Your machine
remembers.
Spike Local is a desktop assistant that reads your files, builds a private memory of your work, and answers with citations you can check — entirely on your own computer. Not a byte leaves.
The renewal locks a 24-month term at the current rate, with a 60-day notice window before auto-renewal 1. Support hours widen to 8×5 with a 4-hour response target for critical issues 2, and the annual review moves to each January 13.
Cloud assistants are brilliant strangers — you re-explain everything, and your files go to someone else's computer. Spike flips it: the intelligence moves to where your data already lives, learns your world once, and proves every claim with a citation.
One assistant, your whole desk.
Five surfaces on one private memory. Everything below ships in the current build.
Cited answers from your files.
Ask across everything you've indexed — contracts, proposals, minutes, spreadsheets. Spike retrieves the real passages and answers with numbered citations you can open, in English or Thai.
Drafts that cite themselves.
Long documents chapter by chapter, grounded in your files — each chapter fact-checked before you see it. Export to PDF.
Meetings become minutes.
Record or drop in audio — Spike transcribes locally (Thai included) and writes structured minutes with action items.
A memory you can read.
Spike extracts facts from your files into a knowledge graph — each with its source. Confirm what's right, fix what's wrong, see conflicts surfaced.
Everyday questions, honestly routed.
Quick questions stay local. When something needs the live web, Spike searches, fuses results with your files, and labels exactly where the answer came from. Images? It reads those too. Every answer gets an automatic fact-check pass against its own sources.
Bring your own horsepower.
Pull any Ollama or Hugging Face GGUF model and assign it per job — one for chat, one for deep answers, one for Thai. Got a GPU workstation? Point Spike at it over your LAN, through an authenticated tunnel, or via Ollama Cloud — and the badge on every answer tells you honestly where it ran.
Privacy you can audit, not just believe.
Every privacy claim in Spike is inspectable in the app. The privacy report lists every route your data can take. If you add an optional cloud model, trust zones decide — folder by folder — what may leave, and redaction masks personal data before anything does.
- ✓
Local by default
Files, index, memory, and models live on your disk. Offline? Everything still works.
- ✓
Trust zones
Mark folders local-only, redact-first, or cloud-OK. The strictest rule always wins.
- ✓
No account, no telemetry
Nothing to sign up for. Even the license check works offline.
From install to "how did it know that?"
Install Spike Local
One installer. Spike provisions its own local engine and models, with a system check that flags anything missing.
Point it at your world
Pick the folders that matter — customers, projects, cases. Spike indexes and organizes them into areas.
It builds your brain
Files become a searchable memory plus a fact graph you can inspect, confirm, and correct.
Ask, draft, capture
Cited answers, grounded documents, meeting minutes — all from your own knowledge, all on your own machine.
Different species, not a better copy.
| Spike Local | Cloud assistants | |
|---|---|---|
| Where your files go | Nowhere — indexed on your disk | Uploaded to their servers |
| Works offline | Fully | No |
| Answers grounded in | Your documents, with citations | Training data + what you paste |
| Memory | A fact graph you can read and correct | Opaque, if any |
| Model choice | Any local model — or your own GPU box | Theirs only |
| Account required | None | Always |
Fair is fair: frontier cloud models are stronger at open-ended reasoning. Spike can optionally use one — with your key, under your trust zones — when you decide a question deserves it.
The three everyone asks.
Does it really work with no internet?
Yes. Models, index, and memory are all on your machine — unplug the network and Ask, Documents, Capture, and Brain keep working. Only the optional web-search and optional cloud-model features need a connection, and both are off by default.
What hardware do I need?
It runs CPU-only on an ordinary 8 GB machine. More RAM or a GPU means bigger models and faster answers — and if your fast machine is a different box, Spike can use it over your LAN.
Is Thai actually supported, or just translated?
Supported natively: Thai documents are indexed and retrieved with a multilingual model, transcription uses a Thai-tuned speech model, and drafting uses a Thai-capable local model. Mixed Thai–English archives are the normal case, not an edge case.
Give your machine a memory.
Download the Windows preview today, or join the waitlist and we'll send the macOS and Linux builds the moment they land.