← Overview
How it works

Global PoP packet transport.

QUIC and MoQ, from your SDK to the modular core — relayed through the nearest edge PoP and observable at every hop. One line, any deployment.

QUICMoQ · media over QUICcloud · on-prem · air-gapped
How it works · The path of one packet

It starts with one packet.

A unit of your data — voice, video, telemetry, or a tool call.

Your SDK puts it on the line.

Idiomatic in every language. Same wire under it.

Nearest edge PoP picks it up.

Global anycast. Closest hop to the user.

Then the modular core takes over.

Modules compose the vertical. The substrate stays the same.

A global PoP network observes it.

Every packet, traced — end to end, every hop.

01 · SDK

Speak QUIC from anywhere.

Point your app packet using our sdk at the line and go. The transport is handled for you — your app just sends and receives on one connection.

  • Idiomatic clients for Python, TypeScript, and C++/Rust
  • Compatible with the features you already use
  • The SDKs live in the Telequick GitHub →
py pip install telequick
ts npm i telequick
++ bare-metal bindings
02 · Edge PoP

Enter at the nearest PoP.

Every connection lands at the closest global point of presence, then relays inward over QUIC and MoQ. Sessions migrate cleanly across networks and survive heavy packet loss where older transports stall.

  • Global edge PoPs — packets take the shortest path in
  • QUIC + MoQ: one connection for audio, video, data and control
  • Connection migration and loss tolerance built into the wire
03 · Modular core

One core, every module.

The core is modular: voice, inference, SIP, robotics and the rest plug into the same transport instead of standing up their own. Run it where you need it — managed cloud, your own racks, or fully air-gapped.

  • Add a modality, not a stack — modules share the line
  • Bare-metal core, zero-copy frame handoff
  • Any deployment: cloud, on-prem, edge, air-gapped
voice ai
inference
sip/webrtc
robotics
modular core · one QUIC transport
04 · Observability

See every hop.

Traces, metrics and logs follow the packet the whole way — SDK to PoP to core and back — on the same line carrying the media. No second pipeline to wire up, no blind spots between vendors.

  • End-to-end traces across every hop and modality
  • Live metrics: RTT, loss, jitter, throughput per track
  • Structured logs on the same transport as the media
traces
metrics
logs
rtt · loss
Inside the core

A shard per core. Nothing shared.

The core is shard-per-core — one shared-nothing shard pinned to every CPU. Each shard owns its connections, memory and queues outright, so there are no mutexes, no locks, and no cross-core traffic on the hot path.

packets in · QUIC datagrams
cBPF steers each connection to a fixed shardin-kernel
core 0
lock-free
core 1
lock-free
core 2
lock-free
core 3
lock-free
core 4
lock-free
core 5
lock-free
shared-nothing · no mutex, no locks, no cross-core traffic
GSO batches many packets into a single writeoffloaded to the NIC
packets out · batched

cBPF steering

A classic BPF program in the kernel hashes every connection to one fixed shard via SO_REUSEPORT, so a flow always lands on the same core — no rebalancing, no shared lookup table.

Shared-nothing shards

Each core is its own shared-nothing reactor with private memory and queues. Work never crosses cores, so there's nothing to lock and nothing to contend on.

GSO batch writes

Outbound, generic segmentation offload packs many packets into a single syscall — the kernel and NIC do the slicing, so one write pushes a whole burst.

Put it on one line.

Telequick moves every modality over QUIC, from your SDK to the core.