BearDrive Review (2026): Pricing, Features & Honest Verdict

BearDrive is an open-source shared folder built for teams whose AI agents write files. Agents drop reports, spreadsheets and docs into real paths, and everyone gets a link and full version history seconds later. Best for: teams running several coding agents who are sick of pasting files into Slack. Price: Free (free plan: yes, self-hosting free forever). Rating: 7.5/10.

What is BearDrive?

BearDrive solves a problem that only started existing about eighteen months ago. Your agents produce files. Claude Code writes a migration plan, Codex generates a spreadsheet, Gemini CLI dumps a report. All of it lands on whichever laptop happened to be running the agent, and then a human copies it into Slack so the rest of the team can read it. BearDrive is the shared folder that removes that last step. As the site puts it: “Real files at real paths, with a link and a full history seconds after they’re written.”

We found it at number six on Product Hunt’s 12 August board with 158 upvotes. It comes from Runbear, the team behind the AI-teammate-in-Slack product of the same name, founded by Snow Lee, Liam Hwang and Shuwn Jung. The code is AGPL-3.0 and the server ships as a single Go binary you run with bdrive serve. The managed version is free during beta with no card required, and the self-hosted version is free permanently.

How does BearDrive actually work?

Your agent mounts the folder and writes to it the way it writes to any local path. No SDK, no upload API, no “export to BearDrive” step in a prompt. That design choice is why it works with Claude Code, Hermes, Codex, Gemini CLI and, per the vendor, “anything else that writes a file.” A tool that requires the agent to know about it would fail immediately, because the agent forgets.

Every write becomes a version. The history is attributed down to the individual agent session, so you can see which agent produced which draft and when, and restore any of them. This is the feature that separates BearDrive from just pointing your agents at a Dropbox folder. Dropbox tracks that a file changed. BearDrive tracks that agent session 4f2a changed it at 14:03 while working on the pricing analysis, which is the information you actually need when the output is wrong and you are trying to work out why.

Sharing is a URL. Generate a link, send it to someone with no BearDrive account, they read the file. For handing an agent’s output to a client or a colleague outside the team, that is the whole workflow, and it is a meaningful improvement over attaching a file to an email and losing the thread.

What are BearDrive’s key features?

Version history attributed to agent sessions

Every change is tracked, attributed and restorable, with attribution reaching individual agent sessions rather than stopping at “some process on some machine.” When four agents are writing into the same folder, this is not a nice-to-have, it is the only thing standing between you and a folder full of files nobody can account for.

Read heat maps

BearDrive shows which files your agents actually depend on and which have gone stale. This is the most interesting feature in the product and the one nobody else is shipping. Agent-generated documentation rots fast, and knowing that six agents read architecture.md last week while onboarding.md has not been touched since March tells you exactly where to spend your review time.

Shareable links with no account required

Public URLs for individual files, no sign-up wall for the recipient. Simple, and the correct default for a tool whose output is meant to leave the team.

Offline work with automatic sync

Teams can keep working disconnected and changes reconcile when the connection returns. Standard for a sync product, but worth confirming it exists, because plenty of AI-era tools assume permanent connectivity and fall over on a train.

Self-hosting with your own storage backend

One Go binary, and storage on S3, Google Cloud Storage, Cloudflare R2, MinIO or plain local disk. The self-hosted build is the full product, not a crippled community edition, and Runbear states plainly that “the open source core is not going behind a paywall.” For teams whose agents are writing files containing customer data, being able to keep the whole thing on infrastructure you control is the difference between adopting this and not.

How much does BearDrive cost?

Nothing, on both routes. The managed cloud version is free during beta with unlimited projects, teammates and history, no card required, and Runbear manages the server, storage and backups. Beta users are promised founder pricing when paid plans eventually arrive. The self-hosted version is free forever under AGPL-3.0, with full functionality and full version-history retention.

You can also move between the two. Export an entire project including every file version and the edit history, then migrate from managed to self-hosted or back. That export path is the single most important detail on the pricing page, because it is what makes “free during beta” a low-risk proposition rather than a trap. If Runbear prices the managed tier at $40 a seat next year, you leave with your data.

For context on what this replaces: a Dropbox Business seat runs around $18/month, Google Workspace Business Standard is $14/user/month, and neither gives you agent-session attribution or read analytics. BearDrive is not cheaper than those, it is free, and the comparison is really about capability rather than price.

Who is BearDrive best for?

Use BearDrive if more than one agent on your team writes files that other people need. That is the entire qualifying condition, and if you meet it the tool costs nothing to try and removes a genuinely annoying manual step.

Use it if you already run Claude Code or Codex across a team and have found yourself asking “where did that report go.” The read heat maps and session attribution are designed for exactly that moment.

Use it if you need this on your own infrastructure. Self-hosting is a first-class path here, not an afterthought, and the storage-backend list covers the realistic options.

Skip BearDrive if you have one person running one agent. A local folder and occasional Slack upload is fine, and you do not need version attribution for a team of one.

Skip it if AGPL is a problem for you. AGPL-3.0 is a strong copyleft licence, and if your legal team has a list of licences it will not allow near your stack, check the list before you deploy the binary.

And skip it, for now, if you need a stable commercial contract. This is a beta with no paid tier, which means no SLA, no committed support response, and a pricing model that does not exist yet. The self-hosted escape hatch mitigates that risk substantially, but it does not eliminate it.

Best BearDrive alternatives

Dropbox is the incumbent answer and costs about $18/user/month on Business. It syncs reliably, everyone already knows how it works, and its version history goes back further than most teams need. It knows nothing about agents, so you get file history without any sense of which agent session produced what, and no read analytics at all.

Git plus GitHub is free for small teams and gives you the best attribution and diff tooling in existence. It is also the wrong shape for this problem. Agents generating spreadsheets, PDFs and hundred-page reports produce exactly the kind of binary churn Git handles worst, and asking an agent to commit cleanly is asking for a folder full of “wip” commits.

Syncthing is free, open source and peer to peer, with no server to run and no vendor at all. It is the right pick if your only requirement is that files appear on other machines. There is no web link sharing, no read analytics and no per-session attribution, so you are trading BearDrive’s whole feature set for a simpler dependency.

Final verdict: is BearDrive worth it?

BearDrive is a small, well-aimed tool that solves a real and very new problem, and the read heat map is the kind of feature that only comes from actually watching a team drown in agent output. Session-level attribution plus restorable history plus a shareable link covers the whole lifecycle of an agent-generated file, and the fact that the self-hosted build is the complete product rather than a teaser makes the open-source claim credible rather than marketing.

The caveats are maturity and licence, not capability. This is beta software with no paid tier, so there is no support commitment behind it, and AGPL-3.0 will disqualify it at some companies before anyone gets to evaluate the features. Neither is a reason to avoid trying it. If your team runs multiple agents and files keep going missing, install the binary this week: the price is zero, the migration path out is documented, and the worst case is you learn something about how your agents actually use each other’s output.