Implementation BlueprintExecution layer

Szept Voice-to-Agent Delivery Sprint (5-7 days)

A five to seven day engagement that installs Szept on a client's Mac fleet, wires the Soniox speech key and DeepSeek cleanup endpoint, and trains developers to dictate, annotate, and hand off context to Claude Code or Cursor. Time: 5-7 days.

By InnovaAI ResearchPublished

AI Code Tools

How do you implement it?

Blueprint

Szept Voice-to-Agent Delivery Sprint (5-7 days)

A five to seven day engagement that installs Szept on a client's Mac fleet, wires the Soniox speech key and DeepSeek cleanup endpoint, and trains developers to dictate, annotate, and hand off context to Claude Code or Cursor.

Prerequisites
  • Client team runs macOS on every machine that will use Szept, since the app is Mac-only
  • A Soniox API key available before first-run setup, because Szept requires it for cloud speech recognition
  • One named internal owner who will hold the Team license and manage the ten named seats
  • Access to an OpenAI-compatible cleanup endpoint (DeepSeek by default) for filler-word removal and punctuation fixes
  • At least one active Claude Code or Cursor workflow the client wants to accelerate
Execution Timeline
  • 1.Install the notarized Szept Mac app on the pilot machine and complete first-run setup
  • 2.Supply the Soniox API key and confirm dictation lands at the cursor in any app
  • 3.Point the AI cleanup endpoint at the client's OpenAI-compatible service and test filler-word removal
  • 1.Verify hold-fn dictation across the client's three most-used apps
  • 2.Test 60+ language support and mid-sentence switching against the client's actual working languages
  • 3.Confirm the searchable local history captures dictations, screenshots, and recordings
  • 1.Configure Command-Shift-A screenshot capture and annotation for the client's review workflow
  • 2.Set up screen region recording with audio and camera for bug reproduction clips
  • 3.Map one recurring developer workflow onto Szept's four core functions
  • 1.Wire the agent handoff so annotated context pastes into Claude Code or Cursor with voice instructions
  • 2.Run a live test where a dictated instruction plus screenshot produces a testable agent task
  • 3.Document the handoff sequence as a one-page SOP for the client's developers
  • 1.Deploy the MDM permissions profile and preset settings across remaining Macs
  • 2.Apply the install notes from the Team MDM deployment kit to each machine
  • 3.Confirm the ten named seats are assigned to the correct people under one invoice
  • 1.Run a 90-minute training session on dictation, cleanup, capture, and agent handoff
  • 2.Have each developer complete one end-to-end task using only Szept input
  • 3.Collect friction points and adjust preset settings before sign-off
  • 1.Hand over the SOP, the agent kit (CLAUDE.md, specs, plans, tests), and the private Swift source repo
  • 2.Walk the client through the twelve-month update window and the $9.99 a year renewal decision
  • 3.Close with a written recommendation on whether to extend Szept to non-technical staff
$199 Team license (ten named people, one invoice) plus $1,350 to $2,400 in agency delivery time, depending on fleet size5-7 days
ROI Logic

The Team license costs $199 one-time for ten named people, so a $1,350 to $2,400 delivery fee returns roughly 6x to 12x on the software line before any retainer. The margin holds because Szept's setup is configuration and training rather than custom development, and the twelve-month update window means no recurring license cost lands on the agency during the engagement. Resale is limited to clients with Mac-based technical teams, so this offer works best as an add-on to an existing dev retainer rather than a standalone product.

Deliverables
  • Configured Szept install across the client's Mac fleet with Soniox key and DeepSeek cleanup endpoint live
  • MDM deployment package built from the Team kit: permissions profile, preset settings, install notes
  • One-page SOP covering dictation, Command-Shift-A annotation, screen recording, and Claude Code or Cursor handoff
  • Training recording and a completed end-to-end agent task per developer as proof of adoption
  • Written recommendation on extending Szept beyond the ten named seats
Definition of Done

Every named developer has completed one real task using only Szept dictation, screenshot annotation, and agent handoff, with the SOP signed off by the client's internal owner.

Open the Szept courseSzept decision recordBrowse the Academy

More on Szept

More for AI Code Tools