Privacy Policy for Mycolab
Last Updated: 18 September 2026
⚠️ Correction notice (15 September 2026)
Every previous version of this policy overstated our encryption. It described a "zero-knowledge encryption architecture", said your conversations and code were "end-to-end encrypted on your device before transmission", and stated that we "cannot decrypt your data even if compelled" using "TweetNaCl (same as Signal)".
Those statements were not true of the software we ship. That text was inherited from the upstream project this app was forked from (slopus/happy, which does implement it) and was rebranded to Mycolab without being re-verified. In our build, the encryption layer is a pass-through: content is serialized and transmitted, not encrypted by us, and our servers can read it.
We are correcting this the moment we found it, rather than quietly softening the wording. If you chose Mycolab because of the previous claims, please read "What we can see" below and decide accordingly — we would rather lose your trust honestly than keep it on a false statement.
End-to-end encryption is planned, and the engineering analysis is public in
docs/E2EE-FEASIBILITY.md. We will not claim it again until it is implemented, and until the keys are genuinely outside our reach.Update (16 September 2026) — it now ships, for part of what we handle
End-to-end encryption is implemented and deployed. Chat messages and terminal output in newly created sessions are encrypted on your devices under keys we never receive. This does not cover session titles, project paths, machine names, timestamps, or uploaded file contents, and it does not apply to sessions created before it was enabled — those stay readable by us, permanently.
Update (18 September 2026) — it now covers commands, calls and file contents
Three more planes are encrypted since the note above, and the note above is left as written rather than widened, because a claim that grows by silent edit is the habit this page exists to break. Now also covered: terminal input (what you type, not just what prints), every call your app makes to one of your machines — arguments and reply — and the contents of the files you browse and edit in a session. Still not covered: session titles, project paths, machine names, timestamps, the names of the files you open, and anything you upload to artifacts.
The correction above stays on this page rather than being deleted. We claimed this once without shipping it; the fix for that is not to quietly replace the retraction with a fresh claim, but to leave both and let the dates speak. The precise boundary is on the Security and encryption page, and "What we can see" below is written to match it.
Overview
Mycolab lets you run and observe AI coding sessions on machines you own, from a browser or phone. Your session content travels through our servers (Hypha Mesh, running on Cloudflare) so that your devices can reach your machines.
This policy describes what we actually do today.
What we can see
For sessions that are not end-to-end encrypted, we can technically read your session content. That is every session created before encryption was enabled for your account, every session imported from a previous install, and any new session created while one of your devices is not yet enrolled.
- Messages and code — your conversations with the coding agent, and any code or file contents that appear in them, are stored on our servers in a form we can read.
- Terminal output — the output of commands run in a session, likewise.
- Files you upload to artifacts — files you put into the artifacts/resources feature, stored in our object storage.
- Files on your own machine — the files in a session's folder are NOT stored by us; they sit on your machine, and we carry the bytes when the app reads or writes them.
- Session and machine metadata — session titles, project directory paths, machine identifiers, and lifecycle state.
For sessions that are end-to-end encrypted, what we can read shrinks to the last item. Chat messages, terminal output, terminal input, the arguments and replies of the calls your app makes to your machines, and the contents of the files you browse and edit are all encrypted on your devices under keys we never receive. What does not change: session titles, project paths, machine identifiers, lifecycle state, the timestamps and sizes of messages, the names and paths of the files you open, and anything you upload to artifacts. Encryption narrows what we can read; it does not make a session invisible to us. The full boundary is on the Security and encryption page.
We do not read this content as a matter of practice, and access is restricted internally. But the architecture does not prevent us from reading it, and it would not prevent a lawful compulsion order from reaching it. Please do not treat Mycolab as a confidential channel for secrets you cannot afford us — or a party who compels us — to see.
What protects your data today
These are real, and they are what we can honestly claim:
- End-to-end encryption, in encrypted sessions — chat messages, terminal output and input, the calls your app makes to your machines, and the contents of files it reads and writes, all encrypted on your devices under keys we never receive. It does not cover session titles, project paths, machine names, file names, or artifact uploads. Scope, conditions and limits: Security and encryption.
- Encryption in transit — all connections between your devices, our edge, and your machines use TLS (HTTPS / WSS). Your data is not sent in the clear over the network.
- Encryption at rest — data stored on Cloudflare Durable Objects and R2 is encrypted at rest by the storage provider.
- Workspace isolation — your data lives in a workspace that is a separate, single-tenant Durable Object. Cross-workspace access is denied by default.
- Scoped, revocable tokens — access is granted by tokens bound to one workspace and one permission level, revocable at any time; authorization fails closed.
- Open source — the implementation is publicly auditable, which is how this correction notice came to be written.
What we collect
Content
As listed under "What we can see" — messages, terminal output, artifact uploads, and session/machine metadata. For encrypted sessions we hold the messages and terminal output only as ciphertext, and the contents of files on your machine pass through us sealed; their names do not.
Operational metadata
- Identifiers — message, session, machine, workspace and device identifiers
- Timestamps — when messages and events were created and synchronized
- Account information — the email address you sign in with
- Connection data — IP address and standard request metadata, as any web service receives
What we do not collect
- Your Claude/Anthropic or other third-party API keys, beyond storing them where you explicitly ask us to so that your sessions can use them
- Location data
- Advertising identifiers; we do not sell data or serve ads
Analytics
The app derives a pseudonymous analytics identifier from your credential. This identifier is linkable back to your account by us — an earlier version of this policy claimed it was not, which was incorrect for the current sign-in mechanism.
How we use data
- Content is stored and relayed solely to synchronize your sessions between your devices and your machines. We do not use it to train models.
- Metadata is used for routing, ordering, presence, billing and abuse prevention.
- Audit records we keep about service calls contain identifiers, timestamps, decisions and cryptographic digests — not message or file content.
Data retention and deletion
- Session content is retained until you delete it or delete the workspace containing it.
- Deleting a workspace removes its content and its stored objects.
- Operational metadata may be retained after deletion for a limited period for security and billing.
Your rights
Depending on where you live (including under the GDPR if you are in the EU/EEA/UK), you may have the right to access, correct, export, or delete your personal data, to restrict or object to processing, and to lodge a complaint with your supervisory authority. Contact us to exercise these rights.
International transfers
Our infrastructure runs on Cloudflare's global network, and your data may be processed outside your country of residence, including in the United States. We do not currently operate EU-only data residency. End-to-end encryption now ships and is available as a supplementary safeguard for the content it covers — chat, terminal output and input, the calls your app makes to your machines, and the contents of files it reads and writes — but not for the metadata it does not cover (session titles, project paths, machine names, file names) or for artifact uploads, which still transfer and rest in a form we can read. EU-only residency remains on our roadmap; we will update this policy when it ships, and not before.
Third-party services
- Cloudflare — hosting, storage, and network (processor)
- Resend — transactional email for sign-in codes
- Anthropic / other model providers — when a session calls a model, the session's prompts and context go to that provider under *their* terms and privacy policy, not ours
Changes to this policy
We will post material changes here with an updated date. Where a change corrects a previous misstatement, we will say so explicitly rather than editing silently, as we have done above.
Contact
Open an issue at <https://github.com/oeway/hypha-mesh> or contact the maintainers listed there.