Manifesto v2 · core 0.9 · 9 domains · RU

Application
description format

At the OpenAPI / JSON-LD level. One artifact — four readers: pixels, voice, agent API, document. LLM helps during authoring; at runtime it's not there.

Get started — 5 slides → Manifesto v2 GitHub
world = fold(Φ)  ·  artifact = f(intents, ontology, projection)
↓
Four objects of the format
Ontology, intents, projections, log.
Everything else is derived. The artifact, the UI, the agent schema, the document — are all functions over these four.
◉

Ontology

Entities, fields with read/write matrix, roles, invariants, rules.

→

Intents

Declarative particles: ownerRole, requiredFields, conditions, effects.

▢

Projections

Which intents form one screen, one dialog, one page.

Φ

Phi — the log

Stream of confirmed effects. World isn't stored — it's fold(Φ).

0
domains
0
intents
0
tests
0
materializations
0
UI adapters
Four readers
One artifact. Four equal materializations.
Pixels, voice, agent API, document — architecturally equal. All read the same viewerWorld, scoped by role.
01 ▦

Pixels

Renderer + adapter + Token Bridge. Mantine, shadcn, Apple-glass, AntD.

ProjectionRendererV2
02 ♫

Voice

Speech-script of turns. JSON for voice-agents, SSML for TTS, plain for IVR.

/api/voice/:domain/:projection
03 ↵

Agent API

JSON-schema of intents, viewer-scoped world, exec. Preapproval guard for limits.

/api/agent/:domain/{schema,world,exec}
04 ▤

Document

Structured graph: sections, tables, fields, badges. HTML for print, JSON for archives.

/api/document/:domain/:projection
Safety is a property of the format: all four readers get filterWorldForRole for free.
Where LLM belongs
Authoring — yes. Runtime — no.
No decision that affects world state or rendering depends on the LLM. The format's determinism excludes it.
Design-time
  • Ontology inference from a domain description
  • Artifact enrichment — labels, icons, placeholders
  • Pattern suggestion — candidate patterns from live apps
Runtime
  • Artifact rendering
  • Intent validation and application
  • Voice / document / agent API
  • Witness promotion
Zones of strength
Nine domains. 585 intents. One engine.
Each domain is a new shape of problem. Different aesthetics, roles, materializations — from one ontology.
225
Sales
Auction. 11 entities, 4 roles. Pure-crystallization.
100
Messenger
Real-time chat. WebSocket, WebRTC, threads.
58
Invest
AntD fintech. 4 roles, 7 rules, 3 ML services.
56
LifeQuest
Goals + habits. Shadcn/doodle, mobile-first.
47
Reflect
Mood diary. Apple visionOS-glass.
45
Delivery
Food last-mile. 5 roles, map, scheduler.
22
Booking
Service booking. Temporal predicates.
17
Planning
Collaborative polls. Quorum, phases.
15
Workflow
Process editor. React Flow canvas.
Authoring environment
IDF Studio. A tool, not part of the format.
Ontology visualization, Pattern Inspector, viewer / role switching, Φ-playback. Plays no role at runtime.
IDF Studio
The invest domain in the authoring environment

Ontology graph

Entities, roles, ownership, references.

Live prototype

Projections render inline on every change.

Pattern Inspector

Off / Preview / Commit for stable patterns.

The artifact is a function of ontology, intents and projection. The UI is a function of the artifact. Everything computable, everything reproducible. No runtime LLM.
— Manifesto v2, Part III