RomanLogic voice agents · 19 Nov 2026

The voice brain: what it does, what it fixed, and what moving an agent onto it would take

The voice brain sits between ElevenLabs and the model. It keeps a record of the call as it goes, checks every sentence before the caller hears it, and makes sure a lead reaches the owner even when the caller hangs up early. It has run on the Ambient Heating demo line since 18 Nov, with real handset calls every day.

Built and tested on Rachel's own ElevenLabs account, Anthropic key and server. Nothing on Roman's account has been touched.

What it is

Today an ElevenLabs agent sends each turn straight to its model, and whatever the model says is spoken. With the brain, ElevenLabs sends the turn to the brain instead. The brain answers with the same model, but it knows what the caller has already said and checks the reply before it goes out.

Keeps a recordName, number, address, job and timing, updated every turn from the caller's own words.
Checks each sentenceA question about something already given is swapped for the next one that's needed. About 3 ms a turn.
Checks the numberA read-back that doesn't match the digits the caller said is corrected before it is spoken.
Sends the lead itselfWhen the caller is leaving, the lead goes at once, without waiting on the model.
Catches hang-upsAfter the call ends, if no lead went out, the brain builds one from the record and sends it.

If the brain is down, ElevenLabs falls back to its own hosted model for that turn, so a call never goes silent. Switching an agent on or off is one command, and the agent's settings are saved before every change.

What it fixed

Before any real calls, we ran 36 simulated calls twice through the LDG Joinery prompt: once with the brain only passing the model's words through (what callers get today), and once with the brain checking. Claude played the caller, and every name, number and address was invented.

MeasureToday's setupWith the brain
Calls where a usable number was given but no lead reached the owner5 of 360 of 36
Leads with a wrong or missing name or number80
Calls that asked for something the caller had already said51 (minor)
Time to the agent's first word, as the caller hears it1.10 s1.13 s

Nothing scored worse with the brain on. The extra 0.03 seconds is not something a caller can hear.

Live testing on Ambient

Ambient Heating has been on the brain since 18 Nov. Each round of handset calls turned up something, and each fix went in with a test so it stays fixed.

Examples of what real calls found

The text a caller gets after hanging up early

Before

Here's what I took down: Caller hung up before the end. Job: Yeah, there's a red light flashing. Address: 12 Mill Lane in Barton. Timing: best time to call: Uh, this week.

Now

Here's what I took down: Boiler stopped working, no hot water, a red light flashing. Best time to call: this week. Address: 21 Mill Lane, Barton

The "after" also carries the caller's spoken correction from 12 to 21, which the old lead missed.

Where it stands

Ambient passes all 27 automated tests on the brain. Each automated run costs about 430 ElevenLabs credits, so the full set runs after every change.

What it costs

The brain runs on Claude Haiku 4.5 through Rachel's Anthropic key, on a server that already runs the demo desks, so it adds no hosting cost. All figures are measured, not estimated.

ItemCostWhere the figure comes from
One real call, 1 to 1.5 minutes$0.035 to $0.066Three Ambient handset calls on 19 Nov: every model call the brain made, including reading the caller's details
Per minute of call, roughlyabout $0.04The same three calls
The model on ElevenLabs' side$0.00ElevenLabs bills a call on the brain as "custom LLM" at no charge, so its own model charge for that call goes away; voice and minutes are billed as before
Building and testing all of it this month$16.97The brain's whole Anthropic spend for November so far: simulated call runs, test packs and every live call
Monthly ceiling$20A hard spend cap in the brain: it stops calling the model and lets ElevenLabs' own model take over before going past it (raised to $40 for November only)

For scale: a client line taking 300 calls a month, at about a minute and a half each, comes to roughly $15 to $20 a month in brain model costs. Today the cap is one total across the brain. Giving each client line its own cap is a small addition, so a busy line could never run up an unexpected bill.

One-off cost to move an agent is time, not money: about two to three days for the first one (see below), plus around 430 ElevenLabs credits per automated test run while it's being tuned.

Moving one of Roman's agents

This would only happen if Roman wants it. Logistically it's straightforward: the switch itself takes minutes, and most of the work is teaching the brain that agent's own way of taking a call.

Already built and reusable as it is

The work for each agent

StepEffortWhat it involves
Copy the agentminutesAll work is on the copy; the live agent is untouched until the end
Write its flow½ to 1 dayWhat the agent collects, in what order, and how to spot each answer. The main piece of work
Match its leada few hoursMap the brain's lead onto that agent's own send tool and webhook, if they differ from the demos'
Point the copy at the brainminutesOne setting on the agent, one line in the brain's config
Test and tuneabout 1 dayThe automated pack plus handset calls, the same way Ambient was done
Switch the live agentminutesOne setting, with a one-command rollback

So the first agent is about two to three days. Later ones are quicker, because most of the patterns carry over.

Before a client's calls go through it

What comes next

Figures from the 18 Nov simulated-call results, the Ambient handset calls of 18 and 19 Nov, and the brain's own spend ledger. Every name, number and address in the examples above was made up for the test calls.