← All use cases
For engineers

For the engineer who’d rather ship than type.

Commits and PRs by voice. Design reviews and standups recorded with on-device Meeting Notes. The chat overlay reading the file you're in. Three tools, one app, all on Apple Foundation Models. Free.

Download for Mac

A field guide for engineers, updated May 4, 2026 by Branden Chong.

In practice

Where Dollop plugs into the engineering loop.

  1. 01.

    Commit messages and PR descriptions.

    Hit ⌥ Space, describe the change the way you'd brief a teammate, paste it as the commit body. Dollop's adapter is tuned to keep technical vocabulary intact, Postgres, gRPC, Tailwind, Anthropic, Vite, instead of phoneticizing them.

    Heard

    refactor the auth middleware to extract the JWT validation into its own service, and switch from synchronous bcrypt to argon2id for the password hash. closes ENG-2841.

  2. 02.

    Record design reviews and standups.

    Hit ⌃ R before the design review or sprint demo. Meeting Notes captures the call (system audio for Zoom, mic for in-person), transcribes locally, and outputs action items you can drop straight into Linear or Jira. No Otter, Fireflies, or Granola; the recording stays on your Mac.

    Heard

    action items: ben to spike the websocket reconnection logic by friday. anu to draft the migration plan for the user_events table. follow up on whether to keep the legacy webhook endpoint.

  3. 03.

    AI chat that reads the file you’re in.

    ⌃ S summons a chat overlay that can read the active window. Ask it to explain the file, generate test cases for the function under your cursor, or rewrite a docstring. Multi-turn, on-device, stays open while you keep coding.

  4. 04.

    Slack and Linear at native speed.

    Per-app tone matching: Dollop knows Slack should sound looser, Linear should be terse and direct, Mail should be a touch formal. Same dictation, three correct voices.

    Heard

    merging the worker pool change once CI goes green. flagged the rollback steps in the PR. ping me if anything looks off.

What you actually get

Dollop mapped to your work.

Voice dictation (⌥ Space)
Commit messages, PR descriptions, code comments, docstrings, Slack/Linear replies, all system-wide in Cursor, VS Code, Zed, Xcode, JetBrains, and the terminal.
Meeting Notes (⌃ R)
Record design reviews, sprint demos, and architecture syncs. Action items drop straight into Linear or Jira from a local transcript.
AI chat reading active file (⌃ S)
Explain a function, generate test cases, rewrite a docstring, all reading what's on screen, locally.
Custom codebase vocabulary
Internal product names, framework names, vendor names, internal acronyms recognized verbatim. "ARR" stays "ARR." "Sora" stays "Sora."
Per-app tone matching
Terse in Linear, looser in Slack, formal in Mail, all from the same voice input.
On-device for proprietary code
No upload of code, comments, or commit messages to a third-party AI vendor. Verifiable with Little Snitch.
A type specimen

How the cleanup reads.

Real before-and-after dictation on terms specific to engineers, processed locally on Apple Foundation Models.

Commit message (technical vocabulary preserved)
Heard

uh refactor the auth middleware to extract the j w t validation into its own service and switch from synchronous bcrypt to argon2id for the password hash. closes e n g 2 8 4 1.

Cleaned

Refactor the auth middleware to extract the JWT validation into its own service, and switch from synchronous bcrypt to argon2id for the password hash. Closes ENG-2841.

Slack reply (casual tone, low ceremony)
Heard

uh merging the worker pool change once c i goes green. flagged the rollback steps in the p r. ping me if anything looks off.

Cleaned

merging the worker pool change once CI goes green. flagged the rollback steps in the PR. ping me if anything looks off.

Why it fits

Why engineers care.

Names recognized verbatim.

Dollop AI is fine-tuned on the names that show up in real engineering work, frameworks, languages, tools, vendors, internal acronyms. "ARR" stays "ARR," not "are are." "Sora" stays "Sora," not "Sora." You can also add a custom vocabulary for your codebase.

0 GB of RAM, no IDE slowdown.

Local Whisper-based tools eat 700 MB to 2 GB of memory. Cloud meeting-note tools (Otter, Granola) are network-bound. Dollop adds nothing in either case, Apple Foundation Models is already in the OS. Your IDE keeps its RAM, your build keeps its CPU.

Audio never leaves your Mac.

If your codebase contains anything sensitive, security work, customer data, unreleased features, cloud dictation and cloud meeting notes are nonstarters. Dollop is on-device by architecture across all three pillars. Verifiable with Little Snitch or Activity Monitor.

Works in every IDE and terminal.

System-wide. Cursor, VS Code, Zed, Xcode, JetBrains, Vim, Emacs, Terminal, iTerm2, Warp. Anywhere there is a text input, ⌥ Space pastes clean text. The chat overlay can read any of them.

Asked & answered

Questions, answered.

Does Dollop work in Cursor and VS Code? +
Yes. Dollop is system-wide, it pastes clean text at your cursor in any app, including Cursor, VS Code, Zed, Xcode, JetBrains IDEs, and the terminal. There is also a dedicated Cursor integration page with extra detail.
Can Dollop record Zoom and Meet calls? +
Yes. Meeting Notes captures system audio (so it picks up your call participants) and your mic, transcribes locally, and produces a summary with action items. No browser plugin, no bot in the call, nothing uploaded.
How does it compare to Otter, Granola, or Fireflies for engineering meetings? +
Otter, Granola, and Fireflies all upload audio to a server for cloud transcription, and most train models on user data unless you opt out. Dollop's Meeting Notes runs entirely on-device on Apple's Neural Engine and Foundation Models. For internal engineering calls (architecture reviews, security work, customer interviews), that's the difference between safe-by-default and requiring a redaction pass.
Can I add custom vocabulary for my codebase? +
Yes. The Custom Vocabulary panel lets you add internal product names, framework names, and acronyms. Dollop's adapter learns them on top of Apple Foundation Models without retraining anything cloud-side.
Is it private enough for proprietary code? +
Yes. Speech recognition runs on Apple's Neural Engine, cleanup runs on Apple Foundation Models on-device. There is no network call in the dictation, chat, or Meeting Notes paths. Full privacy posture here.
What about latency? +
Dictation: first token in tens of milliseconds, full clean response in under a second for typical commit-length dictations. Meeting Notes: transcript appears in real time during recording, summary in seconds after stop. Comparable to raw Apple FM, faster than any cloud round-trip.
Hardware requirements? +
Apple Silicon (M1+) on macOS 26 with Apple Intelligence enabled. Intel Macs are not supported.

Your IDE has autocomplete. Your terminal has tab completion. Your prose, your standups, and your design reviews should keep up. Download Dollop and ship the next PR description in the time it takes to make coffee.

Download for Mac
Dollop

The voice OS for your Mac. Built with care in California.

© Dollop 2026