Breadcrumb AI Plugin for macOS: Local Context Management and Recording

What Breadcrumb does and why it matters

Breadcrumb records everything you do on your Mac—screens, meetings, and AI sessions—and makes that history searchable for any AI you use. By keeping all data encrypted on‑device, it gives developers a privacy‑preserving way to supply rich context to Claude, ChatGPT, Cursor, and other models without sending raw logs to a third‑party server.

Core features delivered out‑of‑the‑box

Scoped rules automate AI behavior per project

A scope is a folder that carries a set of user‑defined rules.
When you work inside a scoped folder (e.g., ~/code/acme), Breadcrumb automatically applies the associated rules to every AI request.
Example from the UI shows a global scope with 22 rules and a project scope /acme with 28 additional rules.

Persistent, searchable AI memory

Breadcrumb captures screen content, meeting transcripts, and daily activity logs.
Users can query the AI about past events, such as disk‑space issues, stand‑up notes, or yesterday’s work summary, and receive concise answers generated from the recorded context.
The system can continue long conversations by loading prior session metadata, eliminating token waste while preserving full context.

One‑word shortcuts turn reports into commands

After generating a report (e.g., a breakdown of hours spent on a project), Breadcrumb offers to save it as a shortcut like \fb-hours.
Typing the shortcut in any integrated AI instantly reproduces the report, streamlining repetitive queries.

Local‑only processing and encryption

All transcription, speaker identification, and day‑organising models run on the Mac; no data leaves the device.
History is encrypted at rest with a user‑controlled key stored in the macOS Keychain.
A 12‑word BIP‑39 recovery phrase secures the encryption key; the developers never see it.
Users can pause recording from the menu bar, and can exclude specific apps, sites, or phrases from being logged.

Privacy and security claims

"Everything stays on your Mac, encrypted with your key. When your AI asks Breadcrumb something, the answer goes to that AI just like anything you type. Use a local AI and nothing leaves. Use Claude or ChatGPT and their privacy policy applies."

No telemetry or voice‑print collection; speaker naming is derived solely from on‑screen information.
The app performs a single daily update check without transmitting identifying details (verified with Little Snitch).
Users retain full control: they can stop recording instantly, and the app does not require an account, email, or subscription.

System requirements and availability

macOS 15 or later on Apple Silicon (M1+), with at least 16 GB RAM.
Free download from the official site; the free version is guaranteed to remain functional regardless of the company's future funding.
Future roadmap includes support for Google Meet, Microsoft Teams, Windows, and Linux.

Community feedback from Hacker News

Positive ROI focus: One commenter notes that Breadcrumb feels like “upgrading to a better model and a better brain simultaneously,” highlighting its value in augmenting AI assistants.
Feature wish‑list: A user compares Breadcrumb to the now‑defunct Windows Recall, indicating strong demand for cross‑platform recall tools.
Security curiosity: A question about security mentions similarity to Microsoft’s Recall feature, suggesting users are scrutinizing the encryption model.
Technical concerns: A comment asks whether all 30+ integrated tools load on every request, hinting at potential performance overhead.

Business model and funding path

The personal version is free forever, with no hidden fees.
Innerloop plans to monetize via a paid “founding seat” for teams ($299, limited to 1024 seats), which includes early‑access benefits and a refundable credit toward future team licenses.

Bottom line

Breadcrumb provides a locally‑hosted, encrypted context manager that records a developer’s macOS activity and feeds it to multiple AI assistants, enabling richer, token‑efficient interactions while preserving privacy. Its scoped rule system, shortcut creation, and zero‑trust architecture make it a compelling tool for developers who rely heavily on AI‑driven workflows.