Buzz hosting that does not run out of disk.

Buzz is an open-source (Apache-2.0) workspace published by Block, Inc., where team chat, code collaboration, and automation share one signed event log, and AI agents participate as first-class members. The source lives at github.com/block/buzz and packaged desktop builds are on the releases page ↗. Hachiflow does not build Buzz; we run it for you: hardware, upgrades, backups, security patching, and the operations around them.

From the upstream project, not from us. Packaged desktop builds for macOS, Linux, and Windows are published on the Buzz releases page ↗ by Block. Note that the Windows build is currently an unsigned alpha upstream, so SmartScreen will warn you on install. Our connect your app walkthrough takes you from download to first message, including how to point the app at your relay.

No. Buzz is Apache-2.0 open-source software published by Block, Inc. Hachiflow is an independent service that hosts it. Buzz and Block are trademarks of Block, Inc., used here only to identify the software we host (nominative fair use). We are not sponsored by, endorsed by, or otherwise affiliated with Block, Inc. The source lives at github.com/block/buzz and anyone can self-host it without us.

Yes, and we encourage you to try it. Buzz is Apache-2.0, the standard stack is the relay, Postgres, and MinIO, and Railway will run it from about $5 a month, as will Fly, Hetzner, or a machine under your desk. Some teams should self-host, and we would rather say so than pretend otherwise. What changes at that price is who the operator is: you make the backups, and you prove they restore, which is the step most self-hosted setups skip until the day it matters. You patch the relay and Postgres, watch the disk MinIO writes your files to (unencrypted in the stock setup, unless you wire up key management yourself), plan for that disk filling up, and answer for all of it at 2am. That work is what our $50 buys: a daily backup of the whole workspace, every file hashed on the way through, verified end to end weekly; restore drills we run by hand on real infrastructure, the last of which brought a destroyed workspace back in 56.6 seconds; media in cloud object storage off the server's disk, encrypted at rest; upgrades that roll back automatically when a new version misbehaves. If running that stack sounds like a good weekend, self-host with our blessing. If what you want is the result of using a Buzz workspace rather than the overhead and ops of running one, that is the job you are hiring us for.

Because what we run for you is one dedicated machine. That cost does not grow by a line every time you hire. Per-seat products bill per person per month, so their bill climbs; ours does not. There is one price for the workspace, and nobody inside it is a line on the invoice. Invite the whole company.

Yes. That is the use we keep seeing: Skool stays your front door, your courses, and your calendar, and Buzz becomes the place your members actually talk between sessions. The usual candidates fit a community badly: per-seat chat bills a growing community like a growing company, and Discord is free but a Discord server lives on Discord's infrastructure under Discord's rules, with no supported way to export your server's history, and your members are Discord accounts first. If free is the whole bar, use it with our blessing. A hive is chat that belongs to you: channels, threads, DMs, and file sharing on your own dedicated server, $50 flat however many members join, every message a signed event you can export any day, and moderators or onboarding helpers running as agent members on your own LLM keys. Your course business stays on Skool. The conversation around it runs on something you own.

No. An agent is a member of your workspace like anyone else: its own keypair, its own channel access, its own signed history. Its runtime runs on your machines and its inference runs on your own LLM API keys, so the compute behind it was never ours to meter. What it uses on our side is storage and relay events, the same as an active teammate. We are not counting agents, we do not sell them by the head, and there is no per-agent add-on because there is nothing to add on to. Run as many as the work needs.

Then it costs $1 per 10 GB a month past the included 100 GB. Your files live in Cloudflare R2 under a namespace scoped to your workspace alone, off the server's disk, so the allowance is priced object storage, not a hope about the hardware. Overage starts with an email from us telling you what you use and what it will cost, never with a silent line on an invoice. If your work runs on very large files, write to hello@hachiflow.com before you sign up and we will tell you what to expect.

Here is the exact map, because a vague yes is how trust dies. Your files: encrypted at rest in cloud object storage, always. The stock Buzz setup does not do this; its bundled MinIO writes media to the server's disk unencrypted unless you wire up key management yourself. Workspace secrets and your escrowed owner key: protected with hybrid post-quantum encryption (ML-KEM-768 + X25519) before they touch storage, and we can still read our escrow copy, on purpose, because that is how a lost key comes back. In transit: TLS everywhere. What we do not claim: end-to-end encryption. Like every Buzz deployment, the relay can read channel messages in order to serve them, and the at-rest keys for files belong to the storage provider, not to you. If a vendor selling the same software tells you otherwise, ask them to show you the code.

Yes, for the agents' inference. Bring keys from any provider and pay them directly at cost. We never mark up tokens, and your prompts never route through our billing. The agent runtime runs on your own machines rather than on the relay we host, so the compute behind an agent is yours as well. What lives on your relay is the agent's identity and its signed history.

Yes. Day to day, Hachiflow runs on keys you already own: join a workspace with your existing key and an admin can enrol it, promote it, and hand it any role short of owner. The workspace's root owner key is different on purpose: we mint it and hold a copy in escrow so a lost key is recoverable, which is only safe because it is not anybody's personal identity. Bringing your own key as the root is on the roadmap as an explicit opt-in, with the trade-off stated up front: we cannot recover what we never hold. The full picture is on the keys page.

Admins run the workspace day to day: invites, adding and removing members, moderation. Members and admins both act with their own personal keys, which we never see. The owner is the one role above that, and role changes are the owner's alone: only the owner promotes or demotes. The owner key is the workspace's root, minted by us and held in escrow, so it is the one key that survives being lost.

Two honest answers. If it is the workspace owner key: yes. We hold a copy in escrow, recovery is verified and audited, and handing it back is the reason the escrow exists. If it is your personal identity key: no. It only ever lived on your device, we never had a copy, and we cannot recover it. That asymmetry is by design: the recoverable key is the one we hold, and the key that is truly yours is the one nobody can recover for you. Nothing else is lost either way: with a new personal key you are re-enrolled and your workspace carries on. Details on the keys page.

Invite links are signed codes with a built-in expiry, 30 days at most, and every outstanding invite also dies at once when the workspace rotates its signing identity. Both are deliberate: a link anyone can forward has to be short-lived and instantly revocable. Ask an admin for a fresh one; minting takes seconds. Our own community door is hachiflow.com/hive, a stable address that always redirects to a live invite.

We are building an importer, and building it honestly: imported messages arrive as clearly labelled archival records, because we will not forge cryptographic signatures on your teammates' behalf. It is in development, not shipped, so do not buy today on its account. One thing worth doing now regardless: export your Slack workspace while you still can. On Slack's free plan, messages older than 90 days are hidden from you, and free and Pro plans only export public channels.

Everything we hold for you, in a form built for leaving. Your history exports as raw signed events, every signature re-verified on the way out. Your files come content-addressed with an index, your git repos come as clonable repositories, and your owner key is deliberately not in the archive: you claim it separately in the dashboard, where every handover is audited. You start the export yourself in the dashboard, watch it build, and download it from there. The export stays downloadable for about a week, and each download mints a fresh private link that is good for fifteen minutes, so no long-lived URL to your data sits in a browser history or an inbox. A workspace too large for that path is still packaged by hand. One more deliberate choice: it is not a database dump. A dump is the shape of lock-in; signed events are the shape of leaving. What the archive is, is a record, and that is the strong part: anyone can verify it offline, forever, with no software of ours, because every event id recomputes from the canonical serialization and every signature still validates. What it is not is a restore button. A relay refuses any event whose timestamp is more than fifteen minutes off its own clock, and that timestamp is inside the signature, so bringing a working workspace up somewhere else is the part we still do with you, out of the backup set. We drill it: the last rebuild returned 52 of 52 events with their original ids and timestamps, 53 of 53 signatures verified, 3 of 3 channels, 7 of 7 threaded replies, the roster and its roles, 19 of 19 media objects hashing to the value their author signed, and both git repos cloning to identical commit hashes.

As big as your workspace. Exports over 200 MB, the usual per-request ceiling on upload paths like ours, ship via multipart upload: uniform parts, reassembled by the storage layer, verified against its own size echo. We have drilled a 260 MB export end to end against production, byte for byte, with the store's own multipart fingerprint confirming the path. The export is a verifiable record of your workspace: signed events, content-addressed files, clonable repos. It is a record you can verify offline, not a restore button.

Honest answer: Buzz is alpha software, and we would rather lose the sale than hide it. Solid today: the relay, channels, DMs, file sharing, agents as members, daily backups we verify, and an exit that hands you everything. Rough today: an unsigned Windows build upstream, no mobile app, and a support motion that is a small team and a status page. If your team can live with that, the $50 is locked for life and the people answering email are the people running your relay.

No. Every workspace is served at its own address, like yourteam.hachiflow.chat, over HTTPS, from your own machine. Pointing a domain you own at your workspace is not something we support, so if that is a hard requirement, this is the honest place to find it out rather than after you have paid.

Yes, physically. One workspace, one server, one relay. No shared databases, no noisy neighbors, and a blast radius of exactly one hive. It also keeps the handover clean: one database, one media store, one set of keys, and all of them only yours.

Yes. Each workspace is its own subscription at the same flat price, and your chat identity is the same key in all of them: join your second workspace and you already exist there. The dashboard lists every workspace you own, and switching between them is one click.