Manage hosts, credentials and keys in a desktop app; sync them across your devices and share them with teammates through a server you run yourself. The server stores ciphertext and never holds a key — the operator cannot read the credentials it stores.

Status: early development. See the milestone plan for what exists today.

Why

Teams either scatter SSH credentials across individual ~/.ssh directories with no sharing story, or pay per-seat for a hosted product that holds their infrastructure credentials. DodoSSH keeps the convenience of a synced, shareable vault while remaining self-hostable and zero-knowledge.

Architecture

Component	Choice
Backend	ASP.NET Core on .NET 10, PostgreSQL + EF Core
Client	Avalonia (C#): a desktop head for Windows/Linux/macOS and a phone-first Android head, sharing one set of view models; terminal pane is a WebView running xterm.js
Auth	OIDC, provider-agnostic (Entra ID, Keycloak, Auth0, Authentik)
Vault	End-to-end encrypted; X25519 + Ed25519 + XChaCha20-Poly1305, Argon2id unlock
Connections	Client-direct SSH by default, with an optional raw-TCP server relay

Three consequences worth knowing before you read further:

Revocation is not retroactive. A removed member keeps what they already downloaded. The real remediation is rotating the SSH credential, so offboarding is built around a rotation checklist rather than a button that implies more than it delivers.
No session recording in relay mode. The relay forwards SSH ciphertext, so it cannot see commands. That is the cost of the relay not being able to read your traffic.
Locking the vault does not close your shells. Lock closes the vault and zeroes every key it held; a session that authenticated before it keeps running, because the remote host never consulted the vault and the credential was already spent. That is deliberate — you lock when you walk away from the machine, which is exactly when a long upgrade or transfer is most likely to be in flight, and an idle auto-lock that killed it would be worse than the exposure it removed. The honest reading is that locked describes the vault and not this machine's access to your hosts. The unlock screen therefore shows how many shells are still connected, and quitting DodoSSH is what ends them.

The reasoning behind each major decision is recorded in docs/adr/, starting with the E2EE trust model.

The desktop client's interface was built from a design covering more product than exists yet — file transfer, teams, saved snippets, port forwarding. Everything that design asked for and this build has not got is written down in docs/design-import-gaps.md, with the layer each piece would land in and what the interface shows in its place. Nothing was rendered with invented data to fill a screen.

Repository layout

src/
  DodoSSH.Contracts        DTOs shared with the client — the real API contract
  DodoSSH.Crypto           DSH1 envelope, AAD derivation, the key hierarchy
  DodoSSH.Domain           entities and invariants, no EF
  DodoSSH.Infrastructure   DbContext, configurations, migrations
  DodoSSH.Api              the server
  DodoSSH.Client.Auth      OIDC code+PKCE on a loopback redirect, and the key binding
  DodoSSH.Client.Api       the typed server client, and client-side enrollment
  DodoSSH.Client.Domain    the decrypted item model and the three-way merge — no I/O at all
  DodoSSH.Client.Storage   the local cache: ciphertext mirror, outbox, offline unlock material
  DodoSSH.Client.Sync      the pull/apply/push loop and the conflict policy
  DodoSSH.Client.Session   where a profile lives, unlocking it, and getting one in the first place
  DodoSSH.Client.Ssh       connections, PTY shells, SFTP, host key trust
  DodoSSH.Client.Terminal  the loopback data plane and credit-based flow control
  DodoSSH.Client.Transfer  the transfer queue, part files and resume, and the local file listing
  DodoSSH.Client.ObjectStore  S3-compatible buckets, behind the same interface as SFTP
  DodoSSH.Client.Import    reading ~/.ssh/config, with no I/O of its own
  DodoSSH.Client.Shell     the view models both heads drive, the renderer's files, the palette
  DodoSSH.Client.App       the desktop head: its views, and its Windows integration
  DodoSSH.Client.Android   the phone head: its views, and its Android integration
tests/                     one test project per source project
docs/adr/                  architecture decision records
docs/design-import-gaps.md what the client's design asked for and this build has not got
docs/platform-flags.md     what differs off Windows, and the gotchas that have cost time
docs/manual-checks.md      what no test can reach, and what to look for when checking by hand
docs/android-port.md       the Android head: what was decided, what is built, what is left
scripts/                   release-windows.ps1 — builds, packs and publishes the Windows client

Everything under src/DodoSSH.Client.* except the two heads and Shell is deliberately free of Avalonia. That is the seam that lets the SSH layer, the terminal's flow control and the OIDC flow be tested without a UI toolkit or a browser engine — which is most of why they are testable at all.

Shell is the narrow exception and it earns it: what it takes from Avalonia is Dispatcher, the asset loader and a resource dictionary, none of which imply a window, and what it holds is the shell's state machine — which two heads have to agree on exactly rather than approximately. The rule's purpose was never Avalonia-avoidance for its own sake; it was that the layers with the hard logic stay testable, and none of them are here.

Building

Requires the .NET SDK pinned in global.json (10.0.x).

dotnet build DodoSSH.slnx

dotnet test DodoSSH.slnx

The tests need a Docker daemon. Everything that touches the database, the identity provider or an SSH server uses Testcontainers rather than a stub or a shared instance, so there is nothing to start first and nothing to clean up after — but with no daemon those suites fail rather than skip.

Installing on Windows

The desktop client is published on the project's own release page as a Setup.exe. It installs per user, under %LOCALAPPDATA%\DodoSSH.Desktop, and never asks for an administrator.

It will warn you, and here is exactly what the warning means. The build is not yet signed with a code signing certificate, so Windows SmartScreen shows "Windows protected your PC" the first time you run the installer; More info → Run anyway gets past it. That is the honest state of things rather than something to click through blindly — it is a statement that Microsoft has not seen this file before, and it will stop appearing when the project buys a certificate. ADR 0013 says what that costs and when it happens. The warning is once per person: updates from inside the application do not raise it.

Updates. The client checks the project's release page every six hours, downloads a newer build in the background, and then waits. Nothing is ever installed while you are using it — a downloaded update runs after a restart you ask for, or the next time you start DodoSSH anyway. Restarting does end every shell you have open, which locking deliberately does not, so the choice of when is left to you. You can turn the checking off on PREFERENCES → UPDATES.

Where it comes from matters, and it is worth one paragraph. A DodoSSH server will never offer you the client, and one that does is not one to trust. Whoever hands you the binary can hand you a binary that copies your passphrase — the client is where your credentials are in plaintext, by construction — and the operator of a deployment is precisely the party the trust model is about. An operator may tell you where to get it. They are not where it comes from, and the update check inside the application points at the project's own forge and nowhere else. See ADR 0011 rule 2.

Uninstalling removes the application and leaves your vault cache at %LOCALAPPDATA%\DodoSSH, so reinstalling asks for your passphrase rather than starting over. Use Sign out inside the application if you want the machine to genuinely forget everything — an uninstall is not a sign-out, and does not withdraw this machine's device key from your account.

Cutting a release is scripts/release-windows.ps1, run by a person on a Windows machine. Deliberately not a CI job; ADR 0013 decision 3 explains why, and it is not only that the runners are Linux.

Running it

Three commands, in order. The first is once per machine.

1. The development dependencies — PostgreSQL and Keycloak, with the dodossh realm imported:

docker compose -f deploy/docker-compose.dev.yml up -d

2. The server:

dotnet run --project src/DodoSSH.Api

It applies any pending migrations before it opens its port, so there is no separate schema step and no window in which a request meets a half-applied schema. Set Database:AutoMigrate to false where something else owns the schema — a migrator job, or a database user denied DDL — and the old behaviour comes back: readiness fails while a migration is pending, and the log says which one. To apply them by hand, dotnet-ef is pinned in .config/dotnet-tools.json (dotnet tool restore first if you have not):

dotnet ef database update --project src/DodoSSH.Infrastructure

With nothing else configured that targets the compose stack above. Set DODOSSH_DESIGN_CONNECTION to point it at another database.

The server listens on http://localhost:5233, serving /healthz/live, /healthz/ready and — in Development — /openapi/v1.json.

3. The desktop client:

dotnet run --project src/DodoSSH.Client.App

In the app, enter http://localhost:5233 as the server. Your browser opens for sign-in — the realm ships alice / alice — then choose a vault passphrase and write down the recovery code, which cannot be skipped and cannot be recovered from the server. You can then add a host and open a shell on it — double-click its card, or select it and press CONNECT in the drawer that opens beside the grid, which is the same command with the password box above it. Keycloak's admin console is at http://localhost:18080 (admin / admin).

You can also add an SSH key, which is stored in the vault like a host and synced the same way: paste the private key, then edit a host and pick that key from its key dropdown. From then on that host authenticates with it — on every machine, since the choice travels inside the host's encrypted payload — and its password box disappears.

The first time you connect to a host you are asked to check its key fingerprint. That decision is stored in the vault, so it is asked once per host rather than once per launch and it reaches your other machines with the next sync. If a server is legitimately rebuilt and offers a new key, the connection is refused outright with no way to continue from the warning — edit the host and choose Forget host key, which is deliberately somewhere you have to go on purpose.

Deleting asks first, and the question is worth reading. DELETE on a host, an SSH key or a stored password puts a question where the buttons were, and what it says is counted rather than generic: how many hosts authenticate with the key about to go — they refuse to connect afterwards rather than falling back to a typed password — whether a terminal is open on the host about to go, and whether this machine can push the deletion yet or is queuing it. There is no undo, which is the other thing it says. Withdrawing host key trust is the deliberate exception: it costs one fingerprint check on the next connection, and the dangerous button there is the one that adds trust.

Signing in once is enough. The refresh token is kept in the local cache, sealed under the vault's own key, so a later launch resumes the session itself and no browser opens — and because it is sealed under that key, resuming can only happen after the vault is unlocked. A machine that unlocks with no network keeps trying: every synchronisation pass asks for a connection, so a laptop opened on a train is online again within a minute of finding a network, with nothing pressed. Unlock takes Enter in the passphrase box, and nothing about unlocking ever waits on the network.

Signing out is under Preferences → Account, and again on the unlock screen, where it is the only answer to a forgotten passphrase — nothing can recover one. It asks first, and says what it costs: it empties this machine's cache (the profile, the cached items, and anything still queued to be sent) and withdraws this machine's device key from the account. The vault itself is on the server and is untouched, so signing in again brings it all back; the count in the confirmation is the one thing that exists nowhere else. Your session at the identity provider is not ended — DodoSSH has no way to end it — so on a machine that is not yours, sign out there too.

Neither a password nor a passphrase has to be typed twice, and both ways out of that are opt-in. A password typed to connect is typed once: tick Remember this password under the box and it is saved to your keychain and bound to that host the moment the remote accepts it — or add one outright with + PASSWORD on the Vault screen. And unlock can be a Windows confirmation instead of the passphrase: Preferences → This machine → REGISTER keeps this machine's device key in the TPM, so a later launch offers USE WINDOWS HELLO on the unlock card. A machine with no TPM — and any desktop that is not Windows — is offered neither button and keeps asking for the passphrase, which Preferences says out loud rather than leaving you to notice. The passphrase never stops working either way: a declined confirmation leaves the box exactly where it was.

Moving files

SFTP and S3 in the tab strip are a two-pane browser: this machine on the left, the remote on the right, and a queue underneath. The right-hand pane opens on an invitation rather than a listing — press SELECT HOST, choose one, press CONNECT — and once something is open, select a file in either pane and press the arrow pointing the way you want it to go.

Two things about it are worth expecting rather than discovering.

It is a second connection, not a second channel. SSH itself would allow the SFTP subsystem to open beside a shell on the transport that is already up; SSH.NET does not offer that — its SftpClient owns its own transport — so pressing CONNECT here authenticates again. The host records a second login, and a host whose password you type each time will ask for it again on this screen. Host key trust is shared: a fingerprint approved for a terminal is approved here, and one approved here reaches your other machines with the next sync.

Nothing is written at its final name until it is complete. Every transfer goes to a .dodossh-part file beside its destination and is renamed into place at the end, so an interrupted transfer can never be mistaken for a finished one — which matters most for what people actually use this for, which is copying a build artefact onto a server and then running it. A destination that already exists is refused outright rather than overwritten; the remote pane has DELETE and MKDIR so that refusal is not a dead end. DELETE asks first and names the full path, and it carries the strongest warning in the application on purpose: everything else DodoSSH deletes is a tombstone against a copy the server still holds, and a file on somebody's host is bytes with nothing behind them. RESUME on a stopped transfer carries on from what the part file already holds.

Resume works within a run of the application and not across a restart, and that limit is deliberate: nothing records which source wrote a part file, and resuming one on the strength of its name matching is how a corrupt artefact gets delivered with nothing reporting a failure. A part file found at startup is started over.

What is not here: transferring a directory, dragging between the panes, and routing a transfer through a bastion — the last needs jump hosts the connection layer has not got. All three are in docs/design-import-gaps.md.

VAULTS in the nav rail lists every vault you can see, makes new ones, adds and invites people to one, shares its key, renames it and hands it over. One distinction runs through the whole screen and is worth having before you use it.

A vault is the thing you make, and the group of people is behind it. The server authorises through a team — VaultAccessService resolves a shared vault through team_membership, and every membership call names a team id — but nothing asks you to make one: naming a vault makes the membership list that carries it, named after the vault and owned by you. So the thing you came to share is the thing you create, and "which team is this in" stops being a question you need an answer to before you can share four servers with two colleagues. Renaming the vault renames that membership list with it, as long as it carries nothing else.

The one case where the distinction resurfaces is a team owning several vaults, which this screen cannot produce and does not hide: the members section then says so, because adding somebody to one of those vaults adds them to all of them.

Adding somebody to a vault and giving them its key are two different acts, and only the first is something the server can do. Adding a member changes what the server will serve them: the vault appears in their list immediately. It cannot make it readable, because a vault key is sealed to each member's public key and this server never holds one — so until somebody presses SHARE KEY from a machine that has the key, their vault sits in the list saying it is waiting for one. That is not a rough edge to be smoothed over later; it is what "the operator cannot read the credentials it stores" costs, and the screen says so rather than implying the server handed anything out.

Sharing verifies before it wraps. The client reads the server's append-only key log, checks its hash chain from the first entry, and refuses unless the key the directory just offered appears in that log unchanged. That converts a key substitution by the server from invisible into visible — a substituted key has to be published in a log every other client also reads. It does not prove the key is the right person's. Compare the fingerprint with them over something this server does not carry; that is the only step that closes it, and the success message says so every time.

Somebody with no account here yet can be invited, and nothing is sent. There is one button — ADD — and it does whichever of the two applies, because which one applies is a fact about the server's account table rather than about what you are trying to do. If the directory knows the address, that account is added straight away. If it does not, the address is invited instead, and the status line says which happened, because the difference decides what you do next.

An invitation is a standing instruction rather than a message: the next account that signs in with that address joins this vault, at the role you chose. There is no link and no token, because this server has no outbound mail path and does not pretend otherwise — telling them to go and sign in is your job, over a channel this server does not carry, and a link nobody can deliver would be worse than no link. An invitation lasts fourteen days, so an address handed on to whoever takes the job next does not carry a standing offer for ever; it can be withdrawn until it is taken up; and like adding a member it grants nothing readable, so somebody still has to press SHARE KEY afterwards.

The one thing the merged button costs is worth knowing. Adding an account the directory knows also hands you the public key you are about to verify and wrap a vault to, and an invitation cannot do that because there may be no key yet. So when you are adding somebody in order to share a vault with them, the useful sequence is still the same one: add them, see them appear in the members list, then share.

Inviting an address that already belongs to a member of the vault is refused and says so. Inviting one that merely has an account here is not — that would make this a way of asking the server which addresses have accounts, which is not a question anybody willing to create a vault first should be able to put to it. Such an invitation simply gets claimed sooner: within the hour, on the same sweep that records they were here, rather than waiting for a first sign-in that has already happened.

An invitation is only claimed if your identity provider says the address is verified, and there is no way to relax that. The access token has to carry email_verified as true. Anything else — false, missing, or sent under another name — claims nothing at all, and no setting turns that off: an invitation decides what the server will serve, and one that could be taken by anybody able to obtain a token asserting somebody else's address is a way into a vault. If your invitations never activate, this is the first thing to check. They sit at pending rather than failing, the server logs a warning each time it declines to claim one, and the two fixes are on your side: set Oidc:EmailVerifiedClaim to whatever your provider calls the claim if it is not email_verified, and make sure the provider puts it in the access token rather than only in the ID token or the userinfo response.

Ownership is sole, and handing a vault over is one act. Transferring names an existing active member: they become owner and you become an admin, in a single transaction. Not two role changes — promoting first leaves it owned twice, demoting first leaves it owned by nobody, and there is nobody with the authority to finish a transfer that stopped in the middle. You are demoted rather than removed, so you keep your vault key grants; removing you would revoke them and rotate the vault, and somebody handing a vault over is usually staying in it.

Six limits, stated rather than discovered:

Deleting a vault does not reach a machine that has already synced it. An admin or the owner can delete a shared vault: it leaves everybody's list at once, every key to it is withdrawn, and the membership list behind it is archived when it existed to carry that vault alone. What it cannot do is take back the copy a colleague pulled yesterday — the same limit revocation has, for the same reason — and the question you are asked before it happens says so. Your personal vault is refused: everything filed nowhere else lives in it, and there is no way to make another.
Removing a member is not retroactive. It revokes their grants, rotates every vault behind that membership list your machine can open, and hands each new key to the members who are left — so nothing written from that point on is readable to them. Everything they already pulled is still on their machine. Rotate the SSH credentials that matter — that is the actual remediation, and it is why there is no button labelled anything stronger.
A rotation moves the vault's contents too, and says so when it could not. The key changes first, in one server transaction; then every item already stored is re-sealed under it, so the key somebody left with opens nothing that is still here. Existing items keep working throughout — everybody still in the vault holds the older keys as well as the new one, which is what stops a half-finished rotation making a vault unreadable, and what makes the pass safe to interrupt and run again. An item somebody else was editing at that moment is left for the next pass, and the message tells you which of the two you got. See ADR 0010.
Adding somebody shares the vaults you can open, including their history. Membership is still one act and a key is still another — nothing changed about that — but the client now performs the second one for you, wrapping every generation it holds so the new member can read the vault back to its first item. A vault your machine holds no key to is skipped and says so; somebody who holds it has to share that one.
Host key trust stays in your personal vault. A pin approved for a shared vault's host is recorded and used from your own vault, not the shared one, so a colleague cannot pre-approve a fingerprint that your client will then trust silently for a host you defined. The cost is that each member approves a shared host's key once on each of their machines. Shared vaults' pins are still listed on the Pins screen, so you can see what has been trusted.
LAST ACTIVE is coarse on purpose. The server records it at most once per account per hour, so a value an hour old means "recently" and not "at that moment". That is the granularity the question is really asked at — whether somebody is still using this deployment — and writing it on every request would put an UPDATE on the hot path of every authenticated call for a number nobody reads that closely. It is shown as roughly-when rather than to the minute, because showing it to the minute would be reading a precision into it that is not there.

Items are filed into one vault at a time, and which one is asked at the moment the item is made. A host's editor has its own picker, beside the name, because it is the decision on that form that decides who can read the host. It is absent when you edit an existing one rather than present and refusing, and that is not because the host is stuck: a host can be moved to another vault — "Move to another vault…" in the detail pane's menu on the desktop, MOVE beside EDIT on the phone. It is a separate act because it is not a save. The two vaults are encrypted under different keys, so a move is a re-seal into one and a tombstone in the other; the host gets a new id, and its group and its tags stay behind, because both are items of the vault it is leaving. A picker inside the form would do all of that as a side effect of correcting a port. What a move cannot do is reach a machine that has already synced the host, which is the same limit everything else about revocation has. Keys, passwords and buckets take theirs from a standing "new items go to" picker on the Keychain screen and cannot be moved yet.

A group can be moved too, and it takes its contents with it — "Move to another vault…" on the group card's right-click menu, beside Open, Edit and Delete, which is the whole of what can be done to a group on the desktop. That is the desktop only, because the phone draws groups as headings in the host list and has never had a way to delete or move one. It is the same re-seal and tombstone underneath, applied to every item involved: the group, the groups nested inside it, and every host filed under any of them, each taking a new id in the destination. Moving less than that was never coherent — the machines and the child groups are items of the vault the group is leaving, so a group that travelled alone would leave half a shelf behind. What stays is the group it was itself nested under, which belongs to the old vault, so it arrives at the top level; the hosts' tags stay for the same reason. Keys and passwords are kept, because those genuinely resolve across vaults, and the sentence afterwards names any that are now outside the destination.

Deleting a group asks what should become of the hosts under it. The default answer keeps them: the reference is cleared and they move to UNGROUPED. Ticking the box deletes them with it. Both answers are a change — the deletion used to leave the hosts holding an id that no longer resolved, which looked the same and cost nothing, and stopped being the right shape once the deletion could take them with it. A group with nothing filed under it is not asked. Whichever answer is given, the groups nested inside take the deleted group's place in the tree rather than being orphaned to the top level. Both default to your personal vault and neither moves on its own, because an item put in a shared vault is visible to everybody holding that vault's key. Choosing a vault in the host editor also decides which groups it can be filed under: a group is an item like any other and lives in exactly one vault.

Changes that do not wait

A client holds a WebSocket open to the server — GET /api/v1/events, subprotocol dodossh.events.v1 — and the server sends a line down it whenever something you can read has moved. The client's answer is the same delta pull it would have run on its timer, only now rather than in up to a minute. Two things you can see: an edit somebody else makes appears while you are looking at the list, and a vault shared with you turns up as soon as they share it.

What is on that socket is a notice, not your data. A frame says which vault changed and how far its change log has got, and nothing else: no item, no ciphertext, not even which item it was. That is the decision the rest of this rests on, and it is deliberate twice over — the server has nothing else it could send, and keeping it that way means there is still exactly one path that applies a change to your keychain, so the socket can be wrong or absent without anything being applied incorrectly.

Polling is still there and is still what guarantees a pass. The minute timer is unchanged. A network that eats WebSockets, a server with Events:Enabled off, an older server, a proxy that will not upgrade, a notice dropped because your machine was too slow to read it — every one of those leaves you with exactly what this product did before the socket existed. Nothing is reachable only this way, and nothing is supposed to become so.

Three limits are worth knowing rather than discovering:

One node. Fan-out is in-process, so a deployment running more than one API replica only pushes for writes that its own replica handled. The rest arrive on the timer. The seam for a PostgreSQL LISTEN/NOTIFY backplane is in place and is not implemented, because an untested backplane would be worse than a documented gap.
The socket does not outlive your access token. It is closed at the token's expiry and the client reconnects with a fresh one, which is a gap you will not see. That, plus re-reading your vault list every few minutes, is what bounds how long a withdrawn grant can keep producing notices — and what it bounds is metadata, because reading a vault needs a key the server has never held.
You are told about your own writes. Your client pushed, so it has already pulled; the extra pass finds nothing. Notices are coalesced over a quarter of a second so that a burst is one pass rather than a dozen.

The reasoning, including why this is a WebSocket rather than server-sent events and where a shared terminal session will attach to it, is in ADR 0012.

The Android head

src/DodoSSH.Client.Android is a phone-first head that shares every view model with the desktop one — the keychain and a terminal, which is the scope docs/android-port.md decided on and the reasoning behind it. Sign in, unlock, browse hosts, open a shell, and read the keychain; the two host-key decisions and the counted delete confirmations are there too, and none of them were softened to fit 360dp.

Its interface is the v2 design: destinations in a bottom bar, with the rest one tap deeper behind the last. The bar is three — Hosts, Connections and Settings — with the keychain, snippets, SFTP, S3 buckets, logs, vaults and preferences behind Settings. A bottom bar is for the places a session moves between, and managing keys is not one of those. Both heads are on that design now; the desktop's own v2 is a 190-pixel labelled nav rail in place of the icon rail, a centred search box in the titlebar, and session tabs as pills, and it keeps its Keychain entry — its rail has the room. Its light theme is not built — see docs/design-import-gaps.md — so the application is dark on both.

Connections is where a connection is made, not only where one is shown. With nothing open it offers a box taking user@host or user@host:port and a password, and lists the machines most recently connected to underneath. That box is the one path in the product to a machine the keychain has never heard of — the case somebody has just been handed an address — and nothing typed into it is saved: a machine worth keeping belongs on Hosts, where it can carry a key, a group's defaults and a name. Tapping a recent machine goes to its host if it has one and back into the box if it does not.

A connected phone shows one bar and then the terminal. The header, the session strip and the bottom bar are collapsed while a shell is up, and a single 35-pixel row replaces them: back on the left, the sessions as pills, and a + on the right offering the three connections this application can make — a shell, a host's files over SFTP, or a bucket. The system back gesture does what the arrow does, and lowers that menu first if it is open.

Widening the rail moved the desktop window's minimum from 880x560 to 1016x574, which leaves every screen exactly the width it was designed against.

A third desktop pass has since moved the furniture. The tab strip belongs to the window rather than to the terminal: Vaults, SFTP and S3 are fixed tabs at its head and open terminals follow them, which took SFTP and S3 out of the nav rail — they are the two destinations you stay in while something runs. The hosts screen became a grid of cards, groups above and hosts below, with a right-hand drawer for whichever host is selected and for both editors; the 268-pixel host sidebar is gone. Text is white rather than the design's blue-tinted #E3E7F4, and the type scale is a point larger.

File transfer is here now, in the shape scoped storage allows: one remote pane and the queue, over either an SFTP host or a bucket. There is no local pane, because there is no browsable local filesystem to put in one. So the way in is ADD FILES, which is the system document picker: point at a document wherever it lives and it goes to the directory showing, rather than choosing on the left and pressing an arrow. What Android hands back is a content:// URI with no path behind it and no promise of a seek, so the document is copied into the app's own cache and the copy is what the queue moves — which is what lets a stopped upload resume from where it stopped. The copy is deleted when the transfer finishes, kept while it is stopped so RESUME has something to read, and swept at the next launch.

SAVE FILE is the way back out, and it is the system's save picker for the same reason: there is nowhere this application could put a file that you would then be able to open. You choose where it goes before the transfer starts, the download runs into the cache, and the finished bytes are copied out to the document you chose. That order has one visible cost, and the screen says it rather than leaving it to be discovered: the picker creates the file when you dismiss it, so a download that then fails leaves an empty one there. The alternative is a picker that appears minutes later over whatever you moved on to — and often while the app is in the background, where Android will not show one at all. Hosts and groups are made and corrected here now, from a floating + on the Hosts screen, and both editors are cards in the list's own row rather than dialogs, so the form never covers the thing it is about. The keychain has no editor of its own: SSH keys and buckets are created on the desktop and sync down, and the phone will delete an item — behind the same counted confirmation — without offering to change it. What this head does make, it makes where the need arises rather than in an editor: a tag from inside a host's editor, and a credential from the connect bar's remember tick, which stores the password just typed and moves the host onto it. Renaming either is still a desktop job. Pins and import have no phone screen either, and importing an ~/.ssh/config has no meaning on a phone at all. VAULTS does have one, behind MORE, and it is there for a reason the design could not have anticipated: an invitation is claimed by signing in, so somebody being told they have been added to a vault is at least as likely to be holding a phone as sitting at a desktop, and a membership visible only on a head they have not installed is a membership they cannot see.

Port forwarding is not built anywhere, and the phone's More screen says so in a paragraph rather than leaving a gap. The v2 design draws a whole screen for it; nothing in the SSH layer forwards anything, so every control on that screen would have had no effect. See docs/design-import-gaps.md.

It is deliberately not in DodoSSH.slnx. Putting it there would make the android workload and a full Android SDK a prerequisite of dotnet build DodoSSH.slnx for everybody; it has its own CI job instead, which builds and packages it, because the two failures it is most exposed to — a native library with no Android ABI, and an assembly that resolves but will not dex — are both invisible to a plain compile.

Building it needs the workload and API 36 specifically:

dotnet workload install android

dotnet build src/DodoSSH.Client.Android/DodoSSH.Client.Android.csproj

API 36 is not a preference. Avalonia.Controls.WebView ships only a net10.0-android36.0 assembly, so anything lower cannot resolve it and the head loses its terminal. The floor is API 28, which is where BiometricPrompt and StrongBox-backed keys exist without an AndroidX shim.

Four things about it are worth expecting rather than discovering.

Signing in does not use the desktop's loopback redirect, and must not. On a shared device any other application can bind a loopback port and race for the authorization code — the attack RFC 8252 §8.3 names. The phone registers a redirect with the system instead and is handed the response as an intent. Everything above that — PKCE, the state check, discovery, the token exchange, the key binding — is the same code the desktop runs, because the only thing that varies is where the response arrives.

The redirect is a private-use scheme (dev.dodotech.dodossh:) rather than an Android App Link, and the limit is worth knowing: another app can declare the same scheme, and Android will offer a chooser rather than refuse. PKCE is what makes an intercepted code useless. An App Link closes it properly and costs an assetlinks.json on your own server's domain.

A fingerprint releases the device key, and re-enrolling a fingerprint destroys it. The key is generated with setInvalidatedByBiometricEnrollment, which is what stops somebody who can add their own fingerprint to an unlocked phone from inheriting the vault. The cost is that adding a finger legitimately means typing the passphrase again and re-registering — which the unlock screen treats as ordinary, because it is.

A notification appears while a shell is open. Android stops backgrounded processes, and the desktop client's promise that locking the vault does not close your shells is only true here behind a foreground service. The notification is the price of that promise; it goes when the last shell does.

The recovery code screen blocks screenshots. FLAG_SECURE is raised for that one state and lowered again afterwards, so the screen's own claim is true and a shell is still screenshotable. It stops the accident worth stopping — the only copy of an unrecoverable code landing in a cloud photo library, or in the recent-apps thumbnail — and stops nothing determined, since a second phone photographs a screen perfectly well.

Nothing has been run on a device. It compiles, links, packages, and carries the right native libraries for arm64 — that is verified, and CI verifies it on every change. Everything about its runtime behaviour is not, and docs/android-port.md says which claims those are.

End-to-end verification

One suite runs against a real server rather than a stub. It needs a Docker daemon and nothing else, so it is part of the ordinary test run:

dotnet test tests/DodoSSH.SystemTests

It brings up PostgreSQL, Keycloak and an OpenSSH server in containers, applies the committed migrations, starts the API as a child process out of its own build output, and then drives the real client: sign in through Keycloak, enroll, unlock, create an SSH key and a host bound to it, sync them, open a shell on the sshd and approve its host key at the real first-contact refusal, then read all three back on a second simulated machine and unlock again with no network. Roughly 25 seconds once the images are pulled.

What makes it worth its weight is that it consumes the artefacts that ship — the realm file from deploy/keycloak, the EF migrations, the API's own appsettings — rather than a fixture written to match them. On its first run it found a loopback redirect URI the realm registered in a form Keycloak rejects, and a JSON configuration gap that made the whole sync surface unreachable from the real client while every other test passed. Both are the same class of bug: two sides of a stub agreeing with each other about something the specification never said.

The one value it cannot take from a committed file is Oidc:Authority, since the container's port is assigned at start. Everything that authority points at is still the real realm.

Development is Windows-first, but the full suite now runs on Linux too, and CI runs it there on every change. Getting there cost three fixes rather than none, and each was a real difference instead of a test being fussy: the local file pane built its roots bar from every mount the kernel holds, one assertion recognised the profile directory only by its Windows capitalisation, and the layout harness pinned a COM error that only Windows raises. macOS is still unverified.

Anything known or suspected to differ is tracked in docs/platform-flags.md, along with the deployment gotchas that have already cost time once. Read it before assuming something works off-Windows.

Android has been audited and scoped, but not started: docs/android-port.md records what ports as it stands (most of the core), what does not (most of the interface), the decisions taken about what an Android client would be — phone-first, keychain plus a terminal — and the spike that gates all of it.

Conventions the build enforces

One version for the whole repository, derived from the nearest v* tag by MinVer. A tag build whose computed version disagrees with the tag fails CI, and every checkout uses fetch-depth: 0 — without it MinVer answers 0.0.0-alpha.0.N rather than failing, and a wrong version here is a client that never updates.
Warnings are errors, formatting included: IDE0055 is an error in .editorconfig, so a misformatted file fails the build itself rather than a separate CI step.
Package versions are centralised in Directory.Packages.props; packages.lock.json is committed and CI restores in locked mode.
BannedSymbols.txt bans DateTime.UtcNow (use TimeProvider), Guid.NewGuid (use CreateVersion7), sync-over-async, MD5/SHA1 and PBKDF2.
Public members of DodoSSH.Contracts must be declared in PublicAPI.Unshipped.txt, so a contract change is a build error rather than a client-side surprise.

Milestones

M0 — foundation. Repo structure, build conventions, CI, ADRs. Done.
M1 — vertical slice. OIDC login → enroll → create a host → open a shell. Server done: the DSH1 crypto core, the data model, sync push/pull for hosts, /me, and enrollment with the identity-provider key binding. Client done: the key hierarchy, the OIDC flow with the key binding, SSH connections with host key trust, the terminal data plane, the encrypted local cache with the sync client — offline unlock, an outbox and a field-level three-way merge, conflict matrix green — and an Avalonia shell that is vault-backed: server URL → browser sign-in → enroll → unlock → host list → terminal. The shell's state machine is covered by tests against an in-memory server, so the states that matter most (the recovery code that cannot be skipped, the unlock that needs no network) are checked rather than remembered.

Its layout is not covered by anything, and that gap has already cost a shipped defect: the setup and unlock screens were layered over the terminal's WebView, which on Windows is a native child window that cannot be covered, so they rendered sliced with their buttons unclickable. No test in this repository loads a .axaml file, and a headless one could not have caught this — there is no native window in headless, so it would have rendered perfectly and confirmed the wrong belief. Screens get looked at, or they are unverified. Verified end to end: tests/DodoSSH.SystemTests drives the whole slice against a real Keycloak, a real API, a real PostgreSQL and a real sshd — sign-in, the identity-provider key binding, enrollment, offline unlock, a host and an SSH key through the vault to a second machine, an interactive shell, and the host key approved at that shell's prompt reaching the second machine as well. See End-to-end verification.

The two gaps this milestone shipped with have both closed since. A vault credential can be created — from the Vault screen, or from the REMEMBER tick beside the connect password, which saves it and binds the host to it once the remote has accepted — so password authentication asks once rather than every time. And a device key is registered where the machine can hold one: Windows keeps it in the TPM under a CNG policy that makes the consent dialog a condition of using the key rather than a prompt this application draws, which is stronger than ADR 0007 originally described and is why that ADR was corrected. What is left is the floor rather than a gap: a machine with no TPM, or a desktop that is not Windows, gets a store that reports itself unavailable and keeps asking for the passphrase — the honest answer rather than a degraded one.

Host key trust is in the vault, which is what makes trust-on-first-use worth having: a fingerprint approved on one machine is approved on all of them and survives a restart, and the server cannot drop a pin to force a fresh first-use decision without the item visibly going missing. A changed host key stays a hard refusal with no way past it; withdrawing a pin is a separate, deliberate act in the host's editor.

Binding a key introduced the first payload schema version bump, and it is worth knowing how it behaves: a host is written at the lowest schema version that can represent it, so only hosts that actually bind a key are written at version 2 and become read-only on an older build. Hosts that do not are still written at version 1, byte-identically to before the field existed — which is what keeps upgrading one machine from making a team's whole vault uneditable everywhere else.
M2 — full personal vault, robust sync, relay. File transfer done: an SFTP session, a two-pane file browser with a real remote listing — names, sizes, modification times and drwxr-xr-x permission bits — and a queue that moves one file at a time with progress, throughput and resume. See Moving files for the two things about it worth knowing before you use it, both of which are consequences rather than choices.

Organising done: hosts can be filed into groups, and commands can be saved as snippets. Both are ordinary synced items — encrypted, merged and pushed like every other — and both are invisible until used: a keychain with no groups draws the flat host list it always did. Two things about them are deliberate. Group membership is a field on the host rather than a member list on the group, so filing two machines at once on two laptops is two independent writes instead of one contested one; and a group carries the port, username and key the hosts inside it fall back to, read at connect time rather than copied in, so changing one changes every host that never overrode it.

Groups nest, and that reverses an earlier decision worth recording. They were flat because a parent pointer merged field by field lets two offline clients build a cycle nothing upstream can see — the pointer is inside the payload, so the server cannot read it, and the merge resolves one item against one item and never sees the pair. That is still true. What changed is that inheritance made the chain something the connect path walks, so the answer had to be a walk that terminates whatever it is handed: every walk carries a visited set and stops at a repeat, which degrades a cycle to a group reading as a root rather than to a shell that never opens. Given a walk that had to be cycle-safe anyway, refusing to nest bought nothing.

Membership living on the host has one further consequence, and it took two goes to settle. Deleting a group could not clear it without rewriting every host under the heading, so at first it did not: the hosts kept an id that resolved to nothing and turned up under UNGROUPED, which reads identically and costs no writes. That held until the deletion had to be able to take the hosts with it — a group is sometimes a heading being tidied away and sometimes a project that has been decommissioned, and nothing in the code can tell which. Once a deletion knows which hosts it means, leaving them naming something that has gone is a state kept for no reason, so both answers now write: N deletions, or N hosts with the reference cleared. The dangling case still has to be survived everywhere it is read, because a group deleted on another machine arrives exactly that way.

Inserting a snippet types it at the prompt and stops. Pressing Enter is a per-snippet decision, off by default, and the reason is worth stating: a terminal is one input stream with no notion of being at a prompt — the remote may be in an editor, or at a password prompt with the echo off — so this client cannot honestly say "run this command", only "type this into whatever is there".

Logs done: what has been connected to, and what has been changed in the keychain. Both are synced, encrypted items rather than local files, because the point of them is auditing a shared vault — a log only one machine can read is a diagnostic, not an audit trail. Two consequences are stated rather than implied. The connection log records the host's name, the address dialled, when, for how long and by which account on which machine — but not the SSH username, which is a detail of the host and is in the host's own logs. The keychain log records the names of the fields an edit touched and never their contents.

What that costs is in ADR 0001: the server still cannot read a single field, but one row per connection with a server-side timestamp tells it your connection rate and the hours you work. Retention bounds it — ninety days or five thousand entries per kind, whichever bites first.

Buckets done: S3-compatible object storage is a second kind of remote on the Files screen, beside a host. Prefixes are directories, objects are files, and transfers go the same way through the same queue. The bucket, its endpoint and its keys are a keychain item like any other, encrypted end to end — which matters more than usual here, because for anybody self-hosting MinIO or Ceph the endpoint is an address on their own network.

Three things a bucket cannot do are refused with a reason rather than approximated: there are no directories, an interrupted upload starts again rather than resuming (an object cannot be written from the middle), and a rename is a copy and a delete rather than one atomic operation. Downloads do resume — a ranged GET is part of the protocol, which is the one place a bucket beats SFTP.

Realtime done: a WebSocket the client holds open, over which the server says which vault has moved so a pull happens now rather than within the minute. What crosses it is a notice and never an item, which is what keeps one code path applying changes and makes a dropped socket cost latency and nothing else — the timer is unchanged and is still the guarantee. Two limits are stated rather than implied: fan-out is in-process, so a multi-replica deployment falls back to the timer for writes another replica handled, and a socket is closed at its access token's expiry rather than outliving the credential that authorised it. See Changes that do not wait and ADR 0012.

It is also the transport a shared terminal session will use — one person's shell, watched or driven by somebody else. Nothing of that exists yet, and ADR 0012 records the one decision made early so it need not be renegotiated: session data will be binary frames on this same socket, because base64 in a JSON envelope is the wrong shape for the one payload here that is continuous rather than occasional.
M3 — shared vaults, sharing, ACLs. Done. Membership with roles, a public-key directory, the append-only key log served for clients to verify against, shared vaults, and vault key grants wrapped by a client and stored opaquely by the server. VaultAccessService now resolves team membership to permissions, so a viewer may pull and may not push; the desktop client reads and syncs every vault it holds a key for, and a real VAULTS screen replaces the placeholder. The screen is vault-shaped rather than team-shaped: naming a vault makes the membership list that carries it, so the team is behind the vault rather than a thing anybody has to create first. A vault can be renamed, deleted and handed to another member; a member row carries when that account was last here; and an address with no account on this deployment can be invited, joining the moment somebody signs in with it. See Sharing a vault for the one distinction the whole design rests on, and the limits worth knowing before you rely on it; the reasoning is in ADR 0009.

Membership changes now move the keys, not just the flag. Adding somebody wraps every team vault the adding machine can open to them — every generation of each, so they can read the vault's history and not only what happens next. Removing somebody revokes their grants, advances each vault it can open to a fresh key generation in one server transaction, and wraps that key to the members who remain. What a rotation buys is exact: everything written from then on is unreadable to the person who left. The items already stored are then re-sealed under the new key as well, item by item and resumably — which is safe to do incrementally precisely because a vault at mixed generations stays readable. A change queued before the rotation is re-sealed as it is pushed, so nothing reaches the server under a superseded key at all. See ADR 0010.

A vault shared with you arrives at once, with no sign-in and nothing to press. Each pass asks the server which vaults this account can reach before syncing the ones it already knows — which is also how a vault that has been deleted, or one whose grant was withdrawn, stops being listed — and the server now says so the moment somebody wraps a key to you rather than leaving it for the next pass. Without a reachable socket that becomes "within the minute", which is what it always was; see Changes that do not wait.

Ownership transfer is here, and it is one write rather than two. The member you name becomes owner and you become an admin, in a single transaction — because ownership is sole, so promoting first leaves the team owned twice and demoting first leaves it owned by nobody, and there is nobody left with the authority to finish a transfer that stopped in the middle. Nothing else is touched: you keep your vault key grants, because removing the outgoing owner would revoke them and rotate every team vault, which is a much larger act than the one being asked for.
M4 — hardening and ops, packaging, self-hosting guide. Decided ahead of the work, because the first release takes it irreversibly: who signs the client and where it comes from. ADR 0011 puts the release key with the project rather than with a store, and rules out the arrangement a self-hosted product reaches for by default — the deployment serving the client binary, which hands it to the one party the whole trust model is about. An installed Android app can only ever be updated by a package signed with the same key, so this is the first release's decision to make and nobody else's afterwards.

The Windows desktop half is built. Velopack packaging for win-x64, a Setup.exe that installs per user with no administrator prompt, and a client that checks the project's own forge every six hours, fetches a newer build in the background, and then waits for a restart the user presses — because a restart ends every shell, and this application has gone to some trouble to make locking not do that. The version is now one number for the whole repository, derived from the v* tag by MinVer, which is also the first time the API has reported a true serverVersion. Releases are cut by a person rather than by CI: the token that writes a release is, for an updater that trusts its feed, the same capability as the signing key, which ADR 0011 rule 1 keeps off runners. See ADR 0013, and Installing on Windows for what a user sees.

Still to do here: signing (the first release is unsigned, and the trigger for buying a certificate is the first release aimed at strangers), and macOS and Linux packaging.
M5 — multi-provider OIDC, identity key rotation, per-item content keys.

Licence

MIT.

README.md

DodoSSH

Why