Asked and answered.
Plain answers, no hedging. If something's missing, it probably belongs on the changelog or in the feature tour.
Does Spike Local really work with no internet?
Yes. The models, the file index, and the memory all live on your machine. Unplug the network and Ask, Documents, Capture, and Brain keep working. Only two features need a connection — web search and any optional cloud model you've added — and both are off until you turn them on.
Where exactly is my data stored?
In a local data folder on your own disk: your file index, the knowledge graph, notes, and settings. Recordings stay where you saved them. Nothing is mirrored to a server — there is no server. Backup & Restore zips this folder into a single portable file you control.
What does the optional cloud mode actually send?
Only what your trust zones allow. You bring your own API key for a provider you choose; folders marked local-only never contribute a byte; redact-first folders have emails, phone numbers, and long ID numbers masked on your machine before anything is sent; and unclassified content fails closed to local. Every cloud-touched answer is labeled with a Cloud badge.
Is there telemetry, analytics, or an account?
No, no, and no. There's nothing to sign up for and nothing phoning home. Even license verification is an offline cryptographic check.
My files are confidential client documents. Is that the intended use?
It's the founding use case. Spike was built against real multi-customer archives, with specific engineering to keep one customer's content from ever leaking into another's answers — and with everything staying on the machine those documents already live on.
What hardware do I need?
Minimum: a 64-bit Windows 10/11 machine with WSL2 and 8 GB RAM — Spike runs CPU-only. Recommended: 16 GB+ and either a GPU or a stronger machine on your network. On a modest CPU, deep answers over large documents genuinely take minutes; that's the physics of running a real model locally, not a bug.
Can I use my own models?
Yes — pull any Ollama model or Hugging Face GGUF from inside Settings, then assign models per job: one for quick chat, one for deep cited answers, one for Thai, one for vision. Defaults are sensible if you'd rather not think about it.
I have a GPU workstation. Can Spike use it?
Yes. The remote model host setting points all local inference at another machine running Ollama — over your LAN, through an authenticated tunnel (API key, custom header, or basic auth), or via Ollama Cloud. Answers served that way carry a LAN badge, and if the host goes down you get a clear error with a one-click "use local for now," never a silent fallback.
Why do some answers take minutes?
Because a several-billion-parameter model is reading your actual documents on your CPU. Quick chat uses a small fast model; deep cited answers use a bigger one. Every long job shows progress and has a Stop button that truly stops the model. A GPU — local or on your LAN — changes minutes to seconds.
Is Thai actually supported, or just translated?
Natively supported. Thai documents are indexed and retrieved with a multilingual embedding model, transcription uses a Thai-tuned speech model, drafting uses a Thai-capable local model, and the routing layer understands Thai questions directly. Mixed Thai–English archives are the normal case here, not an afterthought.
How do I know an answer isn't made up?
Three ways. Answers from your files carry numbered citations you can click through to the source passage. Every chat answer gets an automatic fact-check pass against its own sources, and unverified sentences in generated documents are flagged inline. And the Brain screen lets you audit the underlying facts themselves — each one is a verbatim quote with a file behind it.
What file types does it read?
The office staples: Word (tables included), PDF, Excel, PowerPoint, text, markdown, HTML, and images via local vision models. Audio and video go through Capture for transcription.
Can it browse the web or operate my computer?
Web: yes — it can search, read pages you point it at, and fuse web results with your files, always labeling which is which. Operating your desktop (Autopilot) exists behind explicit permission gates and is honestly labeled a preview: real, supervised, and still rough.
When do macOS and Linux ship?
They're on the waitlist for a reason: we'd rather ship them right than fast. Join the waitlist and you'll get each build the day it's real.
How do updates work?
During the preview, updates ship as new installers announced to the waitlist. In-app updating is being built — the plumbing exists and is being hardened before it's turned on.
Can I move Spike to a new machine?
Yes — Backup & Restore produces one file with your notes, memory, settings, and (optionally) the index. Restore it on the new machine and re-enter any API keys, which deliberately never travel in backups.
What happens to my data if Spike-the-company disappears?
Nothing. The app runs offline on your machine with no server dependency, your files never left, and the free core doesn't expire. That's the point of local-first.
Still curious? Ask it yourself.
The download is free and the first answer takes ten minutes of setup.