Self-hosted business telephony

A business phone system you actually own.

Saj Serve is a complete SIP PBX. Browser softphone, trunks, routing, queues, one-command install on your own Linux server. Your calls, your database, your rules.

core is free · unlimited extensions · 5 concurrent external calls · forever

or talk to the engineers who built it

headless install early access build
One command · Debian · Ubuntu · Fedora · RHEL · Rocky · Alma · Arch Standard SIP (RFC 3261) · Digest MD5/SHA-256 · STIR/SHAKEN PCMU · PCMA · Opus Browser softphone · WebRTC · DTLS-SRTP · ICE-Lite 7 client SDKs · Go · Kotlin · PHP · Python · Ruby · Swift · TypeScript Managed TLS via Caddy, or bring your own nginx

We're new, so there is no logo wall. There is a security page with everything you can verify yourself.

Deploy your way

One binary. Our cloud, your VPC, or your rack.

Same software in all three homes. Move between them whenever you want.

Hosted by us

We run it.

You still own the data and can leave with it. The export is a documented procedure.

In your cloud

Your VPC runs it.

Your AWS, Azure, GCP or DigitalOcean account. Your network boundary, your storage.

On your hardware

Your rack runs it.

Your building, your jurisdiction. The answer regulated industries keep asking cloud vendors for.

The AI engineer

Tell it what you want done. It's the engineer.

Saj Serve ships with an engineer you talk to. It holds your permissions and never more, it confirms before it changes anything, and it points at your own model if you want — nothing leaves your box unless you allow it. It configures and operates the system. It never answers your customers' calls. And it drives the same built-in MCP surface your own AI can use — point any MCP client at your PBX and it gets the same tools, under the same permissions.

For installers
Deploy a phone system for my customer. 10 extensions, 2 of them cordless, and one desk phone is a reception-style console.
Sure. How do they want the ring groups — one main group that rings everyone, or reception answers first and overflows to the floor?
Reception first. Overflow after 15 seconds.
Here's the plan: extensions 101–110 · 101 set as reception · ring group "Main" rings 101 first, overflows to 102–110 after 15 seconds. Apply it?
Apply.
Done. The extensions and ring group are live. Anything else for this site?confirmed · 12 changes · run as you
For your team
Monday's a public holiday — set every extension to do-not-disturb.
DND across 101–110 from Monday 00:00. I'll lift it Tuesday 09:00 and confirm when I have. Apply?
Apply. Also two new starters begin Friday — the boss wants them ringing after reception.
Extensions 111 and 112 created and added to ring group "Main", after 101. Unlimited extensions — this cost you nothing.confirmed · 2 extensions · ring order updated
And rename extension 104 to "Priya — Accounts".
Renamed. The directory and every wallboard now show Priya — Accounts.
For telcos & MSPs
Run the monthly audit across my tenants. Anyone trending over their concurrent-call tier?
3 of 41 tenants peaked above 80% of tier this month: Harbour Dental hit 16/16 twice, NorthPoint Legal 14/16, Kestrel Freight 29/32. Want upgrade quotes drafted for the first two?
Yes. And run the June billing.
Quotes drafted against your rate card. June invoices are synced to your accounting platform, per tenant, with your margins added.confirmed · CDR-backed · nothing sent without you

simulated conversation · the engineer runs with your permissions, confirms every change, and works against your model or ours

A real PBX

If it's missing from this list, it isn't in the product.

Most "modern" phone platforms deleted this vocabulary from their homepages. We build with it. Everything below is shipping code.

SIP transaction & dialog state machinesRFC 3261
Full INVITE / BYE / REFER call lifecyclegateway
Digest authenticationMD5 · SHA-256
STIR/SHAKEN call identityRFC 8224/8588
Dialplan engineper-tenant
RTP media with adaptive jitter buffermedia plane
DTMF detectionGoertzel
Conference mixingN-way
T.38 fax relaygateway
ACD queues with skills routinground-robin · ring-all · longest-idle
Ring groups with call pickupgroups
Call park, retrieve, transfer, DND*88
Overhead pagingmulticast
Inbound DID-first routing, fail-closedno cross-tenant guessing
Trunk failover by priorityordered
Outbound diallerpredictive · progressive · power · preview
Do-not-call enforcement — refuses to dial if lists won't loadfail-closed
Toll-fraud spend capsper-tenant
Workforce managementforecast · schedule · adherence
Live wallboardqueues · SLA · agents
Least-cost routing from rate deckscost & sell prices
CRM adaptersSalesforce · HubSpot · Dynamics +9
Signed outbound webhooksdelivery-stamped · testable
OpenAPI-first — SDKs & agent tools generated from itlive contract
Admin UI locales, RTL includedEN · AR · HI
Call detail recordsCDR export
MCP server — your AI tools can drive the PBXbuilt in
PBX discovery adapters that inventory what you run todayAsterisk · FreePBX · 3CX +4
Built for operators

One system. Every customer on it.

If you run phone systems for other people — an MSP, an IT shop, a telco — the multi-tenant machinery is already in the box. This is the platform the white-label program ships on.

Multi-tenant park

Tenant hierarchy, hard isolation.

Every customer is a tenant with row-level isolation enforced in the database, scoped admin, and a wholesale API for the park. Cross-tenant access is regression-tested in CI.

Toll-fraud telemetry

The bill-shock watcher.

Toll fraud is how operators lose real money overnight. Per-tenant spend caps and fraud telemetry are built into the platform and exercised by CI on every merge — not a bolt-on you remember to buy later.

Terminal admin

SSH in. No browser required.

A real terminal client ships alongside the web console — inspect and administer the system from the shell you already live in, scriptable one-shots included.

Rate decks carry both your cost and your sell price, least-cost routing picks the trunk, and per-tenant usage rolls up to the wholesale API. Full white-label — your brand down to the softphone — is in the pipeline below. The tenancy underneath it works today.

In the pipeline

What's coming. Engineering underway, no promised dates.

The list above is what's real today. This is what's being built. When something ships, it moves up — that's the whole system.

Full white-label for telcos & MSPs — your brand on everything, down to the softphonein build
iOS & Android softphone appsin build
Zero-touch handset provisioning — Yealink · Poly · Grandstreamin hardening
Voicemail, music on hold, operator panel — full parlour setin hardening
Saj Serve Edge — a branded appliance for the comms cupboarddirection set
Site survivability — phones keep working when the internet dropsin engineering
Video & screen share on the same lineroadmap
Wi-Fi to cellular call handoffroadmap
Microsoft Teams direct routingroadmap
Hosted tier with a documented move to self-hostedearly access

Nothing here is for sale yet. If one of these is the thing you need, tell us which — it genuinely moves the order.

Where your call actually goes

Count what your audio passes through.

the self-hosted path. real components, real ports. full data-path on the security page made to be dropped into a procurement deck
Pricing

Priced per system, not per seat.

Per-seat plans grow your bill every time you hire. Saj Serve is licensed per system, sized by concurrent calls. Your hundredth extension costs exactly what the first one did: nothing.

Policy 01

Unlimited extensions. Always.

No per-user fees, no extension caps, no "fair use" policy waiting at renewal.

Policy 02

Internal calls don't count.

Only external calls count toward your tier. Desk-to-desk is your network and your business.

Policy 03

The whole table is published.

Every tier, every number, on the pricing page. No quote gate in front of a price.

See the full table Pro from $495/yr at 8 concurrent calls · Free tier below that, forever.
Trust, without the badge wall

We can't hand you a SOC 2 report for your deployment. It isn't ours to audit.

What we publish instead, with no form in front of it:

The architecture and full data-path. The per-leg encryption facts, stated exactly. The disclosure policy, with a human behind it.

And the argument only self-hosting can make: it runs inside the environment you already had audited.

The export path is never paywalled. Config and data export works on every tier and keeps working if your licence lapses — that path is hard-coded to never gate, like emergency calling.

Read the security & architecture page →

Honesty cuts both ways

Who this isn't for.

  • You want someone else to be responsible for the servers.
  • You need an agentic AI contact centre today.
  • Nobody on your team can run a Linux box.
  • Your procurement pack requires a Gartner citation.

If that's you, the incumbents are genuinely good at what they do. If it isn't, you're exactly who we built this for.

Early access

Own your phone system.

Early access is open to teams who run their own infrastructure. Send your current PBX and your peak concurrent external calls. An engineer answers within one business day.

this server answers to you.