Apricity

A local, zero-dependency front-end for MemPalace — browse, search, write, edit, and curate your personal-memory store and knowledge graph from your browser.

What is Apricity?

MemPalace gives your AI assistant a structured, durable memory — wings, rooms, drawers, and a knowledge graph, all stored locally. Apricity is the front-end for it. Once you have hundreds of memories, you need a way to see, search, and curate them without dropping to SQL or living inside an MCP shell. That's what Apricity is for.

It is not a memory system of its own — it reads and writes an existing MemPalace install on the same machine, through the official mempalace package. Think of it as the window onto your palace.

• 🏠 Local-first. Binds to 127.0.0.1. Nothing leaves your machine. No telemetry. No accounts. No cloud.
• 🪶 Zero runtime dependencies. Pure-Python standard-library server + plain HTML / CSS / vanilla JS front-end. No pip install, no npm, no Docker, no build step.
• 🛡️ Safe by construction. Every write goes through the official mempalace Python package. Every destructive action is snapshotted and recoverable. ETag concurrency control prevents lost edits.
• ⚡ Fast. Reads the SQLite/Chroma backends directly; a localStorage cache, gzip, and serve-time minification make reloads feel instant.
• ⌨️ Keyboard-driven. ⌘K to search, E to edit, F to maximize/browse — core actions are remappable.

⚠️ Apricity expects an installed and initialised MemPalace instance on the same machine. It does not ship MemPalace itself — install it first from MemPalace/mempalace.

Contents

• Features
• Quickstart
• Configuration
• Securing Apricity
• Keyboard shortcuts
• API reference
• Safety model
• Architecture
• Troubleshooting
• FAQ
• Contributing
• Security
• License

Features

Browse & navigate

_{The search bar's always-on rotating ring — Apricity's “alive and listening” affordance. ⌘K from anywhere.}

• Search across content and metadata, filter by author, and sort by date or title (with a configurable default). ⌘K focuses it from anywhere; Enter jumps to the first match.
• Three panes — a Rooms sidebar (wings → rooms, iOS-style drill-down), a Memories list, and a Detail view.
• Browse mode — maximize the Memories panel into an adaptive multi-column grid; toggle with F.
• Wiki-style links — [[drawer_id|Display Title]] references in memory content become clickable jumps to the linked memory, with one-click back-navigation.
• URL state — the current wing / room / memory / query / sort survive reloads and are linkable.

Write & edit

_{The inline WYSIWYG editor — circular toolbar, live bold/italic, ## → heading autoformat, and callouts. No modal, no raw-Markdown round-trip.}

• Inline visual editor — a WYSIWYG editor for bold/italic, headings, lists, links, and callouts, with Markdown autoformatting and HTML→Markdown conversion on save. No modal, no raw-Markdown round-trip.
• Edit metadata — change content, title, wing, and room, or move a memory between rooms, with content-hash ETag protection against concurrent edits.
• Optimistic saves — edits apply instantly and revert automatically (with a notification and a retry affordance) if the server rejects them.
• Drafts inbox — stage partial memories, edit them in place, and file them into the palace when ready.

Curate & recover

_{Nothing is ever lost — every delete is snapshotted into Recently deleted, one click from restore.}

• Delete a single memory, a whole room, or a whole wing — every delete is snapshotted to a recoverable log.
• Recently deleted — restore a deleted memory with one click (optimistically), or purge the log.
• Knowledge graph — view triples, add new facts, invalidate stale ones, and explore a force-directed graph visualisation.
• Tunnels — cross-room / cross-wing links surfaced inline as room chips; create and delete them, bind them to specific memories, and inspect them on a dedicated tunnel page.

Stay in the loop

_{The bell rings as each model files or revises a memory — per-model avatars, created vs updated at a glance.}

• Notifications — a bell that surfaces newly created and updated memories with the originating model's avatar (Claude, Codex, …), distinguishing created from updated.
• Recently-updated markers — memories touched recently get a relative-time marker that clears once you've viewed them.
• Live polling — optional background refresh (15 / 30 / 60 s) that pauses when the tab is hidden, with an optional sound and LAN-wide "seen" sync so dismissing on one device clears it everywhere.

Make it yours

_{Auto / Light / Dark — follow the system or pick a side; your choice is persisted.}

• Tools panel — a power-user sheet that surfaces MemPalace's MCP tools: knowledge-graph query + timeline, tunnels (list, create, delete, find, follow, traverse), diary (read / write), stats, and maintenance (taxonomy, duplicate check, hook settings, sync, reconnect, AAAK spec). Hidden by default; enable it in Settings. See the API reference for the underlying endpoints.
• Settings — a redesigned sheet with sidebar navigation: Display (theme, reduce-motion, relative time, title/name polishing, panel-control reveal), Notifications, remappable keyboard Shortcuts, Trash, Backup, Account, and an About pane with an update-available check.
• Theme — Auto / Light / Dark, overriding or following the system preference, with your choice persisted.
• Auth — optional username + password lock with PBKDF2-SHA256 hashing and HTTP-only session cookies (12-hour default, 30-day "Remember me").

Quickstart

Prerequisites

• Python 3.11 or newer (Apricity itself uses only the standard library — no pip install required).
• A working MemPalace install with its own venv. Apricity shells out to it for writes, so it must be importable from $MEMPALACE_PYTHON_BIN.

Run from a clone (no install step)

The whole server is one file of standard-library Python, so the simplest way to run Apricity is straight from a checkout:

git clone https://github.com/epinethrone/apricity
cd apricity
python3 server.py            # equivalent to `python -m mempalace_dashboard`

Then open http://127.0.0.1:8765.

Install as a command

Prefer an isolated, on-PATH install? Use pipx (or uv tool install):

pipx install apricity
apricity

Note

Published on PyPI as apricity — pip install apricity or uv tool install apricity work too.

There's no build step and no third-party runtime dependencies — the package is published purely so installing it is one command. If your MemPalace install lives somewhere non-standard, copy .env.example to .env (or export the variables in your shell) and adjust the paths before launching.

First-run checklist

1. ✅ MemPalace itself is installed and you've successfully filed at least one memory through it.
2. ✅ ~/.mempalace/palace/chroma.sqlite3 and ~/.mempalace/knowledge_graph.sqlite3 exist (or you've pointed MEMPALACE_PALACE_DB / MEMPALACE_KG_DB at where they actually live).
3. ✅ MEMPALACE_PYTHON_BIN points at a Python that can import mempalace.
4. ✅ You opened Apricity, clicked Settings, and set a username + password.

Configuration

All filesystem locations default to the standard MemPalace home (~/.mempalace). Override with environment variables if your installation differs (see .env.example for a starter file):

Variable	Default	Purpose
PORT	8765	Port Apricity listens on.
MEMPALACE_HOME	~/.mempalace	Root for all of Apricity's data.
MEMPALACE_PALACE_DB	<HOME>/palace/chroma.sqlite3	The Chroma SQLite backend.
MEMPALACE_KG_DB	<HOME>/knowledge_graph.sqlite3	The knowledge-graph SQLite database.
MEMPALACE_INBOX	<HOME>/dashboard-inbox	Where drafts are staged before filing.
MEMPALACE_VERSIONS	<HOME>/dashboard-versions.jsonl	Append-only log of deletions (powers Recently deleted).
MEMPALACE_CREDENTIALS	<HOME>/dashboard-credentials.json	Username + PBKDF2 hash. Mode 0600.
MEMPALACE_SESSIONS	<HOME>/dashboard-sessions.json	Active session tokens. Mode 0600.
MEMPALACE_PREFERENCES	<HOME>/dashboard-preferences.json	Server-side UI preferences (sort, theme, notification settings).
MEMPALACE_PYTHON_BIN	~/.local/share/mempalace-venv/bin/python	Python that can import mempalace.
MEMPALACE_TOKEN	(unset)	Optional shared-secret used by scripts via the X-Auth-Token header. Coexists with the cookie flow.

Securing Apricity

1. Start the server, open the UI, click Settings.
2. Choose a username and password (≥ 8 characters).
3. Save — you're logged in immediately and Apricity refuses every other client until they sign in.

Credentials are stored as pbkdf2_sha256$200000$<salt>$<hash> and never leave the host.

If you forget the password, delete ~/.mempalace/dashboard-credentials.json and ~/.mempalace/dashboard-sessions.json and restart the server — Apricity reverts to open setup mode so you can re-enroll.

For scripted access, set MEMPALACE_TOKEN=<some-secret> in the environment and send it as the X-Auth-Token header. This coexists with the cookie flow; it does not replace it.

Keyboard shortcuts

Four actions are remappable under Settings → Shortcuts (defaults shown); ⌘K and Esc are fixed.

Shortcut	Action	Remappable
F	Maximize / Browse mode	✅
E	Edit the selected memory	✅
X	Close / collapse the current panel or sheet	✅
Backspace	Delete the focused item (with confirmation)	✅
⌘K / Ctrl+K	Focus the search box (inserts a link inside the editor)	—
Esc	Close any open sheet — permanent fallback for X	—
Enter (in search)	Jump to the first matching memory	—

Single-key shortcuts only fire when you're not typing in a field. Full details in the wiki.

API reference

All endpoints are JSON over HTTP. Auth is by session cookie (mempalace_session) or X-Auth-Token header when MEMPALACE_TOKEN is set.

Method	Path	Auth	Purpose
GET	/health	open	Liveness + whether auth is active.
GET	/api/session	open	Current auth state.
POST	/api/login	open	{username, password, remember} — issues a session cookie.
POST	/api/logout	open	Clears the session.
GET	/api/palace	auth	Full palace snapshot (wings, drawers, triples, stats).
GET	/api/search?q=…	auth	Filtered subset of drawers + triples.
GET	/api/versions	auth	Recently deleted snapshots (powers Recently deleted).
POST	/api/memories	auth	File a new memory.
POST	/api/memories/update	auth	Patch content / wing / room of an existing drawer; supports ETag.
POST	/api/delete	auth	Delete a drawer, room, or wing ({scope, …, confirm: "DELETE"}).
GET	/api/drafts	auth	List drafts; ?id=<draft-id> returns one with body.
POST	/api/drafts	auth	Save a new draft.
POST	/api/drafts/update	auth	Replace a draft in place.
POST	/api/drafts/delete	auth	Remove a draft.
POST	/api/drafts/commit	auth	File a draft into the palace.
POST	/api/versions/restore	auth	Recreate a deleted memory from a snapshot.
POST	/api/versions/delete	auth	Remove one snapshot from the log.
POST	/api/versions/clear	auth	Wipe the snapshot log ({confirm: "CLEAR"}).
POST	/api/facts	auth	Add a knowledge-graph triple.
POST	/api/facts/invalidate	auth	Mark a triple as ended.
GET	/api/settings	auth	Whether credentials are configured + username.
POST	/api/settings/credentials	auth	Set or rotate the username + password.
GET	/api/kg/query?entity=…&direction=…&as_of=…	auth	Query KG facts about an entity.
GET	/api/kg/timeline?entity=…	auth	Chronological view of an entity's facts.
GET	/api/kg/stats	auth	Knowledge-graph aggregate stats.
GET	/api/graph/stats	auth	Palace graph stats.
GET	/api/taxonomy	auth	Current wing/room taxonomy tree.
GET	/api/checkpoint	auth	Memories-filed-away state (recent checkpoint).
GET	/api/aaak-spec	auth	AAAK spec reference.
GET	/api/diary?agent_name=…&last_n=…&wing=…	auth	Read agent diary entries (latest last_n, capped at 200).
POST	/api/diary	auth	Write a diary entry ({agent_name, entry, topic?, wing?}).
GET	/api/tunnels?wing=…	auth	List tunnels (optionally filtered by wing).
POST	/api/tunnels	auth	Create a tunnel ({source_wing, source_room, target_wing, target_room, label?, source_drawer_id?, target_drawer_id?}).
POST	/api/tunnels/delete	auth	Delete a tunnel by ID. Snapshotted to the versions log before deletion.
GET	/api/tunnels/find?wing_a=…&wing_b=…	auth	Find tunnels between two wings.
GET	/api/tunnels/follow?wing=…&room=…	auth	List tunnels reachable from a room.
GET	/api/traverse?start_room=…&max_hops=…	auth	Graph-traverse from a starting room (max_hops capped at 5).
POST	/api/check-duplicate	auth	Check whether content is a near-duplicate ({content, threshold?}).
GET	/api/hooks	auth	Current hook settings (silent-save, desktop-toast).
POST	/api/hooks	auth	Update hook settings ({silent_save?, desktop_toast?}).
POST	/api/sync	auth	Sync from a project tree ({apply?, wing?, project_dir?}). Slow — runs with a 5-minute timeout.
POST	/api/reconnect	auth	Force-reconnect the MemPalace backend.

Example — file a memory from the command line:

curl -X POST http://127.0.0.1:8765/api/memories \
  -H "Content-Type: application/json" \
  -H "X-Auth-Token: $MEMPALACE_TOKEN" \
  -d '{"wing":"notes","room":"general","title":"hello","content":"first memory from curl"}'

Safety model

• No raw DB writes. Every mutation routes through the official mempalace Python package — Apricity never modifies SQLite directly.
• Snapshot before destruction. Every delete (memory, room, wing, tunnel) appends a record to the versions log so the content is recoverable. Edits aren't snapshotted as history — they're protected from accidental overwrite by ETag concurrency instead.
• ETag concurrency. The edit form sends the content hash it was opened with; the server refuses to overwrite if it has changed.
• Confirmation values. Bulk deletes and snapshot wipes require an exact-string confirmation in the request body.
• No plaintext secrets. Credentials are hashed with PBKDF2-HMAC-SHA256 (200 000 iterations, 16-byte salt). Sessions are stored server-side; cookies are HttpOnly + SameSite=Lax.
• Loopback by default. The server binds to 127.0.0.1. If you put it behind a reverse proxy, terminate TLS there and keep the upstream on loopback.

Architecture

┌────────────────────┐   HTTP/JSON   ┌───────────────────────┐  subprocess  ┌─────────────────┐
│ Browser (vanilla   │ ────────────► │ server.py             │ ───────────► │ mempalace venv  │
│ HTML + JS, no      │ ◄──────────── │ (stdlib only,         │ ◄─────────── │ (writes only)   │
│ build step)        │   snapshots   │  ThreadingHTTPServer) │              └─────────────────┘
└────────────────────┘               │                       │  read-only
                                     │                       │ ─────────────► palace/chroma.sqlite3
                                     │                       │ ─────────────► knowledge_graph.sqlite3
                                     └───────────────────────┘

• Reads go directly against the SQLite files for speed.
• Writes are dispatched to a child Python that imports mempalace — keeps Apricity in lock-step with the canonical schema.
• State that belongs to Apricity (drafts, snapshots, credentials, sessions, preferences) lives in $MEMPALACE_HOME alongside the palace data so a single backup captures everything.

The Python import package is still named mempalace_dashboard (hence python -m mempalace_dashboard and the directory layout below) — only the distribution, command, and brand are apricity.

Troubleshooting

Apricity loads but the palace is empty

The UI rendered, but Wings: 0 and the list is empty.

• Confirm ~/.mempalace/palace/chroma.sqlite3 exists. If it doesn't, MemPalace itself hasn't been initialised on this machine.
• If your palace lives elsewhere, set MEMPALACE_PALACE_DB and MEMPALACE_KG_DB to the correct absolute paths and restart.
• Check the server log — Apricity reports the exact paths it tried to open.

"mempalace package not found" when writing

Reads work, but every save / delete fails with a subprocess error mentioning ModuleNotFoundError: No module named 'mempalace'.

• MEMPALACE_PYTHON_BIN points at a Python that can't import mempalace. Run it manually:

  $MEMPALACE_PYTHON_BIN -c "import mempalace; print(mempalace.__file__)"

• Fix the path, or install mempalace into that venv, then restart Apricity.

I forgot my password

rm ~/.mempalace/dashboard-credentials.json ~/.mempalace/dashboard-sessions.json

Restart the server and re-enroll via Settings. This does not touch your memories.

Port 8765 is already in use

PORT=9000 python3 server.py

I want to reach Apricity from another machine on my LAN

The server binds to 127.0.0.1 deliberately. Put it behind a reverse proxy (Caddy, nginx, Tailscale Serve) that terminates TLS and forwards to the loopback port. Never expose the raw port to the public internet — Apricity is designed for a trusted host.

FAQ

Is this an official MemPalace project? No. Apricity is a community front-end that talks to the official mempalace package. The MemPalace project itself lives at MemPalace/mempalace.

Does it work offline? Yes. The server is loopback-only and the front-end is vendored — no CDNs, no analytics, no outbound calls (except an optional, cached check of the GitHub releases API to tell you when a new Apricity version is out).

Do my memories leave my machine? Never. Apricity reads and writes only files under $MEMPALACE_HOME. There's no network code in the data path.

Can I run multiple instances on one host? Yes — give each one its own PORT and (if you want isolated palaces) its own MEMPALACE_HOME.

Why no pip install requirement? The whole server is one file of standard-library Python. Keeping it dependency-free means you can audit it, vendor it, or run it on a fresh box without thinking about supply chains.

Can I script against it? Yes. Set MEMPALACE_TOKEN=… and pass it as X-Auth-Token. See API reference.

Contributing

Issues, ideas, and PRs are welcome. See CONTRIBUTING.md for the development loop, code conventions, and how to file a useful bug report.

If you're adding a feature, please open an issue first to talk through the shape of it — Apricity values being small and dependency-free, so not every feature is the right fit.

Security

If you find a vulnerability, please do not open a public issue. See SECURITY.md for the disclosure process.

License

MIT © Apricity contributors.