Short answer: our current dev-board architecture has no Aiden-hosted backend. Longer answer: that doesn't mean no data is ever stored anywhere, and we'd rather explain the actual data path than let you assume either way.
The architecture
Aiden runs on a bring-your-own-provider model. The board captures a connected device's display via HDMI capture, sends screenshots to a multimodal model endpoint you configure, and controls the target device through USB HID (keyboard, pointer, touch). Voice works the same way — audio goes to the STT, LLM, and TTS endpoints you've selected.
What that means concretely:
- No Aiden-hosted backend in the current architecture.
- Screenshots, audio, and text go to your configured endpoints — not to us.
- You can point it at a cloud provider, a self-hosted model, or something fully local.
- Firmware is open source — inspect it yourself.
This is a deliberate split from the usual pattern where the product company, the storage layer, and the model provider are the same entity. We separated those roles on purpose.
Where it gets more nuanced
Aiden also supports persistent context, preferences, and skill optimization. So "no backend" doesn't mean "nothing is ever stored" — a deployment can retain local context. We haven't yet published the exact schema, retention period, or deletion process for that, and we're not going to pretend we have.
Same caveat applies to screenshots specifically: they may leave the board if you've configured a remote endpoint, and we haven't published a complete account of how caching/buffering works or what each provider retains on their end.
What we haven't documented yet (full list, no spin)
| Topic | Status |
|---|---|
| Data retention periods | Not published yet |
| Local memory deletion workflow | Not published yet |
| Data export | Not published yet |
| Default logging/diagnostics | Not published yet |
| Telemetry/analytics use | No public statement yet |
| Storage jurisdictions | Not published yet |
| Encryption/key management | Not published yet |
We're listing this because the absence of a published policy shouldn't be read either way — not as "they're hiding something" and not as "everything's fine." We just haven't written it up yet.
The part that actually matters for builders
Your configuration is doing as much work as the hardware. A cloud endpoint, a self-hosted model, and a fully local model produce three different data-flow outcomes from the exact same physical device. Before you point Aiden at sensitive workflows, check what your chosen model/STT/TTS provider actually does with inputs — that's now part of your privacy surface, not an implementation detail.
Repo: github.com/AidenAI-IO/aiden-firmware
Discord: discord.com/invite/bcJavjcnYz — our engineers answer technical questions directly.
Top comments (0)