InnarisAnswers your technicians can stand behind.
Innaris reads your manuals, drawings and your team's verified field experience, and answers with citations, a verification verdict and a confidence score — on a laptop, tablet or phone, offline in the plant room and synced to your company's Central Hub when you're back online.
How it works
A knowledge graph of your own corpus, a retrieval pipeline that has to prove relevance, and a delivery gate that verifies every answer before it reaches the screen.
Ingest everything
PDF manuals, SOPs, work orders, spreadsheets, photos. Figures inside documents are OCR'd and stored with their caption and surrounding page text, so nameplates and diagrams are searchable and citable.
Retrieve with proof
Filename similarity, exact keyword grep, embedding similarity and knowledge-graph relations — a passage only enters an answer when its own content meets the question's threshold.
Generate, then verify
The model drafts from the retrieved evidence only. An NLI verifier checks each claim against the cited passages; the answer is delivered, marked inferred, or withheld.
Capture experience
Technicians file what actually fixed it. Submissions are signed on the device, reviewed by an admin, and become citable knowledge with who, seniority, when and validation attached.
Built for liability, not just convenience
Every answer carries the evidence behind it and an honest number for how much to trust it.
Confidence you can audit
Each answer shows Verified ✓ · NN% confidence. The percentage is computed from the verifier verdict, per-claim entailment coverage, retrieval tier, evidence count and validated experience — and the tooltip lists every factor.
Grounded vs. inferred, always labelled
No evidence in your corpus? The answer is marked inferred and capped at low confidence. Alert-linked diagnoses with no evidence are withheld until an operator acknowledges.
Signed, tamper-evident
Experience submissions are Ed25519-signed and chained; the store carries a Merkle manifest. The hub refuses submissions whose signature doesn't verify.
Your corpus stays yours
Documents live on the device or your own Central Hub. Online mode sends only the question plus the sanitized retrieved context to the model — never your library.
Central Hub: one master knowledge graph per company
Every install is a node. Nodes work offline and sync to the company's Central Hub when online, so every technician's verified experience keeps building the same master graph.
Stays local until verified
A technician's experience remains on their device and in the local review queue. It's pushed to the hub inbox as pending and only joins the master graph after a company admin verifies it.
Everyone gets everyone's fixes
Nodes pull the master corpus, figures and every verified experience — with provenance — into their local store, cursor by cursor, even after weeks offline.
Users, devices, keys
Admins issue a client key per user and device, revoke it any time, and see last-seen and sync state for the whole fleet.
Online models on your account
The same client key authorises online model calls through the hub relay (Amazon Bedrock / Anthropic / OpenAI / Gemini), metered per company with a spend cap.
Personal or enterprise
Same application, two ways to run it.
Personal
One device, your own documents, local model. Nothing leaves the machine unless you enable online mode.
- Offline answers with citations and confidence
- Ingest PDFs, images (OCR), spreadsheets, notes
- Your own experience library
- Install as an app on macOS, Windows, Linux, iOS, Android
Enterprise
A Central Hub for your company: master corpus, verified team experience, users and keys, online models on your cloud account.
- Everything in Personal, on every node
- Admin verification queue → master knowledge graph
- Per-user keys, fleet view, audit trail
- Hosted on your AWS (EC2 + Bedrock) or by Innaris
Download Innaris
We detected your device. Desktop installs run the local model on your hardware; phones and tablets install the web app and talk to your node or Central Hub.
macOS
Runs the local model on Metal. Install as an app from Safari (File → Add to Dock) or Chrome (install icon).
git clone the repository, then `bash start_mac.sh` (creates the venv, installs requirements, starts on http://localhost:3131)Windows
CUDA offload when an NVIDIA GPU is present; CPU otherwise. Install as an app from Edge or Chrome (install icon).
git clone the repository, then `powershell -ExecutionPolicy Bypass -File start_windows.ps1`Linux
CUDA/ROCm or CPU. Install as an app from Chrome/Chromium (install icon).
git clone the repository, then `bash start_linux.sh`iPhone & iPad
Open your Innaris node or Central Hub address in Safari, tap Share → Add to Home Screen. Runs full-screen with offline shell; answers come from your node/hub.
Android
Open your Innaris address in Chrome and tap Install (or ⋮ → Install app). Works as a standalone app.
ChromeOS & other browsers
Use the address-bar install icon to pin Innaris as an app.
.env as INNARIS_HUB_URL and INNARIS_HUB_CLIENT_KEY.