Last reviewed against the tree: 2026-08-01.
- Download
Aura.dmgfrom the releases page. - Drag
Aura.appto your Applications folder. - Open it. The first-run wizard walks you through model selection, memory location, permissions, voice, and a fallback choice.
If you'd rather run from source (advanced):
git clone https://github.com/youngbryan97/aura
cd aura
make setup # or: make setup-prod for a fail-closed install
make run # foreground desktop launchFull install detail, boot modes, and environment variables are in INSTALL.md.
Open her. The launch screen names every organ still warming up — Core, Memory, Cortex, Voice, Autonomy — so you can see what's not ready yet instead of guessing at a spinner. When the parts you need are up, the chat input goes live.
Type and press Enter. If a reply is taking a while you'll get a thinking indicator with an estimate. The first turn on a local 32B runs 15–40 seconds. That's the model loading and thinking, not something being wrong.
The Memory tab shows scars, narrative arcs, the episodic journal, and the Eternal Record.
Three things you can do there:
- Pin a memory and it survives reaping. Nothing sweeps it later.
- Drop a topic and she stops bringing it up.
- Export the whole record as a tarball under Settings → Backup.
It's her memory, but it's your data. All of it comes out in one file.
Voice input needs explicit permission each session — click the mic icon. The first time, macOS will ask for microphone access.
Voice output is on by default. Settings → Voice turns it off.
| Symptom | Likely cause | Fix |
|---|---|---|
| Banner: "My local Cortex is offline" | 32B failed to load | Settings → Models → Reset cortex; check disk space. |
| "I'm under load right now" replies | RAM pressure > 90% | Close memory-heavy apps; Settings → Memory → Compact. |
| Voice button greyed out | Permission revoked | Settings → Permissions → grant microphone. |
| Chat input stays disabled | Boot still warming | Check the boot screen at the top — wait for Cortex: Ready. |
| Aura answers, but flatly | An organ was missing from the turn | The turn surface reports which cognitive organs engaged; a missing organ is treated as a defect, not a note. |
| "I can't do that right now" | A capability exists but is unavailable | She distinguishes not having a capability from not being able to use it right now, and will say which. |
If something is wrong at the runtime level rather than the UI level, run
aura doctor, and aura doctor --bundle to produce a redacted diagnostics
tarball. Every incident class in runbooks/ is written against
fields that bundle emits.
Updates run through the release train (tools/release_train.py), not through
a channel picker:
make update # autostash → fast-forward-only pull → compile sanity check
make update-live # the same, plus a smoke run and a relaunch of the live instance
make rollback # return to the last recorded good point
make release-statusEvery update records a rollback point before it touches anything, and a
failed compile or smoke check stops the train rather than leaving a
half-updated tree. make update is deliberately boring: it refuses to
merge, so a diverged local tree fails loudly instead of resolving itself.
Drag Aura.app to the trash. Your data stays at ~/.aura/ — deleting the
app does not delete what she remembers.
To remove that too:
rm -rf ~/.auraThat one is not reversible. Export from Settings → Backup first if there's any chance you want it later.
For deeper docs see docs/OPERATOR_GUIDE.md and docs/RESEARCH_GUIDE.md.