EchoBrain 0.1.0

Semantic memory for a person, captured by voice, on DynamoDB vector search.

Download

Apple Silicon (M1–M4) — EchoBrain_0.1.0_aarch64.dmg

Intel — not in this release. The speaker-identification model runs on ONNX
Runtime, and ort publishes no prebuilt binary for x86_64-apple-darwin, so the
build cannot be produced without either dropping that feature or bundling the
runtime by hand. Open an issue if you need an Intel build and it will be done.

First launch will be refused

This build is ad-hoc signed, not notarised — the project has no Apple Developer
certificate. macOS quarantines anything downloaded from the web, so the app will
be refused with "EchoBrain is damaged and can't be opened". That message is
about the missing signature, not about a corrupt download; the checksum below
tells you which.

After dragging the app into Applications:

xattr -dr com.apple.quarantine /Applications/EchoBrain.app

It needs a backend of your own

EchoBrain is a client. The memories live in your AWS account, in the sibling
agent-memory
stack — deploy that first, then point the app at it under Configurações →
Conexão.

Two things are optional and off by default: dictation needs the whisper model
(~570 MB, make model), and answering needs answer_model set. Without it the
app never spends a token.

SHA-256

c8cb8939b3b49aed1d5dae6173543a02e970ff17c79c09ed52a9e01eb04d141c  EchoBrain_0.1.0_aarch64.dmg

Verify with shasum -a 256 EchoBrain_0.1.0_aarch64.dmg.