# BearDrive > A shared folder that syncs across all your AI agents and teammates, with full version history, public share links, and read-heat insights — open source under AGPL-3.0. BearDrive is an open-source, agent-native shared folder built by Runbear, Inc. and backed by Techstars. It mounts any local directory as a synced volume that every AI agent and teammate on the team reads and writes simultaneously — no uploads, no copy-pasting, no "send me that again." The project is licensed under AGPL-3.0, available on GitHub, and currently in a free managed beta alongside a self-hostable server option. ## What It Is BearDrive describes itself as "Google Drive for AI agents." The core idea is simple: instead of each agent starting from an empty folder or relying on a memory API, every agent on the team shares the same real files at real paths on disk. A lightweight CLI daemon (`bdrive`) keeps the folder synchronized across devices and teammates through a central hub, which can be the managed cloud service at beardrive.ai or a self-hosted server backed by S3, GCS, Cloudflare R2, MinIO, or a plain directory. Because the files are ordinary files, any tool, editor, or agent that can write a file already works with BearDrive — no integration required. ## How the Sync Architecture Works BearDrive uses an append-only journal model: each device writes only its own journal, so no locking service is needed and any dumb object store suffices. Key mechanics include: - **Offline-first** — the working folder is always fully usable with no network; changes are journaled locally and pushed when the remote becomes reachable. - **Conflict-safe** — concurrent edits resolve deterministically (last-writer-wins), and the losing version is preserved as a conflict-copy file rather than silently dropped. - **Delta sync for large files** — files larger than 4 MiB move as content-defined chunks (~1 MiB each), so a small edit to a large file transfers roughly one chunk, not the whole file. - **Agent hook integration** — `bdrive hooks` registers turn-boundary hooks in Claude Code, Codex, Gemini CLI, and Hermes: a blocking pull before each agent turn (so the agent always reads fresh team files) and an async push after every file edit (so artifacts are on the hub seconds after the agent writes them). ## Setup Path Setup is designed to be conversational. Users paste a single line into any supported agent — Claude Code, Codex, Gemini CLI, Hermes — and the agent fetches the setup runbook from `beardrive.ai/setup`, installs the CLI, signs the machine in, asks which folder to sync, registers the hooks, and prints the project link. For those who prefer manual control, the CLI is installable via Homebrew (`brew install runbear-io/tap/beardrive`) or from source via `go install`. ## Knowledge Insights and Read Heat Beyond sync, BearDrive tracks which files agents and humans actually read and surfaces this as a per-file heatmap in the project Dashboard. Four views are available: a treemap of every file sized by reads and colored by staleness, a reads-versus-freshness scatter plot (the hot-but-stale quadrant is described as "the danger zone"), a hot-path list showing the team's effective agent context window, and an agent-coverage matrix showing which agent devices read which folders. The heat data exposes only aggregate counts and last-read times — never who read a specific file. ## Deployment Model and Open-Source Lineage The entire server is open source under AGPL-3.0 — teams, history, share links, and the web UI are all included, with nothing held back in a paid tier. Runbear chose AGPL-3.0 deliberately to keep BearDrive self-hostable forever while preventing a cloud provider from offering a closed BearDrive-as-a-service. The managed service at beardrive.ai runs the same core plus hosting, backups, and support; provider-specific and billing code stays out of the public repository permanently. Projects can be moved between the managed service and a self-hosted hub at any time using `bdrive export` and `bdrive import`, which carry every version of every file and full authorship history. ## Update: v0.15.0 The latest release is **v0.15.0**, published on August 11, 2026. The repository was last pushed on August 13, 2026, indicating active development. The project launched on Product Hunt during this period. The public roadmap (ROADMAP.md) lists upcoming work including `bdrive restore @