Pricing
Pay for what you ingest
Pay for uncompressed ingress bytes delivered to your S3.
Includes 10 GiB/month with the Starter plan.
Feature comparison
Side-by-side plan details. Reach out if you need custom throughput, payload, or compliance.
| Feature | Starter | Pro | Enterprise |
|---|---|---|---|
| Base price | Free | $99/mo | Custom |
| Usage pricing | $0.10/GiB | $0.06/GiB | Volume discounts |
| Included ingest | 10 GiB/mo included | - | - |
| Output formats | Segments + Parquet (Raw) | All formats (+ Schema-Aware) | All formats (+ Schema-Aware) |
| Schema Views | - | Up to 3 views | 10+ views included |
| Fields per view | - | Up to 50 | 100+ |
| Edge regions | 1 | Up to 3 | Unlimited |
| Autoscaled CUs per edge region | 1 CU (max) | Up to 3 CUs | Custom |
| Sustained throughput | ~0.7-1.0 MiB/s | ~3-4.5 MiB/s (per region) | Custom SLO |
| Max payload size | 1 MiB | 10 MiB | Custom |
| Burst bucket | 100-150 MiB | 2× reserved (≤5 min) | 3× reserved |
| Reserved throughput | - | Add-on (per edge region) | |
| Dedicated outbound IP | - | Add-on | |
| Webhooks & alerts | - | ||
| Usage exports | - | ||
| SLA | Best effort | 99.9% | 99.99% |
| Support | Best effort | Business hours | 24×7 + TAM |
| Compliance (SOC2, DPA) | - | - | |
| Private networking | - | - | |
| Cross-region HA | - | - | |
| BYO KMS | - | - |
What's a Compute Unit (CU)?
A Compute Unit (CU) is EdgeMQ's atomic scaling unit. Each CU handles HTTP ingestion, writes to a durable Write-Ahead Log (WAL), and ships sealed segments to your S3 bucket.
CU Specifications
- ▸CPU: 1 shared vCPU
- ▸Memory: 1 GB RAM
- ▸Storage: 10–20 GB NVMe
- ▸Processes: API + S3 shipper
Performance Envelope
- ▸Sustained: ~1–1.5 MB/s
- ▸Burst: 200 MB token bucket
- ▸p95 Latency: <40 ms at 1 MB payload
- ▸Concurrent: 100–200 connections
Scale horizontally: Pro accounts can run multiple CUs per edge region (up to 3) for higher aggregate throughput. Each CU operates independently with its own WAL and S3 shipper.
How EdgeMQ works
POST /ingest
Send NDJSON, RecordIO, or JSON payloads via HTTP. No brokers, no shards-just HTTP.
Durable append
Data appends to a local Write-Ahead Log with CRC checksums. Sub-10ms p95 latency to durable storage.
S3 artifacts + commit
EdgeMQ seals segments, then materializes your S3 artifacts: compressed Segments, Parquet (Raw), and Schema-Aware Parquet with typed columns (Pro+). Commit markers tell your jobs what's safe to read.
HTTP Request → Durable append (<10ms p95) → Seal segment → Materialize S3 artifacts (segments / Parquet) → Upload → Commit marker (manifest) ✓
Frequently asked questions
Start a free 14-day Pro trial from your billing page with 200 GiB included (or trial ends after 14 days, whichever comes first).
The trial unlocks Schema-Aware Parquet output-automatic schema extraction from your JSON events into typed columnar files perfect for analytics.
No surprise billing: If you don't add a payment method before trial ends, your account automatically reverts to Starter (Segments + Raw Parquet outputs). Your ingest continues - no disruption.
EdgeMQ supports three output formats, configurable per endpoint:
- Segments (all plans) - Compressed NDJSON derived from the WAL. Always enabled as the baseline format.
- Parquet (Raw) (all plans) - Columnar Parquet with opaque payload column. Faster queries than NDJSON for analytics tools.
- Parquet (Schema-Aware) (Pro+ only) - Define views to automatically extract typed columns from JSON payloads. Perfect for data warehouses and BI tools. Up to 3 views on Pro, 10+ on Enterprise.
Starter plan includes Segments + Raw Parquet. Pro trial unlocks Schema-Aware to test before upgrading.
We notify you at 80% and 100% of your included amount. Usage continues seamlessly at $0.10/GiB beyond 10 GiB.
Starter accounts automatically suspend their compute after ~5 minutes of inactivity to minimize infrastructure costs.
Wake time: 1-3 seconds on the first request after idle. Subsequent requests are fast.
Data gap risk: Requests during the wake period may timeout or be lost. For always-on availability, use Pro (scale-to-zero disabled by default).
Best practice for sporadic traffic: If your workload is low-volume and data arrives intermittently, you can pre-warm the ingest node by calling /live on your ingest endpoint host. This liveness check wakes the node and returns 200 when it's ready, so your producer can send the real ingest request immediately after.
The 1 MiB limit on Starter prevents memory exhaustion and ensures low latency for shared-CPU instances.
If you need larger payloads (batch uploads, large analytics events), use Pro for the 10 MiB limit or Enterprise for custom sizes.
Reserved throughput is the guaranteed MiB/s you can sustain per edge region with consistent low latency (p95 ≤10ms). This is purchased as an add-on for Pro accounts.
Burst allowance is a token bucket that lets you temporarily exceed your reserved rate. Pro accounts get 2× reserved capacity for ≤5 minutes; Enterprise gets 3× by negotiation.
Above your burst allowance, requests receive 503 + Retry-After headers to signal backpressure.
Yes. Every EdgeMQ account runs on its own dedicated compute instance with:
- Private NVMe-backed WAL volume (no shared disk)
- Dedicated public IPv4 and IPv6 addresses
- Isolated process boundaries
Starter uses shared CPU scheduling (time-sliced with other tenants) but maintains full data isolation. Pro+ can optionally use dedicated CPUs for guaranteed performance.
We meter uncompressed ingress bytes using the HTTP Content-Length header.
This means you pay for the raw data you send, not the compressed size stored in S3. EdgeMQ automatically compresses segments with zstd before shipping to reduce your S3 storage costs.
Example: Send 100 MB of NDJSON → Charged for 100 MB → Stored in S3 as ~30 MB (with typical compression).
Ready to get started?
Starter is free and includes 10 GiB/month. After that, pay $0.10/GiB. Upgrade anytime as your needs grow.