PersonalOps — Founder Intelligence Platform (FIP) — Repo Digest

The producer at the heart of the Intelligence Platform layer, and the home of the existing Intelligence Dashboard. Platform-level view (4 architecture diagrams, subsystem map) lives in architecture/intelligence-platform.md.

What it is

A structured knowledge-management platform — the Founder Intelligence Platform (FIP) — that “extracts, stores, and retrieves actionable intelligence from a multi-venture founder’s knowledge estate. It transforms unstructured markdown documents and ad-hoc intelligence captures into a queryable, AI-accessible asset library.” (docs/founder-intelligence-platform/01-architecture.md). PersonalOps also hosts the Intelligence Dashboard (a TanStack Router app rendering markdown + Supabase asset sections).

Why it exists

“The layer between raw knowledge documents and the AI agents and workflows that use that knowledge to produce output.” A founder accumulates intelligence continuously but loses it inside un-queryable documents. FIP makes every material insight a retrievable, scoped asset an AI agent can pull on demand — and publishes distilled intelligence as versioned products for downstream apps. Its hardest problem is capture, not retrieval (Principle 9: capture first).

At a glance

Business capability provided

Founder intelligence as a queryable, AI-accessible asset library — turning scattered knowledge into scoped, maturity-gated intelligence that both the founder and AI agents can retrieve, and into published intelligence products downstream ventures consume.

Technical responsibilities

• Own the canonical intelligence store (knowledge_assets) — one place, no duplication (Principle 4).
• Run capture → stage → human-approve → structure workflows across all sources.
• Run the extraction/harvest pipeline (docs → candidates).
• Provide scoped retrieval via the search_assets() RPC.
• Render the Intelligence Dashboard (markdown + Supabase asset sections; Principle 11).
• Author intelligence packs (via builder skills) for publication to intelproducts.
• Explicitly not: auto-approve, write markdown directly, or act as a public/multi-tenant product.

Core concepts

Key workflows

1. Capture — manual form / voice / article / agent research / legacy KB → staging queues.
2. Harvest/extract — scripts turn documents into harvest_candidates.
3. Review & approve — founder approves each candidate/capture (mandatory).
4. Structure — approved intelligence becomes a scoped, maturity-tagged knowledge_asset.
5. Retrieve — search_assets() filters by type/maturity/scope/free-text for 8 retrieval use cases (proposal gen, marketing, strategy, solution design, opportunity eval, agent context injection, lead qualification, website assessment).
6. Synthesise & publish — builder skills author versioned packs to intelproducts.

Technologies used

• Data: Supabase (PostgreSQL) — knowledge_assets + 5 lookup + 5 join + staging + registry tables + search_assets() RPC; RLS.
• App/UI: TanStack Router + TanStack Query, TypeScript (src/lib/ka.ts, src/routes/ka.*.tsx).
• Pipelines: Node/JS harvest & extraction scripts (harvest-os.js, extract-pdf.js, capability-harvest/, estate-map.mjs).
• Content: Git markdown (content-intelligence/, docs/).
• Retrieval is FTS for MVP; vector/pgvector is post-MVP (non-goal now).

Major modules / components

Capabilities

• knowledge-capture · document-extraction · intelligence-structuring · scoped-retrieval
• intelligence-dashboard · pack-authoring (producer of intelligence products)

Upstream dependencies

Downstream consumers

• intelproducts — receives the published packs this platform authors.
• leadplatform, outreachagent — consume packs (downstream of intelproducts). (Layer 4.)
• OpenClaw — planned programmatic consumer (FIP Phase 5, not started).
• The founder’s ventures (Inexis Digital, Consulting, Inbound Lanka, City Retreats) via retrieval.

Major interfaces & integration points

• Consumes: knowledge estate (read-only), Supabase, Shared Skills.
• Exposes: search_assets() RPC (scoped retrieval); the Intelligence Dashboard; published intelligence packs (to intelproducts); doc_update_suggestions (suggested, human-applied — never auto-written).

Reusable assets exposed to other repositories

• The published intelligence packs (via intelproducts) — the primary reusable output.
• The knowledge model + Supabase schema (a reusable pattern for founder/knowledge intelligence).
• The Intelligence Dashboard rendering model (markdown + asset + hybrid sections).

Architectural decisions

FIP records ADRs in docs/founder-intelligence-platform/06-decisions.md and 11 architectural principles in 01-architecture.md. The portal-level modelling decision is ADR 0003. Several FIP principles seed the ecosystem Architecture Principles (intelligence-over-documents, human-approval-mandatory, one-canonical-store).

Architecture snapshot

graph TD
    EST[[Knowledge Estate<br/>markdown]] --> HARV[Extraction/harvest ✅]
    SRC[Captures · voice · research · legacy] --> Q[/staging queues/]
    HARV --> Q
    Q --> REV{Human approval ✅}
    REV -->|approve| KA[(knowledge_assets<br/>scoped × maturity)]
    KA --> RPC[search_assets RPC]
    RPC --> DASH[Intelligence Dashboard + agents]
    KA --> PACKS[(Published packs → intelproducts)]

Current maturity

Early-development. Rationale (evidence): FIP design is v1.0, approved, June 2026; Phase 0 is complete (Supabase migration + seed + TS foundation, harvest/extraction scripts, estate-map, dashboard markdown rendering). Phases 1–5 are not started (asset library, capture, review queue, harvest UI, AI classification, legacy migration, agent ingestion + OpenClaw). So the design and foundation exist and some pipeline code runs, but the core product surfaces are unbuilt. (Source: .../README.md status table.)

Roadmap

Phase 1 asset library + capture + review queue → Phase 2 document harvest + coverage dashboard → Phase 3 AI classification + doc-update suggestions → Phase 4 legacy migration → Phase 5 agent ingestion + OpenClaw integration. Post-MVP: asset relationships graph, pgvector semantic search, automated doc updates, third-party source ingestion. (Source: 01-architecture.md Future Evolution + README phases.)

Known limitations

• Core product surfaces (capture, review, harvest UI, classification) not yet built (Phases 1–5).
• Retrieval is FTS-only for MVP; no semantic search, no asset relationship graph yet (explicit non-goals).
• Single-founder, private tool — no multi-tenant, no public API.
• Hermes learning loop and OpenClaw integration are future, not implemented.

Future opportunities

• Complete the capture→retrieve loop to realise the “capture first” thesis.
• Add the Hermes feedback loop to close intelligence quality on real outcomes.
• Semantic search + relationship graph once the asset library reaches scale (≥200 assets).

Relationship to the wider AI venture ecosystem

The producer of the Intelligence Platform (layer 2): it consumes Shared Skills (layer 1), publishes intelligence products (intelproducts, layer 3), and feeds Applications & Agents (leadplatform, outreachagent — layer 4) and the founder’s Ventures (layer 5). See Intelligence Platform architecture and the Portfolio Overview.

Links

• Source: local git repo PersonalOps/PersonalOps (private)
• System digest: intelligence-platform
• Platform doc: Intelligence Platform
• Registry row: repo registry