Live multiview engine
The multiviewer that never blinks.
Conspect takes your cameras, encoders and streams, composites them on the GPU into one clean, templated wall, and serves it back out in the formats your facility already runs. One fixed-cadence clock, one correct frame every tick — no matter what your inputs do.
Pure Rust · Linux & macOS · NVIDIA / Intel / AMD / Apple Silicon · software fallback
Why Conspect
Built around one promise: the wall stays up.
Inputs are sampled — never allowed to pace the output. One misbehaving source can never warp, stall or corrupt the picture your gallery depends on.
01 / CLOCK
Output that never falters
A single fixed-cadence clock emits one valid, correctly-timed frame every tick — independent of any input. A dead camera gets a "no signal" tile. It never gets your wall.
02 / GPU
Commodity hardware, broadcast discipline
NVIDIA, Intel, AMD or Apple Silicon — or pure software. Decode at display size, stay NV12 end-to-end, keep frames on the GPU, and shed load tile-by-tile before the program output is ever touched.
03 / FAN-OUT
Encode once. Deliver everywhere.
The wall is composited and encoded once per rendition, then the same packets fan out to every destination. No per-output re-encode tax, no per-output drift.
Signal path
Many in. One engine. Many out.
Inputs
Supervised reconnect, jitter buffering, per-input timestamp normalisation.
Conspect engine
Outputs
Plus AES67 / SMPTE ST 2110-30 audio — open standards that still reach Dante® networks.
Operations
Built to be operated, not babysat.
Declarative config, an embedded web UI, and a complete REST / WebSocket / SSE API. Everything the UI can do, your automation can do.
Declarative TOML config
The whole wall — sources, layout, outputs — in one reviewable file. Hot-reconfigurable layouts.
OpenAPI 3.1, interactive docs
Live spec at /api/v1/openapi.json, playground at /docs. Scriptable end to end.
Preview isolated from program
Sub-second WHEP/MJPEG confidence preview that can never touch the program path.
[[input]] id = "cam-1" url = "rtsp://10.0.40.11/main" [layout] template = "2+8" # PGM/PVW + 8 tiles [[output]] kind = "ll-hls" path = "/multiview.m3u8"
$ curl -X PATCH conspect.local:8080/api/v1/tiles/6 \ -d '{"input": "replay"}' → 200 OK · wall re-composited, output clock untouched
Mesh
Nodes that find each other.
Every Conspect node announces itself over mDNS — Wi-Fi mesh today, Bluetooth on the roadmap — across the whole broadcast space, not just one subnet.
01 / CLAIM
New machines surface themselves
Power up a fresh node anywhere your fleet can hear it and it appears in your dashboard, ready to claim in one click — no codes to carry across the building.
02 / RELAY
No-Internet networks stay licensed
Nodes on a dark production LAN piggyback their licence leases and telemetry through any mesh peer with a path out. End-to-end signed — a relay can't read or alter what it carries.
03 / ALWAYS ON
Always on, disclosed up front
Like the licence heartbeat, broadcast and discovery can't be switched off. What's announced: presence and claim state as salted digests — never media, never config. Relaying for neighbours is your call.
In the room
From the home lab to the gallery.
The same engine drives a spare mini-PC under a desk and the monitor stack of a production control room.
The tools that run live television have always lived behind high price tags and closed doors. We think the next generation of engineers deserves a way in. No gatekeeping.
That’s why Conspect is free for students, hobbyists and home labs — full engine, full source, no trial clock. The commercial licences on this site are what keep it that way.
Licensing
Free where it should be. Licensed where it pays its way.
Conspect is source-available: genuinely free for personal, home and non-commercial use. Commercial use funds the project — and keeps the free side free.
forever
Non-commercial use is free. Personal, home, hobby and learning use — plus defined free exceptions: First Nations Owned Broadcasters, small Community Broadcasters, and smaller Content Creators.
No trial clock · no watermarks · full source on GitHub · start at conspect.dev
Studio
Consumer-grade machines — desktop CPUs, Apple silicon, GeForce / Radeon RX.
- Everything — no feature gates, channels never metered
- Own it outright or licence it ongoing
- Multiple consumer GPUs, expandable
Broadcast
Commercial-grade machines — server CPUs, pro GPUs, SDI / ST 2110 I/O.
- Everything in Studio, on any hardware class
- Own it outright or licence it ongoing
- Priority support, LTS builds, air-gapped activation
Appliances
Build-to-order hardware we configure, test and stand behind.
- Build-to-order appliance classes, sized to load
- Server-grade: BMC, ECC, redundant power
- Licence and support included
Per device, never per channel. Every licence starts with a full-function evaluation — and our licensing can never take you off air. Short engagement instead? Run it on Conspect Cloud.