Camera intelligence infrastructure for developers

Cameras in.
Events out.

Coming SoonExpected beta 2027
scroll

Skip the pipeline. Ship the product.

Drop the open-source agent next to your camera, draw your zones and triggers in the dashboard, and Upstream streams typed events into your stack. RTSP ingestion, GPU inference, tracking, and event delivery are handled — so you ship product logic instead of pipelines.

Sub-100ms end-to-end latencyRTSP · IP cameras · webcam ingestionWebhooks · REST · WebSocket event delivery
ZONE_A · front_door
agent · cam_dock_0328.4 fps · 1080p
lat 84ms · gpu a1011:42:08 PDT
Live events200 OK
person.entered_zonecam_dock_03 · front_door
11:42:08
vehicle.crossed_linecam_dock_07 · loading_bay
11:42:04
forklift.dwellcam_aisle_02 · aisle_3
11:41:55

Four steps from camera to callback.

The whole pipeline — ingest, inference, event logic, delivery — runs as managed infrastructure. You configure it; we keep it flowing.

01

Connect a stream

Point an RTSP, WebRTC, or ONVIF camera at Upstream. Edge devices, IP cameras, drones — we ingest them all at the source.

02

Configure visually

Draw tracking zones, pick detection models, set thresholds and dwell rules in a hosted studio. No model wrangling required.

03

Author event logic

Compose conditions across detections, zones, and time — “forklift in aisle 3 for >30s” — without writing infrastructure code.

04

Trigger anything

Push to your webhook, queue, database, or workflow. Upstream guarantees delivery; you decide what happens downstream.

Everything upstream of your product.

RTSP ingestion, GPU inference, orchestration, scaling, monitoring — all of the plumbing that usually takes six months to build, run as a hosted service.

Tracking zones

Draw the world, watch what flows through it.

Polygonal zones, virtual tripwires, dwell timers and entry/exit logic — all editable in the browser, versioned, and deployed in one click.

zone_a · loadingzone_b · staging
Event pipeline

Detections become events become actions.

Compose typed events across cameras, zones, and time. Upstream handles fan-out, retries, and ordering — your handlers stay simple.

detection zone enter dwell 30s webhook Slack alert
Managed infra

Hosted compute that scales with the stream.

GPU autoscaling, regional pinning, monitoring — without an MLOps team.

0ms
P50 LATENCY
0.00%
UPTIME · 30D
0×
REGIONS
STREAMS
upstream/handlers/forklift.ts
webhook
// Triggered when a forklift dwells in a restricted zone
import { upstream } from "@upstream/sdk";

export const onEvent = upstream.on("forklift.dwell", async (e) => {
  if (e.zone === "aisle_3" && e.duration_s > 30) {
    await alerts.page({
      team:     "floor-safety",
      camera:   e.camera_id,
      // signed, 5-min URL to the clip frame
      preview:  e.frame({ format: "jpeg", w: 1280 }),
    });
  }
});

// Latency: 84ms · Delivery: at-least-once · Region: us-west-2

Live cameras, real product.

Teams use Upstream wherever a stream of pixels needs to become a stream of business events.

WH
Warehouse & logistics
Dock occupancy, pallet counting, PPE compliance
RT
Retail analytics
Queue length, dwell heatmaps, conversion funnels
MF
Manufacturing
Line throughput, defect flagging, safety perimeters
SC
Smart cities & mobility
Curb activity, traffic counts, incident detection
AG
Agriculture & field ops
Livestock tracking, equipment telemetry, intrusion

Stop building pipelines.
Start shipping what they enable.

Spin up a free workspace, point a camera at it, and have a working event flowing into your stack in under ten minutes.