Spacedrive Technology Inc. founder Jamie Pine announced Spacedrive v3 on March 10, 2026, repositioning the project from a cross-platform file manager into a local-first data engine explicitly designed as infrastructure for AI agents. The v3 codebase went public on March 15, 2026, under an AGPL-3.0 license. The pivot follows two prior versions that never reached stable release: v1, a viral open-source alpha that attracted 37,000 GitHub stars and 600,000 downloads before collapsing under infrastructure debt with a 12-person team, and v2, a solo 183,000-line Rust clean-room rewrite that Pine completed alone but could not reduce to a shippable surface area across five operating systems. Pine's own post-mortem is candid: "the most valuable part of Spacedrive was never the file browser. It was the index."

V3 is built around a single Rust library crate, spacedrive-core, that ingests data sources — email, notes, bookmarks, Slack exports, GitHub, calendar, contacts, and browser history among eleven launch adapters — and makes everything searchable on-device. Every record must clear a mandatory four-stage processing pipeline before reaching the search index. Stage 1 runs Meta's Prompt Guard 2 injection classifier locally on CPU in under 50ms per chunk, quarantining records that contain <a href="/news/2026-03-14-rag-document-poisoning-attack">adversarial prompt injection instructions</a> before they can reach AI consumers — a direct response to OWASP's top-ranked LLM vulnerability. Stages 2 through 4 handle content classification, adapter-specific processing such as email signature stripping and Slack bot detection, and final indexing into both an FTS5 full-text index and a LanceDB vector index using FastEmbed embeddings that run entirely on-device.

V3 also introduces a trust tier system that classifies repositories by data provenance. User-authored notes receive balanced screening, shared Slack workspaces receive collaborative classification, and email inboxes default to strict screening because any sender controls text that enters the search corpus. Per-repository visibility controls allow users to mark sources as private or agent-excluded, and the desktop application surfaces agent access status per repository. Pine describes this as a deliberate contrast to competing local data tools: "every tool in this space pipes raw content straight into LLM context with zero screening."

The commercial model is structured around Spacebot, an AI agent product that links spacedrive-core directly as a Rust crate with no IPC overhead. Spacedrive itself operates as the <a href="/news/2026-03-14-captain-yc-w26-launches-automated-rag-platform-for-enterprise-ai-agents">knowledge layer</a> — screened, classified, and vector-indexed source data — while Spacebot's separate memory system handles weighted associations and importance scores. Pine says that arrangement eliminates the monetization pressure he blames for v1's over-hiring, with the business model framed as Spacedrive making Spacebot more valuable rather than requiring its own revenue stream. The v2 codebase will be preserved on a dedicated branch, leaving GitHub to host a public record of all three iterations of the project.