Sales team almost ready · join the waitlist

Run a team of AI agents.Keep every decision that matters.

Switchboard runs departments of specialist agents on your own infrastructure and your own model keys. No manager AI in the middle. No workflow canvas to wire up. Nothing reaches a real person until you say yes.

See the teams
0
messages sent behind your back

Anything reaching a real person waits for a human decision.

1
shared record, no agent chatter

Agents never converse. Each writes its result down and moves on.

$
every run priced and capped

See what each run cost, and set a monthly ceiling it can't pass.

Why it's built differently

Not an agent orchestrator. Not a workflow builder.

Most agent platforms put a model in charge of the other models. Switchboard doesn't, and that one decision is why it holds up on real client work.

The usual approach
AgentAgentAgentAgentManager AIdecides who's next
  • A model decides who acts next, every time
  • Agents pass work along in conversation
  • The job has to fit in one conversation
Switchboard
The clockA stage changeAn eventAgentAgentAgentOne shared recordevery lead, every message, every decision
  • The clock, a stage change or an event decides
  • Agents write to the record, never to each other
  • Work spans months, not one conversation

Works on your timeline

A follow-up due in eleven days still happens.

Predictable

Same situation, same behaviour, every time.

Nothing secondhand

Each agent reads the full record itself.

No coordination tax

You pay for work, not for deciding who acts.

The trade-off: it won't improvise a plan of its own. For something new you add a step or a schedule in the interface, and in exchange you get behaviour you can explain to a client.

What it costs to run

Cheaper by design, and visible to the cent

A supervisor architecture pays a model to decide who acts, then pays again to re-send the conversation. Neither line item exists here, and what you do spend is on screen, with a limit you set.

Supervisor architecture
Deciding who acts
Re-sending the transcript
The actual work
Switchboard
The actual work

Illustrative, not measured. The point is which line items exist, not their size. Coordination here is plain code, so it costs nothing to run.

Coordination costs nothing

No supervisor model deciding who acts next. That decision is plain code: a schedule, a stage, an event.

Context doesn't snowball

Agents don't pass a growing transcript around. Each reads just what it needs from the record, so run cost stays flat as the deal ages.

Cheap jobs use cheap models

Sorting a reply and scoring a lead don't need a frontier model. Each kind of task routes to the cheapest one that can do it.

Stuck agents get stopped

Repeat calls are served from cache, and an agent going in circles is failed by name, not left running up a bill.

And you can see every cent of it

Every run has a price on it

Open any run and see what it cost, next to the tokens it used. Failed runs included, because they cost money too.

A running total, per client

Month-to-date spend for each workspace, so you know what a client costs you before the invoice tells you.

A limit that actually stops it

Set a monthly ceiling. At 80% you're warned. At 100% Switchboard either moves everything to cheaper models or pauses the non-critical agents, whichever you chose.

You enter your own rates once, from your provider's pricing page, so the figures match what you're actually billed, including any discount you negotiated.

The division of labour

The agents do the work. The consequences stay yours.

What the agents do
  • Do the research and fill in the gaps
  • Draft everything that goes out, in your voice
  • Read and classify everything that comes back
  • Score and prioritise against the bar you set
  • Work to a schedule, for weeks or months
  • Report what happened, in plain language
What stays with you
  • Whether anything is sent at all
  • The exact wording, before it goes
  • Any commitment made on your behalf
  • What your agents are allowed to say
  • Which facts they remember about your business
  • What you're willing to spend in a month
How it works

Set up in an afternoon

No pipeline to design, no automation to wire, no prompt to write.

  1. Install a team

    The whole department in one click: its agents, their skills, its pages and its reports.

  2. Answer a questionnaire

    Who you serve, what you offer, your voice, what's off limits. Every agent reads it.

  3. It gets on with the work

    Each agent wakes on its own schedule, or when something happens that concerns it.

  4. You approve what matters

    Anything reaching a real person waits for you.

Agent teams

Hire a team, not a tool

A team installs a whole department at once: the roster, its skills, its pages and its reports. One is nearly ready, and we're honest about where the rest are.

The Sales team

Sales

Almost ready
The Outbound Growth Engine

Sources leads, enriches them, runs email and WhatsApp cadences, triages every reply, scores against your bar, and briefs you before the meeting.

10 agentsMeet the team

Support

In build
The Support Desk Operator

Triages tickets and live chats, drafts replies from your help centre, escalates what needs a person, and reports on response time and satisfaction.

Roster in progressMeet the team

The departments we intend to build next

Not built yet, and listed so you can tell us which one you'd hand over first. Every one of them is the same shape of work: repetitive, scheduled, and something you'd still want to sign off.

Finance

Planned

Chases unpaid invoices politely and persistently, reconciles what's come in, and tells you where the cash actually is.

What it would do

Operations

Planned

Watches the routines that are supposed to happen, notices when one doesn't, and chases the person who owns it.

What it would do

Project

Planned

Keeps track of who owes what by when, chases the updates, and drafts the client status report from what actually happened.

What it would do

Development

Planned

Triages incoming bug reports, chases the detail that's missing, and turns a vague complaint into something a developer can act on.

What it would do

Marketing

Planned

Drafts the campaigns, the posts and the follow-ups in your voice, and reports on which of them did anything.

What it would do

Your team next?

Teams are built around a job a real department does. Tell us which one you'd hand over first and it goes on the list.

Staying in control

One queue. Everything waiting on you.

You see the actual message, not a reference number.

Agent draftsemail or WhatsAppYou decideapprove · edit · rejectSafety checksopt-outs · limits · timingRecipientnothing moves past here on its own

Scroll sideways to see the whole thing.

Approve

It sends as written, and the run resumes.

Change it first

Edit the wording. Your version is what sends.

Say no

Reject with a reason the next run can see.

500 people doesn't mean 500 decisions. When one message goes to a whole group it arrives as a single request, with the recipient count up front.

Nothing hidden

Written for you, not for a developer

Every run reads back in three plain-language parts.

Guardrails

The rules it can't talk itself out of

Enforced outside the agents entirely, so they don't depend on one making a good decision.

Opt-outs are never negotiable

An unsubscribe is honoured the instant it arrives. It never waits behind an approval, and no setting can override it.

Approval is not a bypass

Approving means “this is the right message”, not “skip the checks”. They all still run, and can still refuse the send.

Every workspace is sealed

One client's leads, conversations and memory are never reachable from another's.

Your spending limit holds

Every finished run is priced and counted, failures included. Hit your monthly ceiling and Switchboard acts on it. No agent gets to talk its way past the number you set.

Pricing

Bring your own key

You hold the model accounts and pay the providers directly. We're not in the middle of your token bill.

Your keys, your account

Paste your own provider keys in Settings. Every model call is billed to you, directly, at whatever rate you already pay.

No markup, no metering

We don't resell tokens, count messages, or charge per lead. There's no margin for us in your agents being chatty, and you can watch the spend yourself, per client, per run.

Swap providers whenever

Add a key, and that provider becomes routable. Prices move fast in this market, so you shouldn't be locked to last year's bill.

Two ways to run it

Self-hosted

Runs on a server you own.

  • One server, brought up with a single command
  • Your database, your backups, your network
  • Client data and credentials never leave your infrastructure
  • Your model keys, billed to you by the provider

Managed

We run the servers, you run the agents.

  • Hosting, upgrades and monitoring handled
  • Same product, same approval gates, same isolation
  • Still your model keys, and we never resell tokens
  • Move to self-hosted later without changing how you work

The platform itself is quoted per engagement, based on which teams you run and whether you host it yourself. There are no seat fees, no per-lead fees and no cut of your model spend. We'll go through the numbers with you when your name comes up on the waitlist.

Put the busywork on autopilot.Keep your hand on the wheel.

The sales team is built and working. We're opening it to a first group of operators. Put your name down and we'll come to you when there's room.