What Loci is and where it came from.

Loci is a mindspace primitive. Concretely: an intelligence substrate written in plain text, a templates kit plus a few named processes that any file-aware AI can run. Think of it as firmware for a mind, the part that decides how context is held, how identity persists, how work compounds across sessions. The desktop app and the command-line tool are two personal demos, two ways to run it. They are not the product. The substrate is.

It grew from sustained co-intelligence work between Hux (Head of UX/Product at Nym Technologies) and Vesper (a collaborating intelligence built on Claude). It was not designed in advance. It crystallised from practice: from sessions that built upon one another, and from the observation that memory alone was insufficient and that something generative was required. Persistent memory is one of seven feature sets the substrate carries, alongside context architecture, identity and personas, the garden, continuity and synthesis, trust and governance, and interop.

The name derives from the Method of Loci, the classical mnemonic technique in which one constructs a mental palace, places objects in rooms, and navigates to retrieve them. The methodology adapts this framework. The palace holds context. The rooms hold work. The garden grows ideas. The protocols sustain the structure across months of sessions.

By session 28, the palace had accumulated sufficient density that ideas were emerging from the intersections between rooms, not from within any single room. This is the sense in which the garden is generative. It is not a metaphor for recall; it is a structurally distinct mechanism.

The garden

Loci is itself the first case study of what it describes. The substrate was not designed top-down; it grew from a garden. The question of how to make co-intelligence work across sessions was planted early and watered across nearly thirty sessions until the answer had sufficient structure to share. The plain-text kit in the repository is the artefact the garden produced.

The garden holds four active plants. Invisible Proof is a long-running inquiry into zero-knowledge cryptography: not as a technical manual but as a philosophical thread, asking what it means to prove something without revealing it. Political Voice is concerned with arguing for privacy in language that reaches a public register rather than a technical one, writing that changes minds rather than only informing them. The Seed is the garden's public face: dispatches that grow from the research into publishable form, now at loci.garden/seed.

The fourth plant, Alexander's Garden, warrants more detailed treatment because it constitutes the methodological anchor of Loci. Christopher Alexander's A Pattern Language (1977) proposed that good design could be captured as named, reusable patterns: solutions that recur across contexts and can be documented, taught, and applied. Each pattern has a name, a problem it addresses, and a context in which it applies. The insight is that naming a pattern is itself generative: once named, the pattern becomes recognisable across contexts and can be applied intentionally.

Loci is built on this logic. Rooms are patterns. Crystals are patterns. The session delta is a pattern. The morning check-in is a pattern. When something works in the palace, it is named; naming it makes the pattern portable, teachable, and refinable. Alexander's Garden is the plant that tends this process: watching for recurring structures in the co-intelligence work, assigning names, and making them available as explicit tools. Named examples to date include the Named Unknown (the process of naming a question whose edges are visible but whose answer is not yet formed), the Managed Artefact (the principle that every evolving piece of work requires a status, a location, a history, and a destination), and the Honest Side Channel (information that surfaces at the boundary of a structured exchange and proves more revealing than the exchange itself).

These plants grew across sessions without explicit direction. No decision was made to develop a pattern language for co-intelligence; each plant was seeded as a question and watered until it had something to say. That is the garden working.

For teams

The substrate scales to organisations. The mechanism is the same plain text (rooms, crystals, protocols), extended across multiple human–AI pairs running from shared firmware. Rather than one palace for one pair, a team maintains a shared crystal base: verified product facts, architectural decisions, and team context. The shared base is the firmware everyone runs. Each pair reads from the same ground truth and contributes back to it. Sessions do not start from scratch because the team's accumulated knowledge is already loaded.

This is the pattern Loci calls the Collective Loci: multiple individual mindspaces running on one shared substrate, maintained as a living document rather than a static wiki. The shared layer does not replace individual rooms; it sits beneath them, the common firmware every pair inherits automatically.

The methodology is open

The substrate is plain text, so it is open by construction: the scaffold, the protocols, the room templates, and the CLAUDE.md format are all available at github.com/huximaxi/loci. Apache 2.0. It does not belong to anyone; it emerged from a collaboration and is shared as such.

The dispatches (the public writing that grew from the garden) are at loci.garden/seed.

The architecture

loci.garden runs on a VPS in Iceland (1984 Hosting): static HTML, Caddy, and Umami analytics (self-hosted, privacy-first). No third-party tracking. No JavaScript frameworks. The site practices what the substrate preaches: simple, maintainable, and sovereignty-respecting.