a versioned, one-click-rollback deployment. Pull requests get a live preview function on their own URL, automatically.

Write in your language

First-class runtimes for TypeScript, Python, Go, and Rust — plus any binary you can drop in a container. Same cold start, same per-millisecond billing, your call.

Tail logs in real time

Stream structured logs and traces straight to your terminal as events fire. Every line is linked back to the exact event that triggered the invocation.

Replay any event

Caught a bug in production? Pull the exact payload from the event log and replay it against your local function, byte for byte, until the fix holds.

Patterns

Five lines wired to the real world.

Ignite is happiest gluing systems together. These are the shapes our teams reach for first — each one a handler, a trigger, and nothing else to run.

HTTP trigger

Stripe webhook to ledger

Catch payment events, verify the signature at the edge, and write to your ledger with exactly-once guarantees. No double charges, even when Stripe retries.

Object trigger

Upload to thumbnail pipeline

An image lands in storage; Ignite fans out to resize, transcode, and tag it across parallel functions, then writes the manifest back when the last one finishes.

Queue trigger

Queue-driven fan-out

Drain a message queue with automatic concurrency limits and dead-letter handling. Back-pressure is the platform's problem now, not a thing you tune by hand.

Cron trigger

Nightly rollups

Schedule a function down to the second across regions. It wakes, aggregates the day, writes the report, and bills you for the ninety seconds it was alive.

Database trigger

Row-change to search index

A row changes in Postgres and Ignite streams the delta to your search index inside the same heartbeat. No polling loop, no nightly reindex, no drift.

Durable workflow

Thirty-day signup drip

A workflow that sleeps between emails for a month and resumes precisely where it left off — with zero infrastructure kept alive in the gaps.

From the on-call channel

The teams who stopped renting idle servers.

We deleted an entire Kubernetes cluster that existed only to run cron jobs and webhooks. The compute bill dropped 71% the first month, and on-call went quiet.

R
Renata Voss
Staff Engineer, Switchyard

Cold starts were the one reason we never went serverless. Ignite's first request comes back in single-digit milliseconds. That objection just evaporated.

D
Daniel Okonkwo
Head of Platform, Northbeam

Durable workflows replaced an orchestration service we were dreading building. A function sleeps two weeks and wakes on the right line. We shipped the feature in an afternoon instead of a quarter.

M
Mei-Lin Zhao
Principal Engineer, Cinderwork
Pricing

You pay for the spark, not the idle.

Billing is per millisecond of execution. A function that sleeps costs exactly nothing. No seats, no reserved capacity, no surprise egress line on the invoice.

Spark

For side projects and the first version of everything.

$0/mo
  • 2M invocations / mo included
  • All trigger types
  • 9ms cold starts
  • 7-day log retention
  • Community Discord
Most popular

Ignition

For production backends real users depend on.

$49/mo
  • 100M invocations included, then $0.18/M
  • Durable workflows + event replay
  • 30-day log retention
  • Concurrency up to 10,000
  • 99.99% uptime SLA

Reactor

For event volume measured in billions.

Custom
  • Unlimited invocations
  • Dedicated regions + private networking
  • Compliance pack (SOC 2, HIPAA, ISO 27001)
  • Named reliability engineer
  • Custom retry and isolation policies

The questions every backend team asks.

How are cold starts 9 milliseconds?

We snapshot a fully-initialized microVM to memory and resume from that frozen image instead of booting an OS and runtime on every cold request. Your function is serving traffic before a traditional container has finished starting its kernel.

What can trigger a function?

HTTP requests, message queues, object-storage uploads, database row changes, scheduled crons, and any event you publish to our event bus. A single function can listen on several triggers at once.

What happens when an event fails?

Failed invocations retry with exponential backoff, fenced by an idempotency key so the retries stay safe. Anything that exhausts its retries lands in a dead-letter stream you can inspect, fix, and replay.

How is this different from running my own functions on a hyperscaler?

No provisioning, no warm-pool tuning, no per-region setup, and no cold-start penalty. You wire a trigger to a handler; we own everything between the event and your code — scale, isolation, and delivery guarantees included.

Am I locked into a proprietary runtime?

No. Functions are plain TypeScript, Python, Go, Rust, or any container. Triggers map to open standards like CloudEvents, so your handlers stay portable the day you want to leave.

Wire your first trigger before this tab loses focus.

Free for two million invocations a month. No credit card, no sales call, no cluster to stand up first.