Spacedrive

Open-source file management and data indexing, honestly reviewed. Including the part where the v1 got abandoned and the team started over.

TL;DR

• What it is: Open-source (AGPL-3.0) local-first data engine — indexes files across all your devices and storage into a single searchable interface, powered by a virtual distributed filesystem written in Rust [1][2].
• Who it’s for: Privacy-conscious founders, creators, and power users with files scattered across multiple devices, NAS drives, and cloud services who want unified search without surrendering their data to Google or Dropbox [4].
• Cost savings: Self-hosted is free. Google Workspace Business Starter runs $6/user/month; Dropbox Business $15/user/month. Spacedrive’s self-hosted server runs on any Linux box or NAS at the cost of electricity [3].
• Key strength: The architecture is genuinely ambitious — content-addressed deduplication via BLAKE3, semantic search via LanceDB vector similarity, peer-to-peer sync without a central server, and AI integration that runs locally [2][4].
• Key weakness: As of this review, v2.0.0-alpha.1 shipped December 26, 2025. V1 development was halted in January 2025, and v2 is a ground-up rewrite. Windows support isn’t available yet. If you need something that works today without babysitting, this is not that tool.

---

What is Spacedrive

Spacedrive started in 2021 as a cross-platform file manager — the kind of thing that hit #1 on GitHub Trending, raised $2M in seed funding, and accumulated 37,390 stars [website scrape]. The original pitch was simple and resonant: one interface for all your files across all your devices, without uploading everything to someone else’s cloud.

Then v1 ran into the classic over-ambitious open source trap. The surface area of building a full file manager — multiple OSes, mobile apps, sync protocols, database layer, UI — was unsustainable. Development stopped in January 2025 [README]. What came next was either a pivot or a collapse, depending on how charitable you’re feeling: the team scrapped v1 entirely and rebuilt from scratch as “a local-first data engine,” releasing v2.0.0-alpha.1 on December 26, 2025 [README].

The new positioning is subtly but meaningfully different. Spacedrive v2 isn’t primarily a file browser — it’s an indexing and search layer. The tagline is now “Index any data source. Search everything from one place. Keep it on your machine.” [website scrape]. The file explorer UI is still there, but it’s framed as the familiar entry point into something more ambitious: a unified index of your emails, notes, bookmarks, browser history, GitHub stars, Slack messages, Notion workspace, and files — all searchable locally, with AI that never phones home [website scrape].

The core is written in Rust on a Tokio async runtime. The desktop app uses Tauri rather than Electron, which keeps bundle size and memory usage down [1][2]. The distributed filesystem uses BLAKE3 for content-addressed hashing, meaning a photo on your laptop and the same photo on your NAS are recognized as one file, not two [README].

---

Why people choose it

The reviews we found break into two camps: people excited about the vision and people reporting on alpha software that crashes. Both are right.

The privacy angle is the strongest pull. The itsfoss.com review [2] focuses on this: creative professionals — filmmakers, musicians, designers — who have TB of assets spread across drives, NAS devices, and old backups, and who balk at the cloud’s implicit bargain of “hand us your data, we’ll make it searchable.” Spacedrive’s architecture genuinely doesn’t require that trade-off. Files stay where they are; Spacedrive stores only metadata, paths, content hashes, and extracted text [4]. The StackFoss review [1] calls the approach “personal distributed cloud,” which is a more accessible frame than “content-addressed VDFS.”

The architecture is credible. The v2 philosophy doc [4] is unusually self-aware for an open source project — it specifically names v1’s failure (“perfect is the enemy of good”), explains what was over-engineered (a custom CRDT sync implementation), and explains what replaced it (domain-separated sync that’s simpler and shippable). The whitepaper [5] covers transactional operations, conflict preview before commits, and offline-first sync. This is the kind of thinking that distinguishes projects that ship from projects that accumulate GitHub stars while staying perpetually in beta.

The Spacebot angle is genuinely interesting. The v3 homepage [website scrape] describes an AI agent natively integrated as a Rust crate — not via IPC or a protocol bridge — that gets access to the Spacedrive index. Every record the AI sees has been through a processing pipeline: prompt injection screening via a Prompt Guard 2 classifier, content classification and quality scoring, and ranked indexing. The practical result is supposed to be an AI agent with “perfect recall of everything you own” that runs locally, not through OpenAI’s servers [website scrape]. This is a plausible competitive differentiator against products like Notion AI or Google’s Gemini integrations, both of which require your data to live in their cloud.

The honest caveat from reviews: Both major third-party reviews [1][2] are from 2023 and 2024, reviewing v1 alpha software. Cloud sync and peer-to-peer were listed as “forthcoming” features then. The v2 rewrite addresses some of this, but the track record means “coming soon” carries less weight than it normally would.

---

Features

Based on the v2 README, website, and documentation:

Core indexing and search:

• Hybrid search combining full-text (FTS5) and vector similarity (LanceDB), merged via Reciprocal Rank Fusion — keyword precision plus semantic understanding in one query [website scrape]
• Adapters for any data source: script-based, any language, TOML manifest — if it can write JSON lines to stdout, it can be an adapter [website scrape]
• Portable repositories: each data source is a self-contained folder with its own SQLite database and vector index — zip it, move it, done [website scrape]
• Cross-device deduplication via BLAKE3 content hashing [README][4]
• Offline volume support: disconnected drives stay indexed, search results show availability [website scrape]
• Reference-only indexing: files stay on disk, Spacedrive stores paths, hashes, extracted text, and metadata [website scrape][4]

AI and Spacebot:

• Processing pipeline: safety screening (Prompt Guard 2), content classification, quality scoring, ranked indexing before anything enters the search index [website scrape]
• Spacebot AI agent: concurrent channels, branching thought processes, parallel task workers [website scrape]
• Local AI via Ollama — inference stays on your hardware [4]
• Natural language file management: “find my design assets from last fall” → safe, previewable action [4]
• Data Guardian: tracks redundancy, flags files that exist only in one location [4][5]

Sync and distributed filesystem:

• Peer-to-peer sync via Iroh — no central server, no single point of failure [README]
• Transactional operations: preview conflicts, space savings, and outcomes before committing [README][4]
• Offline-first: full functionality without internet, delta sync on reconnect [README]
• Domain-separated sync instead of full CRDT — simpler, more reliable [5]

Self-hosted server:

• Headless server component for NAS or Linux boxes [3]
• Docker images for x86_64 and ARM64 (including Raspberry Pi) [3]
• Static binary option — no container runtime required [3]
• HTTP RPC endpoints with basic auth (SD_AUTH env var) [3]
• systemd service support for auto-restart on boot [3]
• Port 8080 for HTTP, port 7373 for P2P connections [3]

Desktop app:

• Tauri-based (not Electron) — smaller memory footprint [1][2]
• Grid, list, and media view layouts [2]
• Tagging system for custom file organization [2]
• Cross-platform: macOS and Linux in v2 alpha; Windows “coming in alpha.2” [README]

---

Pricing: SaaS vs self-hosted math

Spacedrive: Free. AGPL-3.0 license. The server runs on your hardware [README][3]. No per-user fees, no storage limits, no bandwidth charges.

What you’re replacing:

Google Drive / Workspace — for teams, Business Starter is approximately $6/user/month for 30GB pooled storage. As your team grows or you need more storage, you’re stacking costs. The real pain isn’t the storage fee; it’s the implicit agreement that Google processes your files to power Search, Drive AI, and ad targeting.

Dropbox Business — approximately $15/user/month, $180/year per seat, with storage and version history capped unless you pay more. A 5-person team is $900/year before you’ve bought anything else.

Self-hosted Spacedrive on a Raspberry Pi 5: ~$80 one-time hardware cost, $0/month ongoing. On a used mini PC or a Hetzner VPS: $5–15/month. For a 5-person team, that’s $60–180/year total — not per user.

The actual math depends on your use case. Spacedrive doesn’t replace cloud storage in the sense of syncing files to a remote backup — it indexes what you already have and makes it searchable. If you need offsite redundancy, you still want Backblaze B2 or similar. If you need real-time file sync across machines the way Dropbox does it, Syncthing (free, purpose-built) is a more mature choice right now. What Spacedrive replaces is the “find the file I worked on last week across five devices” problem — currently solved by paying Google Workspace and uploading everything.

---

Deployment reality check

The v2 self-hosting documentation [3] is clean and well-written, which is a good sign for a project still in alpha. Three deployment paths exist:

Static binary (simplest): Download, extract, verify SHA256 checksum, move to /usr/local/bin, run. ARM64 builds available for Raspberry Pi [3]. This is as simple as self-hosted software gets.

Docker: Single docker run command with volume mounts and the SD_AUTH environment variable [3]. x86_64 and ARM64 images both available.

Docker Compose: Production config with health checks and restart policies lives in apps/server/docker-compose.yml in the repo [3]. Mount your existing storage as read-only volumes for indexing without touching the files.

What you’ll need:

• Any Linux system or NAS (macOS server support not documented)
• 1–2GB RAM minimum; more if you’re running AI processing
• Network access for remote clients to connect; reverse proxy with TLS if internet-facing
• The SD_AUTH environment variable set — the server refuses to start without it or explicit disabled flag [3]
• Port 7373 open for P2P sync between devices

What can go sideways:

The alpha status is real. v2.0.0-alpha.1 shipped December 26, 2025. Windows isn’t supported yet. Mobile apps are “coming soon” — the same phrase that described features in v1 that never shipped before the team abandoned it [README][2]. This isn’t FUD; it’s the project’s own stated status.

The v1 abandonment precedent. The team stopped v1 development and rebuilt from scratch. This is described in the README as learning from hard lessons. It’s plausible that v2 is the result of that learning — the architecture docs are more coherent than most alpha projects. But for a founder making a long-term infrastructure bet, the one-time full rewrite is a fact to weigh.

Adapter ecosystem is early. The adapter system (script-based, any language, JSONL to stdout) is elegant but the directory of available adapters is not mature. You’ll likely be writing your own for anything beyond the basics.

No mature community reviews of v2 yet. The two substantive third-party reviews [1][2] are of v1 alpha. v2 is too new to have accumulated real-world production feedback.

Realistic setup time for a technical user: 30–60 minutes for the server on a fresh Linux box. Non-technical users should have a technical person handle it — the documentation is good, but the software is alpha, and there’s no installer wizard.

---

Pros and cons

Pros

• Architecture is genuinely sound. Content-addressed deduplication, BLAKE3 hashing, peer-to-peer via Iroh, transactional operations with conflict preview — these are the right primitives for the problem [README][4][5].
• Local AI that actually means local. Spacebot runs as a Rust crate inside the process, not via API call to OpenAI. Combined with Ollama integration, the entire AI processing loop can stay on your hardware [website scrape][4].
• Tauri, not Electron. Smaller binary, lower memory usage than Electron-based alternatives [1][2].
• Adapter-based data sources. Any script that writes JSONL can be an adapter — this is an extensible model that doesn’t require the Spacedrive team to build every integration [website scrape].
• Safety pipeline baked in. Prompt injection screening, content classification, and quality scoring before anything enters the index is thoughtful architecture for an AI-adjacent product [website scrape].
• 37K+ GitHub stars, $2M seed. There is real community and investor interest. This isn’t a one-person weekend project [website scrape].
• Self-hosting docs are clean. Docker, static binary, systemd service, ARM64 support — the deployment story is well-documented for an alpha [3].

Cons

• Alpha software, not production-ready. v2.0.0-alpha.1. Windows not supported yet. Mobile not available. This is a genuine blocker for most use cases [README].
• V1 was abandoned. The team stopped development on v1 and rebuilt from scratch. The v2 team is the same team making the same ambitious promises. The rewrite rationale is credible, but the precedent exists [README].
• AGPL-3.0 license has commercial restrictions. Unlike MIT (Activepieces) or Apache-2.0, AGPL-3.0 requires any software that uses Spacedrive’s code to also be open source. If you’re building a commercial product on top of it, get a lawyer first [README].
• Not a Dropbox/Drive replacement for file sync. Spacedrive indexes files; it doesn’t sync them. If you need offsite copies, you still need Backblaze, rclone, or Syncthing alongside it [4].
• Adapter ecosystem is immature. The architecture is elegant but the library of ready-made adapters is thin. Expect to write your own [website scrape].
• No third-party reviews of v2 exist yet. The two independent reviews [1][2] are of v1 software. v2 launched December 2025 and hasn’t accumulated real-world feedback.
• P2P sync is unreliable in alpha. itsfoss.com [2] noted cloud sync and peer-to-peer were “forthcoming” in v1 — they’re now built into v2 architecture, but alpha reliability is unknown.

---

Who should use this / who shouldn’t

Consider Spacedrive if:

• You’re a developer or technically comfortable founder with files scattered across a laptop, NAS, external drives, and multiple cloud services, and you want one search interface.
• You’re building something AI-native and want a local-first index your AI can query without shipping everything to OpenAI’s infrastructure.
• You’re comfortable running alpha software and willing to hit bugs in exchange for being early on something architecturally interesting.
• Privacy and data sovereignty are non-negotiable — you want semantic search over your files without a cloud company processing them.

Wait (or pick something else) if:

• You need something production-stable today. v2.0.0-alpha.1 is not that.
• You’re on Windows — v2 doesn’t support it yet [README].
• You need mobile file access — apps are “coming soon,” same status as v1 features that never shipped.
• You need real-time file sync across machines — use Syncthing, which is mature and purpose-built.
• You need a commercial-friendly license — AGPL-3.0 has copyleft requirements that MIT or Apache-2.0 don’t [README].

Stay on Google Drive / Dropbox if:

• Your team is non-technical and needs something that just works with zero maintenance.
• You need mobile-first file access from anywhere.
• Your compliance team requires audited, enterprise-supported infrastructure.

---

Alternatives worth considering

• Syncthing — purpose-built peer-to-peer file sync, mature, production-tested, no indexing or search layer. If sync is the core problem, Syncthing is the more reliable answer today.
• Nextcloud — mature, self-hosted cloud platform with file sync, calendar, contacts, docs. Heavier than Spacedrive but production-ready. The closest thing to a self-hosted Google Workspace.
• Immich — self-hosted photo and video library with AI tagging and face recognition. Narrower scope than Spacedrive, but ships features.
• Eagle / DEVONthink — paid desktop tools for organizing creative assets and documents. Not open source, but they work today without babysitting alpha builds.
• Jellyfin — if the use case is specifically media organization across devices, Jellyfin is mature and purpose-built.
• Meilisearch + your own indexing scripts — if you’re technical and want a production-grade local search layer now, roll your own with Meilisearch. More work upfront, zero dependence on one project’s alpha status.

---

Bottom line

Spacedrive’s vision is correct. Computing really did land in a fragmented, cloud-dependent mess, and the right fix is a local-first index that understands content rather than just file paths. The v2 architecture — Rust core, content addressing, P2P via Iroh, AI integration that stays on-device — is more coherent than most projects at this stage. The pivot from full file manager to data engine was probably the right call.

The honest question isn’t whether the architecture is good. It’s whether this team will execute it to completion. V1 attracted 37K stars, raised $2M, and was abandoned. V2 shipped an alpha in December 2025. If v2 ships Windows support, stabilizes the P2P sync, and builds out the adapter ecosystem over the next 12 months, this becomes one of the most interesting self-hosted tools in the category. If development stalls again, it joins a long list of open source projects with beautiful READMEs and no production users.

Watch the GitHub releases cadence. If alpha.2 ships Windows support on schedule and alpha.3 ships mobile, the trajectory is real. If releases slow and issues pile up unanswered, you have your answer.

---

Sources

1. StackFoss, Medium — “Spacedrive: A File Explorer from the Future” (June 25, 2023). https://stackfoss.medium.com/spacedrive-a-file-explorer-from-the-future-45b5732b4582

2. Sourav Rudra, It’s FOSS — “Spacedrive: Rust-Based Open-Source File Manager” (September 26, 2024). https://itsfoss.com/news/spacedrive/

3. Spacedrive Official Documentation — “Self-Hosting Spacedrive”. https://v2.spacedrive.com/overview/self-hosting

4. Spacedrive Official Documentation — “The Spacedrive Philosophy”. https://v2.spacedrive.com/overview/philosophy

5. Spacedrive Official Documentation — “Spacedrive V2 Whitepaper”. https://v2.spacedrive.com/overview/whitepaper

Primary sources:

• GitHub repository and README: https://github.com/spacedriveapp/spacedrive (37,390 stars, AGPL-3.0 license)
• Official website: https://www.spacedrive.com
• V2 documentation: https://v2.spacedrive.com