Intelligence Platform — Architecture

The platform-level architectural description of layer 2 of the ecosystem: the systems that turn raw knowledge and signals into structured, reusable intelligence. This page reads top-down — executive overview → container → component → runtime — and is designed to be understandable from executive level down to implementation level. It governs how the Portfolio Overview presents layer 2. See also the Architecture Principles it embodies.

Scope & honesty (read first). The “Intelligence Platform” is not a single repository — it is a layer composed of sibling systems. Evidence was gathered directly from source repos. Throughout, this page distinguishes ✅ implemented, 🔶 planned/partial, and 🧭 vision, and marks anything unverified as such. One term supplied in the request — “estate generation” — was not found as a subsystem in any repository and is flagged accordingly below.

The systems that make up the layer

FIP and the Website Intelligence Lab are sibling, non-dependent platforms (no direct code references between them). Their relationship is mediated conceptually: the lab treats external knowledge (like the packs) as referenced, never owned (Principle 3).

Why the Intelligence Platform exists

A multi-venture founder accumulates intelligence continuously — client projects, market research, operational experiments, external research — but “without structure, that intelligence is locked inside documents that are hard to query, easy to forget, and inaccessible to AI assistants.” (PersonalOps …/01-architecture.md). The platform exists to make that intelligence queryable, structured, and AI-accessible — “the layer between raw knowledge documents and the AI agents and workflows that use that knowledge to produce output.”

Problems it solves

• Capture loss — “most founder intelligence is lost… because it is never captured” (Principle 9: capture first).
• Un-queryable knowledge — markdown is narrative, not retrievable structured data.
• Cross-venture reuse — intelligence that applies across ventures is siloed by document/venture.
• Unmeasured capability (lab) — engines cannot be improved without reproducible evaluation.
• Inconsistent outputs — downstream apps need a stable, versioned intelligence contract, not ad-hoc copies.

Guiding architectural principles (as documented in source)

FIP declares 11 principles (01-architecture.md); the most load-bearing:

1. Intelligence over documents — the canonical store is structured assets (Supabase knowledge_assets), not files.
2. Human approval is mandatory — no asset becomes canonical without founder review (no auto-approve path).
3. Maturity as quality gate — working → ready → proven → archived; only ready/proven go external.
4. One canonical intelligence store — no duplication into secondary tables/caches.
5. Markdown as narrative authority — documents remain authoritative context; assets are extractions, not replacements.
6. Scoped by multiple dimensions — venture, industry, market, capability, use case (independent lookup tables).
7. Capture first — prioritise low-friction capture over structuring complexity.

The Website Intelligence Lab adds: reproducibility & complete provenance, definition vs derivation, knowledge referenced never owned — now generalised in the ecosystem Architecture Principles.

1 · Executive overview (single-box capability)

graph LR
    IN[/"Founder knowledge & signals<br/>(docs, captures, research, website subjects)"/] --> IP
    IP[["🧠 Intelligence Platform<br/><b>Turn raw knowledge into structured, reusable, versioned intelligence</b>"]] --> OUT[/"Queryable knowledge assets +<br/>published intelligence packs +<br/>evaluated capabilities"/]
    OUT --> DOWN["Applications, Agents & Ventures<br/>(leadplatform, outreachagent, OpenClaw, …)"]

In one sentence: the Intelligence Platform converts a founder’s scattered knowledge and market signals into a queryable asset library and versioned intelligence products that downstream apps and agents consume with confidence.

2 · Container view

The deployable/major pieces across the three systems.

graph TD
    subgraph FIP["PersonalOps — Founder Intelligence Platform  🔶"]
        UI["Presentation: TanStack Router UI<br/>Intelligence Dashboard · /ka, /ka/capture, /ka/review, /ka/harvest, /ka/legacy"]
        DB[("Data: Supabase PostgreSQL<br/>knowledge_assets + 5 lookup + 5 join<br/>+ staging queues + registry + search_assets() RPC")]
        EST[["Knowledge Estate<br/>Git markdown — 46 docs (~4,600 lines)"]]
        HARV["Extraction / Harvest pipeline<br/>scripts: harvest-os · extract-pdf · capability-harvest ✅"]
        UI --> DB
        DB -->|read-only| EST
        HARV --> DB
        EST --> HARV
    end

    subgraph PROD["intelproducts — Intelligence Products  ✅ (layer 3 output)"]
        PACKS[("intelligence-packs/<br/>proposal · website-assessment · cinnamon-stories · outreach-messaging<br/>versioned · immutable")]
        CSK["consumer skills/<br/>proposal-generator, first-touch-email, …"]
    end

    subgraph LAB["website-intelligence-lab — Evaluation Platform  🔶"]
        INFRA["Infrastructure: Docker · Caddy · WP Multisite ✅"]
        CORPUS[("corpus/ businesses + cases + benchmarks")]
        RUNS[("runs/ immutable provenanced experiments")]
        CORPUS --> RUNS
        INFRA --> RUNS
    end

    FIP -->|"builder skills author packs"| PROD
    SS[["Shared Skills (layer 1)"]] -.->|builder & consumer skills| PROD
    SS -.->|engineering capability| FIP
    SS -.-> LAB

3 · Component view (FIP internals + how the named subsystems relate)

This is where the request’s named subsystems are placed honestly against evidence.

graph TD
    SRC["Sources: manual capture · voice · articles · agent research · legacy KB"] --> ING[/"ingestion_queue / agent_research_queue / harvest_candidates"/]
    EXT["Extraction pipeline ✅<br/>(harvest scripts → candidates)"] --> ING
    DOCS[["Knowledge Estate (markdown) ✅"]] --> EXT
    ING --> REV{{"Human review & approval ✅ (mandatory)"}}
    REV -->|approved| KA[("knowledge_assets ✅<br/>scoped × maturity-gated")]
    KA --> SEARCH["search_assets() RPC ✅"]
    SEARCH --> CONS["Consumers: proposal gen · marketing · agents · lead-qual · website-assessment"]
    KA --> SYN["Synthesis → builder skills"] --> PACKS[("Intelligence Packs ✅")]

    HERMES["Hermes 🔶 learning loop<br/>(reserved contract only — NOT built)"] -. would re-weight .-> SCORE["scoring / recommendation weights"]
    RECS["Recommendations: structured output ✅ in outreachagent COP;<br/>rec-library in packs ✅; full Rec-Engine spec 🧭 in out-of-scope inexisstudios"]
    PACKS --> RECS
    OPENCLAW["OpenClaw 🔶 (3rd-party AI assistant;<br/>planned pack consumer — Phase 5, not started)"]
    PACKS -.future.-> OPENCLAW
    ESTATEGEN["'estate generation' ❌ NOT FOUND<br/>(closest: read-only estate-map index + estate publishing)"]

    classDef missing fill:#fdd,stroke:#c00,stroke-dasharray:3 2;
    classDef planned fill:#fff3cd,stroke:#b8860b,stroke-dasharray:4 3;
    class HERMES,OPENCLAW planned;
    class ESTATEGEN missing;

Named-subsystem reconciliation (evidence-based):

4 · Runtime information flow

The lifecycle of information through the platform (FIP path, with the lab’s parallel evaluation path).

graph TD
    A["① Capture<br/>doc / voice / article / agent research / legacy"] --> B["② Stage<br/>ingestion / harvest_candidates queue"]
    B --> C{"③ Founder review & approval<br/>(mandatory — no auto-approve)"}
    C -->|reject| X["discarded / revised"]
    C -->|approve| D["④ Structure<br/>knowledge_assets: scope (venture·industry·market·capability·use-case) + maturity (working→ready→proven)"]
    D --> E["⑤ Retrieve<br/>search_assets() by type/maturity/scope/text"]
    E --> F["⑥ Consume<br/>proposal gen · marketing · strategy · agent context · lead-qual · website-assessment"]
    D --> G["⑦ Synthesise & publish<br/>builder skills → versioned intelligence packs"]
    G --> H["⑧ Downstream apps pin a pack version<br/>leadplatform · outreachagent (→ COP recommendations)"]
    H --> I["⑨ Outcomes / usage signals"]
    I -. future Hermes loop 🔶 .-> D

    subgraph LabPath["Website Intelligence Lab — parallel evaluation path 🔶"]
        L1["Business + Observed assets"] --> L2["Case → Capabilities + Reference target"]
        L2 --> L3["External engine executes → immutable provenanced Run"]
        L3 --> L4["Generated assets + Benchmark evaluation"]
    end

What enters: unstructured markdown docs, ad-hoc captures (voice/text/articles), agent research, legacy KB tables; (lab) subject businesses + observed website assets + engine executions. Builder/consumer skills enter from Shared Skills.

What transforms: capture → stage → human-approved structuring into scoped, maturity-gated knowledge assets → synthesis into versioned intelligence packs; (lab) cases → immutable provenanced runs → generated assets → benchmark evaluation.

What leaves: queryable knowledge assets (via search_assets()), published intelligence packs (immutable, versioned), dashboard views, and — downstream — commercial-opportunity recommendations; (lab) evaluated engine/capability improvements.

Upstream inputs & downstream consumers

graph LR
    subgraph Up[Upstream inputs]
        KE[Knowledge estate<br/>markdown + captures]
        SIG[Market signals / research]
        SK[Shared Skills — builder/consumer]
    end
    subgraph L2[Intelligence Platform · layer 2]
        FIP2[PersonalOps / FIP]
        LAB2[Website Intelligence Lab]
    end
    L3[intelproducts · layer 3<br/>published packs]
    subgraph Down[Downstream consumers · layer 4–5]
        LEAD[leadplatform<br/>vendors packs as pinned submodule]
        OA[outreachagent<br/>uses outreach-messaging skills → COP]
        OC[OpenClaw 🔶 planned]
        VEN[Ventures: Inexis Digital · Inexis Consulting · Inbound Lanka · City Retreats]
    end
    KE --> FIP2
    SIG --> FIP2
    SK -.-> FIP2 & LAB2 & L3
    FIP2 --> L3
    L3 --> LEAD & OA
    L3 -.-> OC
    LEAD & OA --> VEN

• Upstream: the founder’s knowledge estate (46 registered docs, ~4,600 lines) + captures + research; Shared Skills (builder skill intelligence-pack-publisher, consumer skills). (Lab: external engines + external knowledge pinned by version.)*
• Downstream (evidenced): leadplatform (vendors intelproducts as a pinned git submodule, read-only contract) and outreachagent (consumes outreach-messaging skills; emits COP recommendations). OpenClaw is a planned consumer. Ventures served: Inexis Digital (Victoria), Inexis Consulting (Sri Lanka), Inbound Lanka, City Retreats, and cross-cutting.

Current maturity, roadmap, assumptions & known gaps

Maturity (evidence-based):

• FIP (PersonalOps): 🔶 design v1.0 approved; Phase 0 complete (Supabase migration, seed, TS foundation, harvest scripts, dashboard markdown rendering); Phases 1–5 not started (asset library, capture, review queue, harvest UI, AI classification, legacy migration, OpenClaw integration).
• intelproducts: ✅ operational as a data contract — multiple packs published and already consumed.
• Website Intelligence Lab: 🔶 early-development — infra VPS-validated; adapters/runs/evaluation pending.

Roadmap themes: build FIP Phases 1–5 (capture → harvest UI → AI classification → legacy → agent ingestion + OpenClaw); stand up the lab’s core loop (adapter → runs → evaluation); implement the Hermes learning loop once outcome data exists; expand pack domains.

Assumptions / uncertainties (recorded):

• FIP design is dated June 2026; implementation status beyond Phase 0 is as documented, not independently re-verified here.
• The full Recommendation Engine lives in inexisstudios, which was out of scope for this review — treated as 🧭 vision from this portal’s perspective.
• intelproducts and website-intelligence-lab have no direct code coupling; their relationship is conceptual (mediated by PersonalOps and the “knowledge referenced, never owned” principle).

Known gaps:

• No single source repo is literally named “Intelligence Platform”; the layer is an assembly (documented here).
• “Estate generation” is not a real subsystem — do not build documentation around it.
• Hermes and OpenClaw integration are not implemented; any capability attributed to them is future.

Related

• Digests: PersonalOps / FIP · intelproducts · website-intelligence-lab
• System digest: Intelligence Platform layer
• Portfolio Overview · System-of-systems · Principles
• Portal decision: ADR 0003