Point your sources at Basalt.
Kafka topics, Postgres change streams, webhooks and SDK calls land in one ordered log. Retries are deduplicated, so every event counts once.
Now in 38 regions
Every event, set in stone.
Basalt catches every event the moment it lands and sets it into typed, joined tables you can query in milliseconds. Write SQL. Read an API. Skip the pipeline.
p994.2 ms
Throughput1.28M ev/s
Setup9 min median
-- live, last 60 seconds SELECT region, event, count(*) AS per_min, p99(latency_ms) AS p99 FROM stream.events WHERE ts > now() - '1 minute' GROUP BY 1, 2 ORDER BY per_min DESC;
Region
Event
Per min
p99
eu-north-1
payment.succeeded
48,210
2.1 ms
us-east-2
model.inference
41,877
3.4 ms
ap-south-1
session.started
36,502
1.8 ms
eu-west-3
cart.updated
29,914
2.7 ms
us-west-1
sensor.reading
27,340
1.2 ms
812 rows · 3.1 ms · eu-north-1
Refreshing every event
Trusted by 2,400+ teams
How it works
Three moves, no glue code.
Shape it with plain SQL.
Write a view once. Basalt keeps joins, windows and aggregates current with each event, with lineage back to the raw row.
Read it as an API.
Every view is a REST, GraphQL and Postgres endpoint the moment it exists, served from the closest of 38 regions.
Capabilities
Solid under every load.
p50
1.7ms
p99
3.9ms
Reads
412k/s
p99 latency · last 60 s
ms
Single-digit milliseconds, every read.
Hot rows stay in memory beside the view that produced them. Your app, dashboards and models read the same answer at the same speed.
stream.orders
stream.payments
orders_enriched
api / v1
Trace any number to its source.
Click a cell and see the exact events that produced it, down to the column.
Close to every user.
Replicas follow your traffic with sub-second failover.
Schemas that evolve.
Add or retype a field with no migration window.
Time travel on every table.
Query any view as of any second in the last 30 days. Answer last Tuesday's question with last Tuesday's rows.
Exactly once, end to end.
Idempotent writes with per-key ordering. Retries, replays and network splits never count a payment twice.
Use cases
One layer, many shapes.
Use case · Fintech
Score every payment before it settles.
Fraud models need the last thirty seconds of behavior. Basalt joins payments, sessions and device signals as they happen and serves a fresh feature row in milliseconds.
- Payments, ledgers and KYC events in one ordered log
- Per-account features updated on every transaction
- Exact replay for auditors with 30-day time travel
Decision time 3.8 ms
False positives -31%
Use case · Commerce
Stock, price and cart, always in agreement.
Inventory, pricing and carts change every second across warehouses and storefronts. Basalt keeps one live view of each SKU, so you only sell what is on the shelf.
- Postgres CDC from every warehouse database
- Live availability per SKU, region and channel
- Cart and checkout events joined in the same view
Oversells 0 last quarter
Views refreshed per event
Use case · AI agents
Give your agents memory that stays fresh.
Agents act on what they can read. Basalt exposes live, row-level context as typed tools, so an agent sees the order that shipped a second ago and follows the same access policies as your app.
- Typed tools for Corvid, Nimbia and Chainweave
- Row-level security on every agent call
- Embeddings stored beside the rows they describe
Context age < 50 ms
Tool calls 1.9M / day
Use case · IoT
Millions of sensors, one queryable map.
Fleets, meters and machines report in bursts and out of order. Basalt orders every reading by event time, windows it per device and keeps the latest state ready for dashboards and alerts.
- Event-time windows that handle late arrivals
- Per-device state with 7-year retention on Enterprise
- Alerts fired from SQL, delivered by webhook
Readings 412k / s
Late events handled 100%
After the switch
Proof from production.
Events dropped
0
Across every customer cluster, trailing 12 months.
p99, 11 → 4.2 ms
4.2ms
Median customer, down from 11 ms on their previous stack.
Pipelines replaced
14avg
Queues, jobs and caches retired per team in year one.
Median setup
9min
From sign-up to the first query on live events.
Customers
Teams that stopped babysitting pipelines.
“Our ledger and our fraud checks finally read the same numbers. Reconciliation used to take a day. Now it runs as a view and nobody thinks about it.”
Ragnheiður Jónsdóttir
Head of Platform, Ferrous
“We track 60,000 trucks. Late GPS pings used to scramble our ETAs. Basalt orders them by event time, and dispatch trusts the map again.”
Daniel Okafor
Staff Engineer, Halcyon
“Our agents read customer state through one typed tool. It is fresh to the second and respects every tenant boundary. That was months of work we never had to do.”
Priya Raman
Co-founder, kiln.dev
FAQ
Before you start.
How long does it take to get started?
Most teams run their first query on live events in under ten minutes. Create a workspace, connect a source with the CLI or the dashboard, and write a view in SQL.
Do I have to replace Kafka?
No. Basalt reads your existing topics with consumer-group offsets and exactly-once commits. Many teams keep Kafka as the transport and retire the stream processors, jobs and caches behind it.
How is Basalt priced?
By events ingested. Hobby is free up to 10M events a month, Scale is $249 a month with 1B events included, and Enterprise is a custom annual agreement. Seats and queries are never billed.
Where does my data live?
In the regions you choose. Scale lets you pick three of our 38 regions. Enterprise adds private regions and bring-your-own-cloud deployments in your AWS, GCP or Azure account.
Can I query data as it looked in the past?
Yes. Every table supports time travel. Query any view as of any second in the last 30 days on Scale, or a custom window on Enterprise.
Get started
Set your data in stone.
Free up to 10M events a month. No card, no sales call. Your first pipeline is live in about four minutes.