Free, open-source learning that runs on every device — you own your credentials and your data.
This software is in active development. It is NOT production-ready. Do not use it with real credentials, real funds, or sensitive data.
- Learn for free, offline. Courses, tutorials and interactive lessons, plus skill assessments graded on the device. See skills and reputation.
- Earn portable credentials. W3C Verifiable Credentials you store, export and verify offline. An issuer can revoke or suspend only the credentials it issued.
- Keep your data on your device. No accounts and no data collection: each profile has its own encrypted database and vault.
- Peer-to-peer. A libp2p mesh plus iroh for content and live media, with no application server. Relays help peers connect but cannot read traffic. See architecture.
- One device, many learners. Isolated profiles for households, classrooms and shared devices, with guardian links for minors. See multi-user profiles.
- Assessment integrity on the device. Sentinel fuses behavioural and environment signals locally; raw signals never leave the device unless the learner releases them to contest a flag.
- Every platform, nine languages. macOS, Windows, Linux, iOS and Android, in 9 languages including right-to-left Urdu.
You need Rust 1.91+, Node.js 24.22.1 (see .nvmrc), the Tauri CLI
(cargo install tauri-cli), and the native packages for your OS listed in
building.
git clone git@github.com:ifftu-dev/alexandria.git
cd alexandria
npm ci
npm run dev # terminal 1: Vite dev server
cargo tauri dev --features tutoring-video # terminal 2 (Linux: tutoring-video-static)cargo tauri dev does not start Vite for you, so keep npm run dev running.
On first launch the app walks you through creating a profile; see the
user guide. For iOS and Android, see
building for mobile.
Everything listed above runs in the alpha today. This is what is missing. No institution has deployed Alexandria, nobody has been hired on a credential it issued, and no external user has stress-tested any of it.
| Area | Where it stands |
|---|---|
| On-chain anchoring | The anchor queue is wired and submits real transactions, but does nothing unless a Blockfrost project id and a funded wallet are configured. No shipped build anchors by default. Validators are on preprod testnet only. |
| Network | New peers are found through the relays in the bundled preprod network profile; DNS seeds and user-pinned bootstrap lists are post-launch. Existing credentials and content survive relays disappearing; discovering new peers does not. Only DHT provider-record keys use the /alexandria/preprod namespace; gossip and request-response protocols keep the unscoped /alexandria/... paths until the app, relay and monitoring switch together. |
| Sentinel | The false-positive rate is measured against a synthetic holdout only; there is no real-world validity study. Android Play Integrity verdicts need the Play Console response keys in the network profile (null in preprod), so Android sessions stay local (runbook Procedure 5). Federated learning is design-phase (sentinel-federation). |
| Plugins | Practical-trade submissions are reviewed by an instructor on the same node; routing to a remote assessor is a later phase. music-reviews has no grader yet, so it cannot produce a score a credential is issued from. |
| Credential trust policy | Credentials from any authentic issuer stay visible. Field-opinion posting requires a credential that satisfies the subject qualification policy pinned in the network profile; self-claims never qualify. Instructor endorsement requests and responses are exchanged as JSON by hand. |
| Push notifications | In progress, using Firebase Cloud Messaging on iOS and Android. |
| Independent security audit | Not commissioned. Internal audits: Pass 1 remediation verified 2026-03-23; Pass 2 completed 2026-08-18 (report). |
| Organisations and governance | SSO, ATS delivery, bulk verification, cohort analytics and the employer console are under test in a separate, proprietary repository; none of it is run by anyone outside the project or purchasable, and SIS/SCIM provisioning has not started (enterprise boundary). Inbound governance changes are rejected until committee outcome certificates exist. The non-profit veto over free learner access in vision has not been drafted. |
Subsystem docs carry their own scope and status, for example the interview assistant.
The full, annotated list is the documentation index.
| For | Start with |
|---|---|
| Using the app | User guide · Demo setup · Optional AI checks |
| Building and contributing | Building · Mobile · Developer CLI · Contributing · Releasing |
| How it works | Architecture · Protocol specification · Skills and reputation · Sentinel · Plugins · Network profiles |
| Reference | Database schema · Project structure · Settings · Changelog |
| Audits | Security · Performance |
Contributions are welcome under the DCO sign-off; no contributor agreement is
needed. Read CONTRIBUTING.md for licensing, the pre-push
checks (./scripts/check.sh), and the commit message format.
No independent audit has been commissioned yet. The internal audit reports and their remediation status are in docs/audits/.
Alexandria is free software under the GNU Affero General Public License v3 or later. crates/alexandria-verify is MIT OR Apache-2.0 so anyone can embed credential verification. Vendored third-party crates keep their own licences (crates/iroh-moq, crates/moq-media and crates/live are Copyright (C) 2025 N0, INC, MIT OR Apache-2.0).
