Skip to content

About

A malleable personal environment for connected ideas, programmable spaces, places and time, with opt-in AI and expandable memory.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Repository files navigation

Nothing–Everything

An experiment in a malleable digital environment: a personal space that starts simple and can grow with what someone learns, creates and needs. Ideas, code, objects, relationships, places and changes through time belong together, with their sources and history retained. The aim is to let people shape both their work and the ways they explore it, while complexity unfolds without losing its underlying detail.

The working prototype combines programmable 3D spaces, connected ideas, geographic anchors and an opt-in tutor with continuous, expandable memory. Wording, controls and panel layouts are adjustable. It is an early local-first foundation for broader exploration and cooperation; sharing, optional public identity and AR/VR remain future directions. Built iteratively with Muse Code and Codex.

Hosted preview: eelsoo.vercel.app — refreshed on 7 November 2026 with the current manual editor, customization controls and opt-in detailed Earth map. AI chat and durable Tutor memory use the local backend and are unavailable on this static preview. Worlds stay in each browser; use File → Export/Import to move them between devices.

Quickstart

Requires Node with the built-in node:sqlite module. Developed and tested only on Node 26.12.0; older versions are untested.

npm ci
npm run dev        # http://127.0.0.1:4317/
npm run build      # production build

Both dev and preview bind to loopback only and start the local API proxy. npm run mock starts an optional test upstream on port 4319; it returns test data and is never the default provider.

What it does

  • Personal + place workspaces and an Earth view in one environment. Create cubes/spheres, write an interpretation ("idea"), keep an optional verbatim source passage or URL, and connect objects with your own stated reasons. Links are authored assertions, not verified facts.
  • Programmable behavior: per-object TypeScript (transpile-checked, not fully type-checked), typed input knobs, evaluated custom meshes, and time-based replay with play/scrub/reverse. Code runs in a timeout-bounded worker — not a hardened sandbox; run code you trust.
  • Editable timeline and guides: experiment start/end seconds, looping, and a programmable grid/guide editor with built-in square/polar/hex examples.
  • Direct-manipulation UI: movable panels, a draggable Tutor, command palette, editable keyboard shortcuts, light/dark themes, and editable wording (Fluent profiles: English/Estonian bundled, custom profiles manual, nothing auto-translated).
  • Optional continuous Tutor: one retained conversation, inspectable context before every request, user-reviewed recursive summary trees, child or exact-original expansion, and optional newest-first fitting within the remaining request budget. See "Privacy" below.

What it does not do (yet)

  • No accounts, publishing, sharing, or sync. Everything is single-device.
  • No physics engine; guide programs draw lines and never move objects.
  • Earth uses cached 2K imagery plus an optional detailed-map layer; it is not live planetary data.
  • VR/AR, multi-device continuity, and a solar-system view are recorded design goals (see docs/design/multi-device-world.md), not implemented features.
  • Mock tests cannot verify live provider replies or real cache hits.

Privacy and storage

  • Worlds live in browser storage; export them from the app, not via Git.
  • The Tutor's durable conversation log is a local SQLite database (tutor.sqlite3) under ~/.local/share/territory/tutor by default, or $TERRITORY_TUTOR_DATA_DIR when set. The app's tutor-export.jsonl export backs up the original messages only — summaries, policy, and memory structure are not in it. For a full memory backup, stop the server and copy the whole data directory. The database is never committed — .gitignore covers it and its WAL/SHM sidecars.
  • AI is opt-in. Without a key, all manual authoring works, but no AI generation at all — that includes tutor replies and AI summary drafts, which are explicit one-call actions when a key is configured. Keys are held only by the local server process (session entry or MODEL_API_KEY env), never stored in the browser, exports, or Git. Never use a VITE_ variable for a key — it can leak into the browser bundle.
  • No app-defined system or developer prompt is sent. Requests are built from your conversation and disclosed context, and can be inspected before sending ("View next request").
  • The Tutor's continuous retained-context design was inspired by UniiChat's memory ideas. UniiChat itself is not integrated; this app implements its own originals-plus-reviewed-summaries model.

Repository layout

src/c/             Active application (variant C)
archive/variants/  Historical A/B prototypes, preserved as-is
public/assets/     Cached Earth textures (see SOURCES.md, THREE-LICENSE.txt)
api-plugin.js      Local credentials + provider proxy
mock-upstream.cjs  Isolated provider test server
docs/              Examples and design documentation (see docs/README.md)
tests/             Self-contained browser/server suites (isolated profiles)

The active app is variant C. A and B were earlier, rejected directions kept for reference; they are not part of the production build.

License

Nothing–Everything source code is licensed under the GNU Affero General Public License v3.0 only (AGPL-3.0-only). See LICENSE for the full text. Third-party notices are retained in place (for example public/assets/THREE-LICENSE.txt and public/assets/SOURCES.md for the cached Earth textures).

Learn more

  • Keyless first experiments: docs/examples/personal-space-experiments.md

About

A malleable personal environment for connected ideas, programmable spaces, places and time, with opt-in AI and expandable memory.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages