§ 05 · The Engine
Eight registries. The AI-First Framework. A compounding skills library. The commitment that 10–15% of every revenue dollar funds the Institute for Human Advancement. Bundled as one specification any AI venture can adopt.
The Engine is what Perpetual Core installs — and what we're publishing as a standard. The reference implementation runs this company. The spec is open to anyone building the next one.
§ 01
Every organization, fund, partner, vendor, and counterparty your operation touches. Resolved, deduplicated, queryable. The first thing your team stops re-typing into spreadsheets.
Staff, beneficiaries, volunteers, board, students, members. Roles, relationships, consent state, audit trail. Every person in your org has one record, not seventeen.
Programs, initiatives, grants, contracts, engagements — the work containers that hold budget, scope, accountability. Tied to entities funding them and people doing them.
Tasks, deliverables, follow-ups. The things your team thought were in Asana, Notion, three Slack channels, and a voice note from last Tuesday. Now in one registry, with provenance.
Documents, voice notes, calls, channels, embeddings. Synthesized into one queryable mind. The registry Vellum operates on directly.
The AI workers your team has built and authorized. Each with a defined scope, audit log, refusal rules. Not "AI" as a vibe — agents as a registered, accountable category of operator.
Automated sequences across registries. Built in the AI-First Framework's Automate phase. Versioned, modifiable, owned by your team after handover.
Every state change, every approval, every refusal. The audit log that lets you answer "who did what when" six months later, in front of a regulator if it comes to that.
§ 02
Four phases. None skipped. The same arc on every engagement, scaled to the band.
Learn. Two weeks of operator-grade reading. Calls, docs, voice notes, channels. We don't ask for an intake form. We sit in the meetings.
Wire. Three to four weeks. The eight registries get installed in your stack — Supabase, storage, auth. Operators are querying live data by week 5.
Automate. Six to ten weeks. Skills built against your real workflows. Anthropic SKILL.md format, per-portco JSON. Versioned, auditable, owned by you.
Scale. Two to four weeks. Your team operates and extends the system. We document, train, hand over.
§ 03
Every skill we build is a versioned, auditable unit of automated work. Markdown frontmatter, prose body, optional code blocks — with a per-portco JSON config that scopes it to your operating context.
Your operators can read every skill. Audit it. Modify it. Add to it. No black boxes in your skills library — the moment a black box appears, your team stops trusting the system, and the system stops being yours.
skills/ ├── intake-triage.skill.md ├── grant-status-rollup.skill.md ├── case-handoff.skill.md ├── donor-thanks-draft.skill.md ├── compliance-flag-check.skill.md └── config.portco.json
By the end of an Operations engagement: 15–30 production skills. By the end of an Institutional engagement: a library that outlives the engagement.
This is the compounding part. Every engagement we run adds skills back to the library. Your operators inherit work we did at other organizations under the same constraint regime — pre-vetted, pre-tested, ready to adapt.
§ 04
The Institute for Human Advancement is our 501(c)(3) parent. It runs workforce development for low-income New Yorkers, healthcare-pathway training in community-college workforce programs, AI-native founder training across emerging markets, and field health work in East Africa. The company exists to fund the mission.
Here's what flows where.
The mixed rate is intentional. Sage carries 15% because it's positioned as an owned-perpetual-asset for individual operators. Engagements stay at 10% to keep the math defensible at scale. The floor is non-negotiable.
A high structural bar — not a moat.
VC-backed AI companies struggle to clear it; their cap tables resist double-digit revenue commitments to a 501(c)(3). JVs face the same constraint from their LPs. The Anthropic-Blackstone venture, the OpenAI-TPG venture, EPAM's Black Belts, the Big-4 AI practices — the structure is hard for them. It isn't hard for AI-native ventures built this way from day one.
This is the part we want to make easier. The spec is published. The methodology is open. The math is named. If you're structuring a new AI venture and want to bake this in from the cap-table stage, the Engine is yours to adopt.
Our books are audited annually. Every invoice line-items the contribution. Imitation is the goal — we're building the reference, not the only one.
§ 05
If you're founding an AI venture and want to structure it the same way — registry-first substrate, AI-First methodology, operator-owned skills, and a structural giving floor — start with the spec on this page. We'll point you at the reference implementation. Email the founder when you're ready to compare notes.
§ —
A 90–180 day engagement installs the Engine inside your organization. A system your team owns. A funding flow that compounds inside the mission, not the cap table.