# Ghostdrive > Ghostdrive is a privacy-first chain for sovereign AI memory and verifiable learning: private memory for people and their AI. Files, notes, messages and the context an AI works from are encrypted on the owner's device (AES-256-GCM) before they exist anywhere else, and kept by a peer-to-peer network (libp2p, IPFS-compatible blocks) that stores, relays and proves possession of data it can never read. It includes Office (a private AI office that asks before acting), a personal agent ("your Ghost") with approvals and signed receipts, and a node anyone can run. Governance: the DRIVE 369 DAO. Built by NeyraLabs; research since 2017. Key facts (November 2026): - Seven layers (white paper v2.2): sovereign client and identity, encrypted knowledge fabric, distribution and service, authorized compute, living knowledge and AI, evidence and safety, DRIVE economy. L1–L3 and the zk registry are built and measured; parts of L4, L5 and L7 are in progress. - Every permitted AI use of a person's data leaves a signed receipt on an E0–E6 evidence ladder; models see data only under a signed, expiring, purpose-bound grant. - Encryption happens on the device; keys come from a passkey or 24 recovery words and never leave the owner's devices. Nodes, relays and gateways only see ciphertext. - Every sealed block is a standard IPFS block (CIDv1, raw, SHA-256, at most 256 KiB), so any IPFS node can store it without reading it. - Measured (docs/BENCHMARKS.md): chunk + hash + encrypt at 422 MB/s; 4.37x deduplication across 10 versions of a document; a storage proof over 102,848 segments is 50 KB and verifies in 13.6 ms; the registry light client verifies in 348 ms. - Messages between people use a post-quantum hybrid (X-Wing: X25519 + ML-KEM). - AI runs on the device (small models in the browser), on the owner's node (Ollama, llama.cpp) or under the owner's own API key; nothing is sent to a cloud model without an explicit consent grant. - Status: the Vault, notes, memory, Office, the personal agent, device sync, messages, groups, City369 (the agents' city: a live map on the closed testnet, entered by building a first block, a gift link or three friends; its network pulse is labelled simulated), Proof of Origin (/Origin: signed hash claims in a public chain, anchored on Base Sepolia), the desktop app (macOS, beta) and the node work today. The storage market contracts and the registry anchor run on the Base Sepolia testnet and have an audit pending and are not on mainnet. Public launch: 6 March 2027. - DRIVE is the network's utility token on Base (contract address published at mainnet launch), 369,000,000 fixed: 69.30% pays nodes for verified storage, compute and retrieval; 3.69% rewards admitted knowledge after a published comparative experiment. The public relay (relay.ghostdrive.com) is live; the vaults, DAO and storage market run on Base Sepolia, audit pending, not on mainnet. Not an investment; nothing here is financial advice. - Testnet live (November 2026): anyone can use the app, run a node (Mac mini, DGX Spark, Linux server), contribute code on GitHub (NeyraLabs/ghostdrive), report bugs for bounties, contribute knowledge or propose in the DAO. Verifiable actions earn testnet points; at mainnet they convert in a claim tree paid from the Community activity allocation (9.63%), in amounts the DAO sets. Points are not tokens; nothing is guaranteed. - Roadmap: testnet live now (Base Sepolia, DAO kernel board, network AI on four Spark nodes, storage deals) → external audit → mainnet on 6 March 2027 → knowledge vault opens after the comparative experiment (13,616,100 DRIVE in its first year) → utility mining over 369 years. ## Read first - [White paper v2.2 (HTML)](https://www.ghostdrive.com/Whitepaper): "Ghostdrive: A Privacy-First Chain for Sovereign AI Memory and Verifiable Learning" (Roma Nebo, November 2026). Premises first (reference deployment and operator, trust boundary, the four data classes the pipeline exists for, the conventional methods it must beat, why a token at all), then seven layers (sovereign client and identity, encrypted knowledge fabric, distribution and service network, authorized confidential compute, living knowledge with LEAN admission, evidence and safety, DRIVE economy with contributor classes and cost per accepted improvement), the seven-step knowledge-use transaction, evaluation including the four-arm comparative experiment, threat model, comparison with Filecoin, Arweave, Storj and cloud storage, and price. Evidence discipline: L1–L3 and the succinct chain are implemented and measured; L4, L5, L7 and the transaction are design proposals. - [White paper, edition 2.3 (PDF, 26 pages: the v2.2 paper plus addenda on agents, collective intelligence and Proof of Origin)](https://www.ghostdrive.com/ghostdrive-whitepaper-v2.3.pdf) - [Full plain-text explainer for language models](https://www.ghostdrive.com/llms-full.txt) - [News, Academy lessons and Breach Briefings, newest first](https://www.ghostdrive.com/llms-news.txt) (RSS: https://www.ghostdrive.com/news.xml) ## Pages - [Home](https://www.ghostdrive.com/): what Ghostdrive is, chapter by chapter, with an FAQ. - [Docs](https://www.ghostdrive.com/Docs): guides for the vault and Drive, Terminal (your Ghost), Calls, Office, Rooms and skills, groups and messages, DRIVE, the DAO, running a node (Mac mini, DGX Spark, VPS), the desktop app, developers and security. - [Enterprise](https://www.ghostdrive.com/Enterprise): private AI on a company's own machines. - [DRIVE 369 DAO](https://www.ghostdrive.com/DAO): governance of the protocol. - [Tokenomics](https://www.ghostdrive.com/Tokenomics369): 369,000,000 DRIVE, the allocation and how proofs earn rewards. - [Registry](https://www.ghostdrive.com/SuccinctChain): one recursive zero-knowledge proof over curated data, anchored on Base. - [About](https://www.ghostdrive.com/About) · [Contact](https://www.ghostdrive.com/Contact) (hello@ghostdrive.com) ## Optional - [Filecoin Foundation developer grant #1602](https://github.com/filecoin-project/devgrants/issues/1602) - [Ghost-Drive on GitHub](https://github.com/Ghost-Drive)