Tally
Early access
01
Cloud · provisional Open source

Know what every voice-agent call costs you.

Tally adds up your transcription, LLM, voice and phone bills for each call, and shows what each customer really costs you and what you earn on them.

Join the Cloud waitlist
Margin per customer Synthetic data
  • Harbor HVAC−10.4%
    Cost$63.07 Revenue$57.11
  • Lumen Realty87.9%
    Cost$37.37 Revenue$309.63
  • Northside Dental78.6%
    Cost$26.98 Revenue$125.90
02

Four invoices, zero answers

A voice agent bills you in minutes, tokens, characters and carrier minutes, from four or five vendors, on different cycles. Your customers pay you per minute or per month. Nothing connects the two.

From a dev on Reddit

"Why did that call cost $2?"

One developer was surprised that a 10-minute call came to $1.50–2.00 once STT, LLM, TTS and telephony were added up.

From a self-hosted LiveKit team

"Which customer is expensive?"

A team running LiveKit on its own servers, paying three provider invoices, asked exactly that. No single bill can answer it.

From several teams

Spreadsheet archaeology

Teams wire up Postgres and Grafana by hand, count TTS characters one by one, or open every provider dashboard and match timestamps.

03

Who it’s for

Anyone paying several voice vendors and charging customers for the result.

Agencies

You pay four vendors and bill per client

Know what every call cost and what every client earns you before you renew a contract.

Multi-tenant voice SaaS

You charge per minute or per seat

See which customers cost more than they pay, and price the next plan on real numbers.

Small self-hosting teams

Three invoices, no total

Cost per call from the metrics LiveKit or Pipecat already emit, reconciled against each invoice.

Most useful between 10,000 and 500,000 minutes a month.

04

What it does

Open source, Python standard library only, runs next to your agent with SQLite.

  1. Collect

    Drop in a collector

    Three lines for LiveKit Agents or Pipecat. It reads the usage metrics the framework already produces: audio seconds, tokens, characters, minutes.

  2. Price

    Measured beats estimated

    Default rates per provider and model, your own rates on top. Real costs reported by a provider replace the estimate. Unknown models are flagged, never counted as $0.

  3. Decide

    Margin per customer

    Cost per call, per customer and per agent next to what you charge. Reconcile against each invoice to see the money no call explains.

05

Try it in two minutes

No accounts, no API keys. The demo writes 30 days of sample calls for three fictional customers.

Demo and LiveKit agent (Python)
# install and run the demo
pip install -e .            # from the repo: source code coming soon
tally demo
tally report --by customer
tally serve              # dashboard on http://127.0.0.1:8787

# in your LiveKit agent
from tally.collectors.livekit import attach
attach(session, HttpSender(), call_id=ctx.room.name, customer_id="acme",
       defaults={"stt": "deepgram/nova-3", "llm": "openai/gpt-4o-mini", "tts": "cartesia/sonic-3"})
Customer (demo data)CallsMinutesCost$/minRevenueMargin
Harbor HVAC141714$63.07$0.0884$57.11−10.4%
Lumen Realty193398$37.37$0.0939$309.6387.9%
Northside Dental259699$26.98$0.0386$125.9078.6%

Synthetic demo data, fictional customers.

06

Pricing

Provisional launch pricing. It may change before general availability.

Open source Free

Self-hosted, Apache-2.0. Collectors, price table, dashboard, reports, invoice reconciliation.

Cloud · provisional 5% of tracked spend

Free up to 10,000 minutes a month, then $49/month minimum, capped at $499/month.

  • Hosted dashboard and retention
  • Team access
  • Provider cost import (roadmap)

Default provider prices are a starting point; check them against your own contracts.

07

Join the Cloud waitlist

We are looking for 2–3 agencies or multi-tenant voice SaaS teams to try it on their own calls.

We only use your email to contact you about this. Privacy.