Event Infrastructure for Real-Time and
AI-Native Systems.
Stream, store, query, and coordinate distributed systems and AI agents on one durable foundation.
One durable log for every real-time path.
Publish from
Applications
services and products
Agents
workflows and memory
Data changes
events and CDC
Durable append log
ordered · retained · replayable
Use as
Streams
consume by offset
Views
query live state
Agents
replay decisions
One ordered history keeps producers simple and every reader independent.
One client, one grammar
Eight primitives.
One grammar.
Every primitive is one accessor and one verb. Publish a record, query a projection, or ask an agent: every call takes the same shape, object.verb(input).await.
laser.topic(name)
Log
Publish and consume records, replay by offset, batch.
laser.query(index)
Views
Filter, aggregate, page, and vector-search declared projections.
laser.graph(name)
Graph
Link entities, traverse, and find neighbors and nearest vectors.
laser.watch()
Change feed
Await a view's advance instead of polling it.
laser.kv(ns) / .fork(id)
State
Point reads and writes, CAS, leases, copy-on-write branches.
laser.memory(scope)
Memory
Remember, recall (semantic, keyword, hybrid), consolidate.
laser.context(id)
Context
Assemble one conversation's record and scope its memory to it.
laser.agent(id) / .workflow(..)
Fabric
Directed asks, deadline contracts, ordered workflows, runs.
Inside the fabric
The part most systems bolt on separately. Here it stays conventions over the log, so agents inherit durability, replay, and ordering for free.
Discovery
Agents advertise a capability card and a live inbox in one health-aware registry.
Coordination
A directed ask with a deadline, a fan-out to every capable agent, or a pause for a human.
Workflows
Dependency-ordered steps with saga compensation and crash-recovery replay from a journal.
Governance
ActionGovernor reviews every send: allow, block, step up, modify, or defer, with digest-chained evidence.
Durable intent
Typed votes bind a threshold of signers to one effect. It runs only once its exact digest is authorized.
Runs
Submit a run, read its state, list by filter, cancel. Folded from the log itself.
Laser SDK
A small streaming API
with room to grow.
The first calls are topics, records, and offsets. Query, state, and agent coordination stay explicit layers.
// npm i @laserdata/laser-sdk
import { Laser } from "@laserdata/laser-sdk"
const laser = await Laser.connect(
"user:[email protected]"
)
const orders = laser.stream("commerce").topic("orders")
await orders.ensure(4)
await orders.publish().json(order).send()
const rows = await laser.query("orders_v1").filterEq("status", "paid").fetch()
const facts = await laser.memory("support").recall().semantic("refunds").fetch()
await laser.kv("sessions").set("user:42").json(session).send()Typed streaming runs on Apache Iggy. Query, KV, and memory light up on LaserData Cloud.
TypeScript examples ↗Use Cases
Four ways to build on
one durable history.
Start with direct streaming and add delivery semantics, live projections, or replayable agent coordination only where the use case needs it.
Apache Iggy foundation
Typed records. Direct control of the log.
Publish, batch, consume, and replay with caller-owned offsets. Reach the Apache Iggy client directly when the hot path needs it.
Surface
streams / partitions / offsets
Result
Replayable hot path
Best when you need
typed JSON, CBOR, or bytes
explicit partitioning
opt-in batching
Each use case keeps the same retention, ordering, offsets, and replay model underneath.
Apache Iggy engine result
Single-node throughput
1 GB/s send, 2 GB/s poll
Real-time starts at the storage engine.
Published Apache Iggy benchmark results on an AWS i4i.4xlarge.
consumer P99
0.357 ms average
producer P99
0.466 ms average
benchmark run
persistent log workload
AWS i4i.4xlarge. Producer and consumer methodology published with the raw results.
One log in the middle
Tools stay at the edge.
Durable data stays in the middle.
Point-to-point wiring turns agent logic into integration logic. Publish once instead, and let every store, protocol, and consumer evolve independently around the same history.
Point-to-point
Orchestration as integration hub
Route agent calls
MCP and A2A
Synchronize context
cache and vector DB
Write durable events
stream broker
Copy every outcome
analytics and audit
Every new system changes orchestrator code.
Log-native
Orchestration stays workflow logic
Protocols at the edge
publish ordinary records
Shared history
rebuilds from offsets
Apache Iggy
retains and replays
Independent readers
subscribe at their own pace
Add a reader without changing the writer.
Into the platform
Ordinary producers
From the platform
Independent edges and readers
MCP and A2A coordinate. AGDX moves what agents know. A2A speaks SendMessage and GetTask over a signed agent card, MCP maps tools, resources, and prompts, and AG-UI streams state snapshots and deltas, all into the same AGDX envelope. No SSE, no side channel.
Explore integrationsReference app: Photon Market. Follow one order through log-only services, recovery, analytics, and governed agent actions.
View the exampleInside the LaserData console
One console for the whole platform.
Provision a deployment, then browse the streams, projections, memory, and agents running inside it.
One deployment, every detail.
Tier, region, storage, and the exact TCP, QUIC, HTTP, and WS ports your connection string resolves to. Logs, heartbeats, and connectors sit one tab away.
LaserData Cloud
Managed where your data lives.
LaserData-managed infrastructure, your cloud account with BYOC, or your own hardware. Same console and APIs across every deployment model.
Fully managed
01LaserData AWS or GCP
We provision, operate, and observe the complete platform.
BYOC
02Your AWS or GCP account
Your VPC, storage, data perimeter, and cloud commitments.
On-premises
03Your infrastructure
Private cloud, VMs, bare metal, or air-gapped environments.
One path from local to managed
Start with a topic.
Add capabilities as you grow.
Keep the streaming path simple and make every higher-level capability an explicit choice.