For messaging teams in private beta

AI that learnsyour voice.Suggests three. You approve one.

A copilot for high-volume DM operators. Paste any incoming message, get three on-tone reply variants in 200 milliseconds. Every approval and reject becomes training data for that voice. Nobody else has it.

Mac, Windows, Chrome Bring your own LLM Human approves every send
smoothtalk
A
incoming, from clipboard
what would you do if I flew out to see you next weekend?
persona, lily · voice match 94% · 41 approvals
click, copies, paste anywhere
copied A
+1 approvaltrains the next reply
⌗ 01, the problem

Generic AI replies sound like a chatbot because they are one.

Most AI reply tools share one persona across every conversation, then charge a flat fee whether you use it or not. Your customers can tell. Your operators override every suggestion anyway. The thing you are paying for is autocomplete you do not actually trust.

Voice fidelity
smoothtalkPer-operator, learned from approvals
othersOne generic voice across the roster
How it improves
smoothtalkEvery approve and reject reshapes the next prompt
othersNightly batch, or never
Fan memory
smoothtalkTracks who they are and what they like
othersStateless. Same opener every time
Where the model lives
smoothtalkBring your own key, any provider
othersLocked to one vendor, marked up
Compliance posture
smoothtalkSuggests. The human still hits send
othersAuto-reply by default, audits later
⌗ 02, how it works

Three keystrokes between message and reply.

No new tab, no context switch, no “open the dashboard.” smoothtalk lives behind a keyboard shortcut on every machine your team already uses. Operators stay in their workflow. The model quietly learns from every click.

01any chat tool

Paste the message

From any chat tool your team already uses. The desktop helper and Chrome extension both read your clipboard or active text field.

clipboard ←
02200ms

Three variants, ranked

smoothtalk pulls the right voice profile and the conversation history, then drafts three replies tagged A, B, C in roughly two hundred milliseconds.

++A
03human in the loop

Approve, edit, or reject

Hit return on the one that fits. Approval logs to that voice. Rejections train against. Every send is something a human signed off on.

return ↵
⌗ 03, what makes it different

Six edges over the flat-fee crowd.

voice

Voice fidelity, not personality cosplay.

Each persona gets its own profile with banned words, mannerisms, signature phrasings, and tone. Drafts that sound like the actual person, not a generic helpful assistant.

memory

Per-fan memory across sessions.

smoothtalk tracks who the person is, what they have said before, what they spent, what they responded to. Replies stop opening with the same line every time.

loop

Every approval makes it sharper.

Approvals and rejections become preference pairs the same hour they happen. The next reply is shaped by what your team said yes and no to today.

control

The human stays in charge.

smoothtalk only suggests. Nothing is sent without an operator hitting return. Compliance teams and platform terms are happier when a person is the final author.

anywhere
A

Mac, Windows, and the browser.

A native desktop helper plus a Chrome extension that pastes into the field you are already in. One shortcut, every chat tool, no rebuild of your stack.

byosk_

Bring your own model.

Claude, OpenAI, Grok, Llama through OpenRouter, your own fine-tune. We do not bundle the model into the price, so you keep cost and provider control.

⌗ 04, the data flywheel

The product that gets better while you use it.

Most AI tools are static. They start good and stay good. smoothtalk is the opposite. The voice profile sharpens with every approval, edit, and reject. By month two it sounds more like the person than they do on a tired day.

01stage

Suggest

Three reply variants, drawn from the persona profile and the conversation history.

02stage

Approve

Operator picks one, edits if needed, and hits return. Approval and edit-distance are both logged.

03stage

Pair

The chosen reply pairs against the two rejected ones to form a preference triple per persona.

04stage

Reweight

Persona prompt is updated within the hour. The next suggestion uses the fresh weights.

loops every send. four stages, no batch jobs.
voice match score, per persona
live
68%day 1
84%day 14
94%day 60

In the closed beta, voice match passes 90% within roughly six weeks at a normal volume of about 600 messages per operator per day.

⌗ 05, pricing

Honest pricing, your own LLM bill.

One transparent seat fee. Bring your own model key, you pay only what the provider charges. The training data, voice profiles, and fan memory stay yours forever.

Starter
$29/seat /mo

For one operator getting started. The full product, smaller training budget.

  • One operator seat
  • Up to two voice profiles
  • Mac, Windows, Chrome extension
  • Bring your own LLM key
  • Per-fan memory, last 30 days
  • Approval-driven training, weekly
Start the trial
most chosen
Team
$99/mo, 5 seats

For small messaging teams. Unlimited voices, shared persona library, hourly retraining.

  • Five operator seats included
  • Unlimited voice profiles
  • Shared persona library, role-based
  • Per-fan memory, unlimited window
  • Approval-driven training, hourly
  • Editable banned-word and tone rules
  • Export DPO-format training pairs
Start the team trial
Enterprise
Talkto us

For larger ops, on-prem or VPC needs, custom fine-tunes, and audit requirements.

  • Unlimited seats and personas
  • On-prem or VPC deployment
  • Custom fine-tuned base model
  • SSO, audit logs, SOC2 mapping
  • Dedicated solution engineer
  • Priority training pipeline
Get in touch
14 day free trial on every planno card to startcancel from the desktop app
⌗ 06, frequent questions

Five things people ask first.

More questions before you trust this with your inbox? hello@smoothtalk.ai answers within a day.

Messaging teams who handle high volumes of conversational DMs in someone else's voice. Customer support, dating, sales SDRs, creator support, virtual assistants. Anywhere a human is replying inside a chat all day and consistency of tone matters.