The privacy story

Proof,
not promises.

Most privacy pages ask you to trust a policy. Spike Local's model is simpler: your data physically stays on your machine, and the app shows you — live — every route it could ever take.

The ledger

What stays. What could leave. Who decides.

Where it livesWhen it could leave
Your files & indexYour diskNever — the index is never uploaded
The knowledge graphYour diskNever
Meeting recordingsYour disk, transcribed locallyNever
Model inferenceThis machine — or your own box on your LANOnly if you configure a cloud model, per trust zones
Web search queriesOnly the search terms, only when you use web features
Telemetry & analyticsDoesn't existNothing to send
Trust zones

If you ever add a cloud model, you set the rules first.

Cloud is strictly opt-in — your own API key, your own provider. Before a single request can go out, every folder gets a trust tier, and the strictest tier touching a question always wins. Unclassified content fails closed: it stays local.

  • L

    Local only

    This folder's content never reaches a cloud model, no matter what. The default for everything.

  • R

    Redact first

    May be used — after emails, phone numbers, and long ID numbers are masked on your machine.

  • C

    Cloud OK

    You've explicitly marked this folder as safe for your chosen provider.

The receipt

A privacy report generated from the running app.

Settings → Privacy & Proof renders this from the app's actual configuration — not from marketing copy. Add a cloud key and the posture flips to "cloud-capable" and says exactly which routes exist. Route inference to your own workstation and it reports the real address, labeled as your own hardware.

The same honesty runs through the UI: every answer carries a badge — Local, LAN, Web, or Cloud — so you always know where a response was computed, per message, not per policy.

Design principles

Rules the code actually follows.

No account. No login. No telemetry.

Spike has nothing to sign up for. There is no usage analytics, no crash reporter phoning home, no "anonymous" metrics. Even the license check is an offline cryptographic signature — it verifies without sending anything, and keeps working 21 days with no network at all.

Fail closed.

Whenever the app can't classify content — a new folder, an unmatched path — it's treated as local-only. Ambiguity never leaks.

Secrets stay out of backups.

Backup files carry your notes, memory, and settings — never your API keys. A portable file can't leak a credential.

Your own hardware isn't "the cloud."

Pointing inference at your GPU workstation over the LAN keeps everything inside your walls — and Spike labels it as exactly that, a third state between local and cloud, never blurred into either.

Audit it yourself.

Install Spike, open Settings → Privacy & Proof, and read the report. Then unplug the network and watch everything keep working.