Architect layer
Design
The name, addresses, meta and appearance of this project. Saved settings are read at runtime — a change shows on the next page load, with no rebuild.
Blocks
Every block this project can build a page from, drawn by the real renderer. Grouped by what the block is for.
A block is the unit a page is made of. A page here is not a laid-out file — it is a LIST OF BLOCKS. The renderers are already compiled; the page says which kinds stand on it, in what order and with what text. That is why every page in the project looks like one product rather than like the work of five different people.
A row of capability labels. The tone is a MEANING group, not a colour.
A section of equal cards with NO order between them — the same strip as `flow`, minus the numbers, the link and the animation. The difference is meaning, not decoration: lighting cards up in turn would show a sequence that does not exist, and an untruth told by a good animation is the more convincing kind. Hence `<ul>`, not `<ol>`. The header (badge, heading, lead) is the SHARED `SectionHead`, identical to `flow` and `noBill`. Equal height comes from the grid, never from measuring.
Three things worth knowing
A lead paragraph under the heading — it says what the cards have in common. The badge above carries the RUBRIC, and its colour is decided by the section, not by the content: a rubric has no semantic group to take a colour from.
A card holds one self-contained statement. Read in any order they still make sense — that is the test for using this kind instead of `flow`.
The cards are the same height because the grid row is as tall as its tallest item and each card fills it. No script measures anything.
A card is a CONTAINER: it holds any blocks — a heading, a list, a paragraph — so one kind serves both three short statements and two long side-by-side panels.
A row of capability cards, each ending in a link. Differs from `cards` by drawing rather than by fields: the order inside is fixed — heading, rule, text, link — and the link is pushed to the bottom, so in a row of three the links sit on one line however uneven the text.
What the server brings
Three capabilities that arrive together, not one after another.
Orchestration
Several agent platforms share one context, so a task started in one continues in another.
Persistent memory
What the project already knows stays in context between sessions instead of being re-read every time.
One machine
Everything runs on the server you own. No cloud accounts to open, no per-token bill to watch.
A grid of capabilities: icon, heading, pill label, description. Ported from the storefront — two columns on a phone, three on a monitor, with row gaps deliberately wider than column gaps because nothing but distance separates one cell from the next. Differs from `badges`, which is the labels themselves: here a label is a mark INSIDE a capability that also has a heading, an icon and an explanation.
What comes with the server
Twelve things that are already there — not a roadmap.
- for everyone
Agents with memory
Ready to answer from what the project already knows, without an hour of setup first.
- for everyone
Voice input
Dictate into any field meant for speech; the rest keep the ordinary keyboard.
- for everyone
Sign-in out of the box
Email links and Google, with roles and sessions. Nothing to configure to start.
- for everyone
Database and storage
Rows and files on the machine you own, behind one door and one key.
- for everyone
Backups by parts
Six parts chosen one at a time: people restore one thing, not everything.
- for everyone
Your repository
The project travels to your GitHub and back with one action.
- for everyone
Static by default
Pages are built ahead of time, so a traffic spike costs nothing extra.
- for everyone
Products on one server
A landing page today, a scheduled watcher next week — neither can damage the other.
- for everyone
Many languages
Enabling one later is a setting, not a rebuild of the way the site works.
- advanced
Maps on your machine
Routes and distances computed on the server, so requests are not metered one by one.
- advanced
Communication channels
A two-way link with the project, not just the ability to send a message.
- advanced
Search by meaning
Next to the rows it describes: one backup, one access contour, one meaning of deletion.
A grid of platforms where the glow comes out of the SEAMS: the cells sit two pixels apart over a radial backdrop, so the lattice is brighter in the middle and fades at the edges — no borders, no shadows. Ported from the storefront. The cell count is a multiple of six: two columns on a phone, three on a monitor, and only a multiple of six fills both without a stub. Differs from `featureGrid`, which lists what the PRODUCT can do; this lists platforms that belong to OTHER COMPANIES — hence the company line and the small print about trademarks underneath.
Five agent platforms, one environment
No API keys, no local setup — all five run on your server with full terminal access and shared memory.
Claude Code
Writes, runs and fixes code in your terminal. The gold standard for AI-assisted development.
Anthropic
Codex
Browser-native coding agent. Full project context, no terminal required.
OpenAI
Gemini CLI
Long-context coding agent. Understands the whole project structure in one prompt.
Qwen Code
Open-source coding agent. No subscription lock-in — powerful and free.
Alibaba
Kimi Code
Context-first model for large codebases. Excellent for refactoring and architecture work.
Moonshot
LightRAG
Your company brain: persistent vector memory shared across all five platforms.
Fractera
* The project runs on your own subscriptions to these platforms — no additional fees or commissions are charged for their use. Connect one, several or all of them, at your own discretion.