For telephony and media-stream providers
Clean audio for every agent on your network.
Offer voice isolation to every AI-agent customer from one place: your media path. μ-law or A-law in, the same format out, 20 ms at a time, on your own servers.
Two ways in.
Media streams
A WebSocket bridge
For streams you already send to agents: take each frame, clean it, forward it. A Go package, with Python and Rust alongside.
engine, err := packethush.NewEngine(packethush.Options{})
call, err := engine.NewSession(8000, packethush.ULaw)
// every media frame of the stream:
err = call.ProcessBytes(in, out) // μ-law in, μ-law outFreeSWITCH
A FreeSWITCH module
Load the module and start it on the leg that carries the caller. Output is identical to the library, sample for sample.
<!-- modules.conf.xml -->
<load module="mod_packethush"/>
<!-- dialplan: on the leg that carries the caller -->
<action application="packethush_start"/>Capacity
280 calls on one server.
Every 20 ms frame on time, 280 calls at once, one 8-vCPU server.
packethushAWS c7i · 8 vCPUone process
- calls
- 280
- per vCPU
- 35
- late frames
- 0
- test
- 60 s, paced
One dot per call, one row per vCPU, slowed down
Licensing that follows your autoscaling.
- One pool of channels
- Shared by all your servers. Servers come and go; nothing is tied to a machine.
- Never drops a call
- If the licence cannot be confirmed, audio passes through unchanged. It is never muted or cut.
- Counted by the second
- Usage is counted in seconds of audio cleaned. Only counters leave your network, never audio or transcripts.
Talk to us about your network.
A trial licence for your media servers, and the evaluation kit to measure it on your customers' calls.
Request access