~/workspace

A private, terminal-themed workspace that runs in the browser: a diary, the notes for work and for uni, and a board per space. One user, one SQLite file, one binary. It was called narl-diary until it grew everything but the diary.

• Rust + axum backend, SQLite storage, Svelte 5 frontend.
• Spaces (diary, work, uni, …) with folders nested to any depth inside them.
• Markdown documents with full-text search across a whole space (SQLite FTS5), and math in them set by KaTeX: $x^2$ inline, $$…$$ or a ```math block on its own.
• Import a folder — or a zip of one — as a tree of documents, links and all.
• A kanban board per space, so what is open at work is not mixed up with uni.
• Drag, drop or paste images, video and audio straight into an entry.
• Files of any kind — PDFs, archives, recordings — live in the tree beside the notes about them, streamed to disk however large — and PDFs, text, code, tables and certificates open in place.
• Share any document behind an unguessable link — revocable at any time.
• A vault: files sealed with AES-256-GCM under a password that is never stored, readable only while it is unlocked.
• Backs itself up to Proton Drive, end-to-end encrypted, as a registered device.
• Mails you — a reminder when the diary is empty, due cards, a weekly digest.
• Real vim keybindings on the desktop (CodeMirror 6 + vim mode), plain taps on mobile.
• The whole frontend is embedded in the release binary, so deploying is copying one file.

Quick start

cp .env.example .env      # set WORKSPACE_USER, WORKSPACE_PASSWORD and WORKSPACE_SECRET
make build                # builds the frontend, then the release binary
./target/release/narl-workspace

Then open http://127.0.0.1:4242.

Every setting is read as WORKSPACE_<NAME>, with the older DIARY_<NAME> spelling still accepted behind it, so a server configured before the rename keeps running untouched.

WORKSPACE_SECRET signs the session cookie; generate one with openssl rand -base64 32. Changing it logs every device out, and so does changing WORKSPACE_PASSWORD or :logout!, which ends every session but needs no restart. Every signed-in browser is a row on the server: :sessions lists them and :sessions rm <n> signs one out — a lost phone, say — without touching the rest, and locks the vault as well. A setting that is set but cannot be read — a number out of range, a switch that is neither on nor off — stops the start with the reason, rather than being taken for its default. The database and uploads live under WORKSPACE_DATA_DIR (./data by default) — that directory is the entire backup, and it is exactly what the Proton Drive mirror copies.

Development

make dev-api   # cargo run, serving the API on :4242
make dev-web   # vite dev server on :4243, proxying /api to :4242

In debug builds the binary reads web/dist from disk, so cargo run alone also works once the frontend has been built at least once.

Keys

The desktop UI is modal, and a key acts on whatever has focus: in the list that is the row under the cursor, even while another document is open beside it; in the document pane it is that document. In the browse pane:

key	action
1 … 9, 0	switch to that space
Shift-1 … 9	send the marked nodes, or the selected one, to that space
Tab, Shift-Tab	next / previous space
w	overview of every space; hjkl then Enter
f	fullscreen the focused pane
Alt-h / Alt-l	focus the list / the document, from the editor too
j / k	move down / up, or scroll the document when it has focus
gg / G	first / last entry, or top / bottom of the document
Ctrl-D / Ctrl-U	half a page down / up the open document
Enter, l	open the highlighted entry
o	a new document here, straight into insert mode
O	a new folder here
i, a	edit the open entry
R	rename what is highlighted — a document, a folder or a file (:name! renames the space)
m / p	pick nodes up — m again puts one back, M takes the whole folder — then drop them all into the folder you walk to
Ctrl/Shift-click	pick a row up / every row from the cursor to it; drag picked-up rows onto a folder, a crumb or a space
b	the board of this space
t	the agenda: what is late, due this week, or in hand, on every board
T	today's diary entry, started if there is none
c	capture a card into this space's inbox without leaving where you are
Esc, q	close the document
h	up one level in the tree
/	full-text search in this space (n clears it)
Ctrl-P	full-text search across every space; Enter goes to the match
s / y	toggle sharing / mint a new share link
x, dd	delete the entry under the cursor (asks y/N)
:	command line
Ctrl-S	save from anywhere
?	help

Inside the editor the full vim keymap is live — w/b, dd, ciw, visual mode, macros, :w to write. :help lists every ex command; the useful ones are :w, :wq, :q!, :new [yyyy-mm-dd], :date, :name, :share, :link, :upload, :media, :search, :backup and :set novim. The rest: :today opens today's entry, :edit [id] (or :12) opens a node by id, :next, :prev, :first, :last and :random walk the list, :delete [id] deletes, :copy puts the whole document on the clipboard, :clear drops the search, :reload re-reads the folder, :stats counts words and lines, :search! searches every space, :set clipboard / :set noclipboard mirrors vim yanks to the system clipboard, :logout ends the session, :logout! every session and :sessions rm <n> one other. The tree has its own: :space [name] switches (:space! <name> makes one, :rmspace [name] deletes one and everything in it), :mkdir <name> makes a folder, :cd <name|..|/> walks it and :mv <folder|..> moves what is selected, and :import uploads a folder into the one being browsed (:import! takes a .zip instead). :upload outside the editor puts files into the folder being browsed, and so does dropping them on the list. :name <text> renames whatever is under the cursor — a document, a folder, or a card on a board — and :name! renames the space itself. The board has :board, :card <title> (:card! links the open document to it), :due <when|->, :done and :list <name> (:list! deletes one).

On a touch device modal editing is off by default (:set vim turns it on) and the same actions are buttons: the ⋯ on a row — or a long press — opens what the keys do to it: rename, move, share, seal, delete. A move is picked up there and dropped with the button the dock then shows, in whichever folder is open. Inside a folder the list starts with .., the way up. The dock also has today's entry, an upload into the folder, a search across every space, and on the board the card moves, renames and deletes; the agenda has a done button per card.

The SFTP mount

The workspace can be mounted as a filesystem over SFTP, so an editor or a file manager on another machine works on it directly. It is off until WORKSPACE_SFTP_BIND is set (127.0.0.1:2222, or 0.0.0.0:2222 inside a container, with the port published in compose.override.yml). Only public keys get in, and only the ones registered from inside the workspace:

:sshkey add laptop ssh-ed25519 AAAA… you@laptop
:sshkey              lists them, with their fingerprints
:sshkey rm laptop    and this one is out, mounted or not
:sshkey vault laptop on    this key may read sealed files while the vault is open

sshfs -p 2222 me@your-host:/ ~/workspace     # the user name is ignored

The root holds the spaces, mkdir there makes one, and anywhere else a folder. A document is <slug>.md, read as # Title followed by its body, and a leading heading written back renames it. The server's host key is made on first start as sftp_host_key in the data directory.

Boards

b, the board door on the rail, or :board swaps the tree for the board of the space being browsed, and swaps it back. Every space has one except the diary — a journal has no backlog — and :set board / :set noboard changes that for the space you are in (the dimmed board door on the rail does the same). A new board starts with backlog, doing and done, and the columns mean something by where they stand rather than by their names: the first is the inbox new cards land in, the last is done. A card is finished exactly when it sits in the last column — moving it there stamps it, moving it out reopens it — so a tick and a column can never disagree; marking a card done through the API moves it there too, and deleting the last column makes the one before it done. :list <name> adds a column in front of the last one, :name! on the board renames the column under the cursor, and done cards older than a week fold away until z shows them.

c captures from anywhere: :card <title> outside the board puts the card in this space's inbox (or the first space that has a board) and leaves you where you were. On the board, o or c puts a card on the column under the cursor.

key	action
h / l	to the list on the left / right
j / k	down / up the cards of that list
J / K	move the card down / up its list
H / L	move the card to the list on the left / right
Enter	the card's notes: its document, made on first use; q comes back
Space	into the done column, or back out to the inbox
o, c	a new card on this list
D	due date — fri, tm, +3d, +2w, 10-31, 2026-10-31
R	rename the card, or the column when it has none
z	show done cards older than a week
x, dd	delete the card
q, Esc	back to the tree

A card can point at a document: :card! while reading one makes a card for it, and Enter on that card opens it again. Enter on a card without one makes a document named after the card in the space's root and opens it to write in, and closing that document returns to the board on the same card. The link is deliberately loose — deleting the document leaves the card standing, because the work is still open once its notes are gone. Only a document can be a card's notes; a folder or a file is refused.

The agenda

t (or :agenda) gathers every unfinished card from every board that is late, due within the week, or sitting in a column between the inbox and done — the work already in hand — and says whether the diary has an entry for today. j/k walk it, Enter goes to the card on its own board, Space finishes it where it stands, and T writes today's entry. The bar shows late and today counts whenever either is above zero, and the agenda opens by itself once a day when something is late or due, so nothing has to be remembered to be looked at.

Not losing things

Every change is written to localStorage as it is typed and to the server a couple of seconds after typing stops, so :w is a habit rather than a necessity. When the server cannot be reached the draft stays in the browser and the save is tried again, less often the longer it fails, and at once when the browser says it is back online. A session that runs out mid-sentence shows the login, and logging in again carries on where it stopped, draft and all.

A document saved somewhere else since it was opened here — another tab, the phone, the SFTP mount — is not written over: :w says so, :e! takes the other version, and :w! writes this one anyway. Search results say which space and folder each match is in. A document made with o and left with neither a name nor a word in it is removed again on the way out, so abandoned starts do not pile up. If a tab dies mid-document, opening it again restores the draft and says so — :e! throws it away, the way vim handles a swap file.

:export downloads the open document as markdown. :export! downloads the whole workspace as a zip: the tree as real directories, empty folders too, one markdown file per document, every file in the tree where it sits and every other embedded file in media/, under their real names, and relative links between documents, folders and files, so the archive reads in any markdown viewer without this application. What plain files have no place for — when each entry was written, names as they were typed, which space is the diary, every board with its cards — goes into one .workspace.json beside them.

The diary is the diary by role, not by name: rename it and the reminder and the agenda still know which space it is, and it cannot be deleted — empty it instead. Before a new build migrates the database it keeps a copy of it as it was, diary.db.pre-<version>, next to it in the data directory.

An upload that nothing embeds any more — pasted, then deleted from the document — is kept a week and then removed, along with what a killed upload or a crash leaves in uploads/. That runs at start and once a day. A file in the tree is never one of them: it is deleted from its folder, not from :media.

Files

A file is a node like a document, only with bytes instead of text: it sits in a folder, shows its size in the list, and Enter opens it. What is shown in place:

• images, video and audio;
• PDFs, in the browser's own viewer;
• text and source code of any kind, recognised by its name or, failing that, by its first few kilobytes, with line numbers and the first 256 KiB of it;
• markdown, rendered, with source to see it as written;
• CSV and TSV, as a table of the first 500 rows;
• certificates, keys and keystores — .pem, .crt, .cer, .der, .csr, .key, .p8, .p12, .pfx — read on the server: subject, issuer, validity, key, names, usage and fingerprints, and which key belongs to which certificate. A .p12 or an encrypted key asks for its password, which opens it for that one look and is never stored. Nothing of a private key is ever sent back but its algorithm and size.

Anything else is a download. :name renames it, and the download is called whatever it is called now; :mv, m and p move it; x deletes it. :upload picks files for the folder being browsed, or drop them onto the list. Each upload is written to disk as it arrives, never held in memory, so the only ceiling is WORKSPACE_MAX_UPLOAD_MB (1024 by default) — which is per request, so a proxy in front needs a body limit at least that large too.

The SFTP mount shows files beside the documents, under their own names — a later file that shares one is told apart by its id, slides (42).pdf — and anything copied in that is not a .md becomes a file in the tree, streamed to disk under the same size limit. A file copied into a sealed folder is sealed on the way in while the vault is open in the browser. Reading one back takes more: a key is a standing credential kept on another machine, so it reads sealed files only once given vault access with :sshkey vault <name> on, and none has it to start with. Such a key, while the vault is open, opens a sealed file decrypted as it is read, and it is sealed again when it is saved, wherever it sits. While the vault is shut, or once the access is taken away, both are refused, and either cuts off a read already under way — though a client that cached what it read keeps that copy until it lets go of it. Nothing done over the mount holds the vault open; only the browser does. A session holds at most 256 handles and 64 MB in memory, the server at most 32 connections, and one whose client has stopped answering is dropped within a few minutes. Editor droppings — swap files, ~ backups, dotted temp files — stay in memory for the session, and a dotted temp file renamed onto a real name is kept as that file, which is how rsync and most editors save.

If a document is saved in the browser while an SFTP client holds it open, the client's save does not overwrite it: it lands beside it as <name>-conflict.md, and both are kept. Deleting a key in the browser ends what that key mounted, not just its next login.

The vault

Some files deserve more than the login: scans of documents, contracts, health records. The vault encrypts them at rest under a password of their own, so they stay unreadable to someone who has the data directory or the Proton account the mirror writes to, and to a logged-in browser while the vault is shut — to anyone, in short, who does not know the vault password.

:vault sets the password the first time (ten characters at least) and says whether the vault is open afterwards. :upload! puts files into the folder being browsed sealed — encrypted as they stream in, so the plaintext never touches the disk. :encrypt seals the file under the cursor; :decrypt takes it back out. :unlock and :lock open and close it, :passwd changes the password. A sealed file carries a lock in the list, and the bar says vault open while it is.

:passwd makes a new master key as well as a new password. The old password stops opening anything at once; every sealed file is then re-sealed under the new key in the background, under a new name, while the vault is open — :vault shows how far it has come, and locking halfway only pauses it until the next :unlock. The old copies are deleted from uploads/ and, on the next mirror run, from Proton Drive for good. After that, a copy of diary.db from before the change (a diary.db.pre-* snapshot, an old revision in Proton Drive) and the old password open nothing that is still kept anywhere. What they cannot take back is a copy someone already made of the files themselves. Re-sealing rewrites only each file's header, but every sealed file is uploaded to Proton Drive once more. A second :passwd waits until the first has finished.

:encrypt on a folder or space seals it for good, not just once: every file in it now, and every file that lands in it later. An upload, an import, an SFTP write and a file or folder moved in are all sealed on the way in. While the vault is shut they are refused rather than stored readable, and the browser asks for the password instead. Folders made inside are sealed too, and a sealed folder carries a lock like a sealed file. :decrypt on that folder unseals it with everything in it; a file or folder inside it cannot be decrypted on its own. Documents are not files and stay readable, as below. The database holds the line as well: a file node whose bytes are not sealed cannot be put in a sealed folder by any path.

The vault closes itself after WORKSPACE_VAULT_IDLE_MIN minutes of nobody using it (15 by default) — the key is dropped from memory then, not merely refused on the next look — on logout, and whenever the server restarts. While it is closed a sealed file shows as sealed and nothing more; a share link never serves one, open or not; and :export! refuses until it is open, because an archive meant to be read without this application has to hold the files opened.

How it works: Argon2id turns the password into a key that seals a random master key, and only that sealed copy is stored. Each file gets a random key of its own, sealed under the master key in the file's header together with its id, and its bytes are AES-256-GCM in 64 KiB records, the last one marked as last — so a file that is swapped, cut short or tampered with refuses to open. Changing the password makes a new master key. Until every file is re-sealed, the old one is kept sealed under the new one, so either kind of file opens and an interrupted re-seal carries on where it stopped; re-sealing a file wraps its own key under the new master key and copies its records as they are.

A sealed file is served with Cache-Control: no-store, so the browser keeps no copy of what it showed once the vault shuts. A sealed image in a document shows as a button while the vault is shut, which asks for the password and puts the image back. :share says how many sealed files a document holds, since its readers will not see them, and :encrypt says how many shared documents just lost a picture to the vault. :encrypt on a folder seals what it can and counts what it could not, rather than stopping at the first.

Only the bytes are sealed. A sealed file's name, and the names and text of the documents around it, stay readable in diary.db — and in the mirror of it — like everything else outside the vault.

There is no recovery. Forget the vault password and every sealed file is gone — from the server and from every backup — which is exactly what makes it worth having. The password is not in .env, not in diary.db, and not in the mirror.

Sealing a file that already exists writes it anew under a new id, rewrites the links of documents that embed it, and deletes the plaintext from the disk. The mirror then deletes the old copy from Proton Drive outright — trash and all — whatever WORKSPACE_BACKUP_PRUNE says, and takes the file out of the readable files/ view. The sealed copy goes up to data/uploads/ like any other. Proton cannot read either; only the vault password can read the sealed one.

Importing a folder

:import picks a folder and uploads it whole — every file carries the path it had inside, and the tree is rebuilt on the other side: directories become folders, .md and .markdown files become documents (titled by their first heading), anything else becomes a file in the tree. :import! does the same with a zip, which is the shape :export! writes. :import! / pours an :export! back in at the root: each directory becomes a space again — or goes into the space already there by that name — and with .workspace.json beside it the dates, names, boards and cards come back as they were, embedded files go back to being embedded rather than a media folder, and a file that happened to end in .md stays a file.

Links between the imported files are rewritten as they land: a link to a markdown file becomes the document's address, a link to an image becomes its media URL, and anything pointing outside the upload — a URL, a file that was not uploaded — is left exactly as it was written. Absolute paths, .. segments and anything a zip smuggles in are refused, and both the upload and what a zip unpacks to are bounded by WORKSPACE_MAX_UPLOAD_MB.

A document has a real address, /n/<space>/<folder>/<slug>, which is what those rewritten links point at and what the address bar shows, so any document can be linked to from outside — behind the login, unlike a share link.

Backing up to Proton Drive

The server can mirror itself into Proton Drive, where it appears as a device — its own sync root, alongside the desktop clients — rather than as a folder dropped in My Files. Everything is encrypted client-side before it leaves the machine, by the same Rust SDK the Linux client uses, so Proton stores a workspace it cannot read.

It is a third-party client: it identifies itself to Proton as external-drive-narl_workspace@<version>-alpha and says so before it asks for account details. It carries no Proton branding and is not supported by Proton.

Set it up once, interactively, because SRP and 2FA need a human. From the workspace:

:backup login <proton-username>

It shows the disclosure above and asks before anything else, then asks for the password and, when the account has one, the two-factor code. If Proton wants a human check first, its own page opens in a sheet; solve it and the login goes on. An account whose only second factor is a security key cannot log in from here — add an authenticator app to it. :backup logout forgets the session again. The mirror starts within a minute of the login; no restart is needed.

The same login runs from the server's terminal, for a check that a terminal can answer:

narl-workspace proton-login                       # or, in Docker:
docker compose exec -it workspace narl-workspace proton-login

That checks the password against the account's keys and stores a session — the tokens, and the key passphrases derived from the password at login — as a 0600 file next to the database. The password itself is never stored: the passphrases unlock the keys but cannot log in, change the password or turn 2FA off. They are still enough to read the account, and the file sits on the same volume as the workspace it protects, so the volume is the thing to keep private. A session stored by an older build, which kept the password, is rewritten without it on the next start; no new login is needed. Built with --features keyring on a host that has a Secret Service, the session goes to the OS keyring instead and the file is only a fallback; the container has no session bus, so there it is always the file.

In a container, WORKSPACE_SESSION_KEY_FILE does what the keyring would: the file is sealed with AES-256-GCM under a key read from somewhere other than the volume, so a copy of the volume alone is no longer a session. A Docker secret is the place for it:

mkdir -p secrets
openssl rand -base64 32 > secrets/session.key
sudo chown 10001 secrets/session.key && sudo chmod 400 secrets/session.key

then the secrets: lines from compose.override.yml.example, and WORKSPACE_SESSION_KEY_FILE=/run/secrets/session_key in .env. A plain session already on the volume is sealed the next time it is read. Lose the key and the session goes with it; proton-login makes a new one.

Afterwards the server resumes on its own and nothing prompts again. Refresh tokens are single-use, so every rotation is written back immediately — which is also why two servers must not share one session file.

The mirror is one-way and change-driven: a write marks the workspace dirty, and once it has been quiet for WORKSPACE_BACKUP_DEBOUNCE_SEC the mirror runs, with WORKSPACE_BACKUP_INTERVAL_MIN as a backstop. An hour of writing is one backup, not sixty. The device folder ends up as a copy of the data directory:

narl-workspace/          the device — Proton allows only folders in a device root
  data/              the data directory, copied
    RESTORE.txt      what this is, and how to put it back
    diary.db         a VACUUM INTO snapshot — consistent, no write-ahead log
                     (the file keeps the name it was created with)
    uploads/<uuid>   every uploaded file, under the name the database knows
  files/             every file node again, as <space>/<folder>/<name>

diary.db becomes a new revision each time, so Proton Drive keeps the older ones and a mistake that was mirrored can still be undone. Uploads are written once and never rewritten. Restoring is copying the contents of data/ back into an empty data directory — no tool in between, which is the point.

command	what it does
narl-workspace proton-login	log in and enable backups
narl-workspace proton-status	account, device, schedule, how much is mirrored
narl-workspace backup-now	mirror once and exit — for cron, or for nerves
narl-workspace proton-logout	forget the session; the mirror stays where it is

From inside the workspace, :backup says when the last one finished and :backup! runs one now. A failed backup is loud: it is reported by :backup, and the next tick retries it.

files/ is for reading, not restoring: the same bytes as data/uploads/, laid out the way the tree shows them, so a PDF can be opened from Proton Drive's own apps by its name. It follows the workspace — a rename or a move there is a rename or a move on Drive, not a second upload, and a deleted file goes to Drive's trash whatever WORKSPACE_BACKUP_PRUNE says, since data/ still has it. Two files that share a name in a folder are told apart by the later one's id, slides (42).pdf. It costs the storage of every file twice; anything put into files/ by hand may be trashed, because the workspace owns that folder.

Deleting an entry does not delete it from the mirror unless WORKSPACE_BACKUP_PRUNE is on. A backup that forgets on command is one accident away from being no backup at all.

Mail

Off until WORKSPACE_SMTP_URL is set, and then four kinds of mail, all plain text and all addressed to WORKSPACE_MAIL_TO:

kind	when
reminder	at WORKSPACE_REMINDER_AT, if nothing was written in the diary space that day
card_due	at WORKSPACE_CARD_DUE_AT, for every card due that day, and again once it is overdue
digest	at WORKSPACE_DIGEST_AT: what was written per space, what was finished, what is still open
login	a browser signed in that has not signed in before, or a run of wrong passwords

WORKSPACE_MAIL_KINDS is the list of kinds that may be sent; drop one to switch it off. Every "today" is asked in WORKSPACE_TIMEZONE, because whether an entry was written today depends on where you are, and a day is not always 24 hours long where the clocks change.

Nothing is sent straight from the code that decides a mail is due. It goes into an outbox table under a key that names the occurrence — reminder:2026-09-18, card_due:41:2026-09-20:overdue — and a tick a minute drains it. A restart cannot resend yesterday's reminder, because the row is already there; a relay that refuses cannot lose one, because the row is still there and is retried with a widening gap, up to an hour apart. After ten refusals a mail is left for mail-status to show and no longer stands in front of the ones behind it.

A card mail is checked against its card before it goes: one finished, deleted or re-dated since it was queued, or on a board that has been switched off, is dropped unsent. Mails that are done with are forgotten after 90 days. The reminder needs a diary to be about, and says nothing when there is none.

The login mail says a new browser, not a new person: the device is remembered as a SHA-256 of its User-Agent, which is honest about what it can tell and needs no proxy header to be trusted for it.

narl-workspace mail-test      # one mail, straight out, bypassing the outbox
narl-workspace mail-status    # what is configured, what is queued, what failed

SMTP settings are wrong the first time, always. compose.override.yml.example carries a mailpit service for exactly that: point WORKSPACE_SMTP_URL at smtp://mailpit:1025 and read the reminders at http://127.0.0.1:8025 before involving a real relay.

Sharing

:share mints two 192-bit random secrets and copies https://your-host/s/<token>#<key>. That page needs no session, renders the entry read-only, and serves only the media files that entry currently embeds — dropping a picture out of a shared entry immediately makes it unreachable through the link.

The two halves do different jobs. The token names the entry; the key proves the reader was given the whole link. A browser never sends a fragment to the server, so the key stays on the reader's machine and out of every access log, referrer and link preview along the way — a crawler or a chat client that harvested only the path holds a URL that answers 404. The server stores nothing but a SHA-256 of the key and compares in constant time, and a wrong key is answered exactly like an unknown token, so guessing never confirms that a token is live.

The page sends the key on to the server once, in the body of the request that reads the entry, never in a URL. That answer gives the browser a cookie scoped to this one link's path, good for 12 hours, and the entry's pictures and videos are fetched with it — so not even this server's reverse proxy logs the key. A fresh key from :link, or :unshare, voids the cookies handed out under the old one.

Because only the hash is kept, a link exists in full exactly once: in the moment :share copies it. It cannot be re-read from the entry afterwards. :link mints a fresh key for the same token — which retires the previous link, making it the way to cut off a reader without unpublishing — and :unshare destroys both halves.

API

Everything except the login, the health check and the two share routes requires the session cookie. Media answers Range requests, so a video seeks, sealed or not.

method	path	purpose
GET	/api/health	public: up, and the database answers — for a supervisor
POST	/api/login, /api/logout	session; /api/logout?everywhere=true ends every one
GET	/api/me	current user
GET	/api/sessions	every signed-in browser, the one seen last first
DELETE	/api/sessions/{id}	sign another browser out, and lock the vault
GET/POST	/api/spaces	the spaces / create one
GET/POST	/api/nodes	children (?parent=) or search (?q=, ?space=) / create
GET/PUT/DELETE	/api/nodes/{id}	read / update / delete (a delete takes the subtree)
GET	/api/spaces/{id}/board	the lists and cards of one space
GET	/api/agenda	open cards due within a week or in hand, on every board, and whether the diary has today
POST	/api/spaces/{id}/lists	add a list
PUT/DELETE	/api/lists/{id}	rename / delete a list and its cards
POST	/api/lists/{id}/cards	add a card
PUT/DELETE	/api/cards/{id}	update (title, body, due, done, link) / delete
POST	/api/cards/{id}/move	to another list and place in it
POST	/api/nodes/{id}/move	reparent, and place among the new siblings
POST	/api/nodes/move	{ ids, parent_id }: several into one folder, all or none
POST	/api/nodes/{id}/import	import a folder or zip (multipart) under this node
POST	/api/import	import an export at the root: directories become spaces
GET	/api/resolve?path=	the node behind a slug path, work/notes/monday
POST/DELETE	/api/nodes/{id}/share	mint or rotate / revoke a share link
GET	/api/export	the whole workspace, boards and dates included, as a zip
GET/POST	/api/backup	Proton Drive mirror: status / run now
POST/DELETE	/api/backup/login	log the mirror in to Proton, body {"username","password"} / forget its session
POST	/api/backup/login/totp, /api/backup/login/captcha	the two-factor code {"code"}, or the solved human check {"token"}, a waiting login asked for
GET/POST	/api/media	list / upload (multipart)
GET/DELETE	/api/media/{id}	serve (with Range) / delete a file nothing in the tree owns
GET	/api/spaces/overview	every space with what was last written in it
POST	/api/nodes/{id}/files	upload files into this folder as file nodes (?encrypt=true seals them)
POST	/api/nodes/{id}/inspect	what a certificate, key or keystore file holds ({password} opens a protected one)
POST	/api/nodes/{id}/encrypt, /decrypt	seal a file, or a folder for good — what is in it and what lands later — or take it out
GET/POST/PUT	/api/vault	status / set the password once / change it
POST	/api/vault/unlock, /api/vault/lock	open / close the vault
GET/POST	/api/ssh-keys	the keys that may mount over SFTP / add one
DELETE	/api/ssh-keys/{id}	remove a key, which ends its mounts
POST	/api/share/{token}	public: read a shared document, body {"key"}; sets the share's media cookie
GET	/api/share/{token}/media/{id}	public: media inside a shared document, with that cookie

Deploying

Put a TLS-terminating reverse proxy in front of it, point it at WORKSPACE_BIND, and set WORKSPACE_SECURE_COOKIE=1. Nothing else is needed: no database server, no runtime dependencies, no build tools on the host. The server will not start while WORKSPACE_SECRET or WORKSPACE_PASSWORD is still the example from .env.example: with the example secret anyone could sign themselves a session.

Wrong passwords are answered more and more slowly — free for three, then 1 s, 2 s, 4 s… up to 30 s. A browser that has logged in before keeps a signed diary_device cookie (__Host-diary_device over HTTPS) and is slowed only by its own mistakes, so someone guessing at the login from elsewhere cannot keep you out of a browser you already use. A browser that has never logged in waits in the same line as the guesses. The cookie opens nothing and survives logging out.

With WORKSPACE_SECURE_COOKIE=1 the session cookie is Secure and named __Host-diary_session, so the browser keeps it to this host and HTTPS, and every response carries Strict-Transport-Security for a year. Turning it on logs everyone out once.

The app sends its own Content-Security-Policy, Referrer-Policy: no-referrer, X-Content-Type-Options, X-Frame-Options, Cross-Origin-Opener-Policy, Cross-Origin-Resource-Policy and a Permissions-Policy that turns off camera, microphone and location; API answers are Cache-Control: no-store unless they say otherwise. A proxy that adds its own should not weaken them. no-referrer matters in particular, because a share token lives in the URL and would otherwise leak to any site a shared entry links to. Share links minted before the key existed are revoked by migration 0005; re-run :share on those entries to publish them again. Uploaded files are only ever served as types that cannot execute — anything else is handed back as an opaque download.

A write that a browser says came from another origin — Sec-Fetch-Site, or Origin against Host where a browser is too old to send that — is refused, so a proxy must pass the Host header through unchanged (traefik and nginx's proxy_set_header Host $host both do). Only the upload routes accept a body as large as WORKSPACE_MAX_UPLOAD_MB; everything else stops at 16 MiB, and the login at 16 KiB, before anyone has proved who they are. SIGTERM gets in-flight requests eight seconds to finish, inside the ten docker stop waits. Any other request that has not been answered after 30 seconds is given up on with a 503, except the ones that move files around: uploads, imports, the export, re-sealing and a backup run.

Every request is logged at info, path and status and time, never a query string, and with a share link's token and key left out. A background loop that panics — the mirror, the mail, the sweep, the SFTP server — is logged and started again rather than lost until the next restart. If WORKSPACE_SESSION_KEY_FILE cannot be read the server starts anyway with backups off, and :backup says why.

Docker

cp .env.example .env   # set WORKSPACE_USER, WORKSPACE_PASSWORD, WORKSPACE_SECRET
docker compose up -d --build

The image builds the frontend with bun and the binary with cargo, then ships only the binary on debian:trixie-slim (~86 MB), running as an unprivileged user. WORKSPACE_BIND and WORKSPACE_DATA_DIR are forced in compose.yml, so the values in .env do not have to change; everything else is read from .env. Database and uploads live in the diary-data volume — named for what it held when it was made — and survive recreation. Keep the project name if you rename the directory (COMPOSE_PROJECT_NAME=narl-diary), or compose will look for the volume under a new name and find an empty one.

By default the port is published on 127.0.0.1:4242 only. To serve it through traefik instead, copy compose.override.yml.example to compose.override.yml and adjust the host — it joins the external narl network, drops the published port, routes to container port 4242, and sets WORKSPACE_SECURE_COOKIE=1, since traefik serves it over HTTPS.