New Renaissance Labs · Vancouver BC · The private equity harness
Create, launch and operate a portfolio of companies from one harness.
Every portfolio company gets its own AI manager, AI workforce, memory, ledger and audit trail. The operator gets a control plane that provisions companies, funds them, prices every action and reads balance and burn across the whole book. Nothing irreversible runs without a human yes.
01 — The category
A business harness runs one company. A private equity harness runs the portfolio.
A business harness uses AI to run one company’s operations. A private equity harness runs every company in a portfolio — and the portfolio itself. It creates companies, launches them, operates them, diligences them and reports on them, under one governance model and one ledger.
The New Renaissance Engine is the first private equity harness: an AI manager and an AI workforce inside every company, and a governed control plane across all of them.
Why now
01
Agents can do the work. Ungoverned, they are a liability.
AI agents can now do the work of a small operating team. Without governance there is no price before a run, no ledger after it and no gate on anything irreversible.
02
The scarce asset is operating bandwidth, not capital.
No partner can sit inside every company. A harness gives each one a standing manager on a fixed cadence, and gives the operator one place to read the whole book.
03
Playbooks become software, with receipts.
The same departments, approval ceremony and price sheet in every company. Every action leaves a receipt: an audit row, a ledger entry where credits moved, a memory event.
The operational gravity of Palantir, the structured clarity of McKinsey and the planning rigor of Oracle EPM — built for people who own and run companies rather than advise them.
02 — The architecture
Three layers. Every company sits on all of them.
The operator works the top layer, each company’s manager and workforce the middle, and the substrate governs both.
Layer 1 — Control plane · the portfolio
Provision, fund, price, watch, govern, resell.
Provision
A claimable company in one action — login, workspace, owner membership, starting credits and a shareable link with an access password. The operating lead claims it with their own email.
Fund and price
Gift credits to any workspace by id, each grant a ledger row with a named actor. Edit the live price per tool and action; every workspace sees the same sheet.
Watch
Every company’s balance, reserved credits and 30-day burn in one table, with CSV export. Mission Control shows the active executor, failover, cadence and errors.
Govern and resell
Global broadcast, cross-company audit, a self-heal sentinel. For partners: brand tokens, a plan builder, template-based client provisioning and a wholesale statement at tier pricing.
Layer 2 — The company · AI manager and workforce
One manager, one workforce, one memory — per company.
The AI manager
Objectives in, a plan as a flow chart out — approved, edited or declined before a step runs. A five-minute patrol, nightly consolidation, a Monday pack, a kill switch, a trust ladder.
The AI workforce
Intelligence, B2B DealFlow, Fundraising & Investments and Social & Online Presence — every department a set of registry actions, priced before they run and metered on one ledger.
Company Memory
14 floors from Theory of Mind at the crown to Market Demand at the foundation, fed by every report, consolidated nightly without a model call, exportable as a checksummed Mind File.
Layer 3 — The substrate · governance and money
A registry, a ledger, a grant, a policy and a router.
Action registry
158 actions across 40 modules — one choke point for every mutation. A button, a plan step and an external agent invoke the same action with the same price and risk class.
Ledger and audit
Both append-only. Credits reserve, settle and release through Postgres procedures; the audit log revokes update and delete at the grant layer, even for the service role.
Isolation and secrets
167 tables under row-level security, 183 policies. Claims minted server-side are the first wall, Postgres policies the second. Secrets are sealed before they reach the database.
AI router and MCP
8 providers on in-house keys with per-prompt model pins. An MCP server with 8 tools lets an external agent drive a workspace through the same registry. The n8n execution layer carries zero AI nodes.
The AI manager
It plans in the open, asks before anything irreversible, and reports on Monday.
Give a company an objective and a commander’s intent. The Daemon Operator drafts a plan as a flow chart — 12 node kinds, bound only to actions in the registry, each step carrying its risk class and credit estimate from the live catalogue. You approve it, edit it on the canvas from a step library of 20 families, or decline it with a reason it recalls. Drafting with the AI planner costs 15 credits; talking to the manager is free.
Approving a plan never approves a gate. Safe steps fire; anything that contacts the outside world, spends credits or cannot be undone waits at its own gate for your release. Every five minutes a SQL patrol checks belief tripwires, error spikes and KPI drift, usually for zero tokens; only a wake triggers one bounded model revision of the situation, rejected if it states a number the system did not supply.
Nightly, memory consolidates in SQL. On Monday the pack arrives — a scorecard composed from the database, decisions taken, the reserved queue and at most one question; silence applies your standing default. Trust is earned one rung at a time, from L0 Watching to L3 Fully trusted, on graded decisions and Brier calibration; demotion is automatic. One button pauses everything within a tick.
| Cadence | What runs |
|---|---|
| Every minute | The job executor drains every company's queue. |
| Every 5 minutes | The daemon patrol checks belief tripwires, error spikes and KPI drift — in SQL, usually for zero tokens. |
| Hourly | Executive Daily Briefing refresh; RFP Radar sweep. |
| Nightly | Memory consolidation: entities, edges, lessons, salience decay, belief re-checks. |
| Monday | The weekly pack: scorecard, decisions taken, reserved queue, risk moves, one question. |
| Monthly · Quarterly | Monthly review and quarterly horizon packs. |
03 — The workforce
Six departments. One registry, one ledger, one memory.
Every department is a set of registry actions — priced before they run, audited when they run, remembered afterwards. The counts are read from the system.
Department of Intelligence
DILIGENCE ENGINE™ reports in minutes with a live ETA, at a depth you choose, priced before you order. Private Investigation on people is purpose-gated server-side. The Legal Department drafts jurisdiction-required documents for your lawyer — never signed, filed or sent.
15 report types · 1–50 pages · 26 legal document types in 6 groups · briefing 10 cr
B2B DealFlow
The Lead Generator prices a plain-English geography before anything runs and charges only for leads found — blanks are never invented. Outreach sends from the company’s own mailbox with one-click unsubscribe. RFP Radar finds public tenders and answers them in the house style.
8 cr per lead found · 14 procurement feed providers · proposal kit 350 cr · CRM
Fundraising & Investments
Capital Command scores fundability with no model in the arithmetic and names the gap behind every missing point. The Valuation Workbench is a pure engine with an assumptions ledger. Data Rooms are token-gated and watermarked; the Investment Thesis and the Wolf Hunt board hold the buy side.
Capital Command · Valuation Workbench · Investor Finder · Data Rooms · Investment Thesis · Wolf Hunt
Social & Online Presence
The Social Media Poster drafts and schedules on a real cadence; nothing is marked ready without a human unless you switch that on. Brand Radar sweeps with per-source proof of life and a Reputation Score. The Chatbot is trained once on Company Memory and answers under ceilings you set.
12 platforms · 14 radar source types · chatbot train 150 cr, reply 2 cr
Company Memory
Event-sourced and rebuildable. Reports feed typed entities into the floors they cover, the Brain Dump files the company’s own records, the Layer Foundry designs a new floor from a plain-English description, and Theory of Mind models every rival, customer and supplier.
14 floors · 17 entity kinds · layer design 250 cr · study a paper 200 cr
Control plane
Root Admin provisions claimable companies, gifts credits, edits the price sheet, reads every balance and burn, broadcasts to every company and audits across them. Partners get brand tokens, a plan builder, template provisioning and a wholesale statement.
158 registered actions · 167 tables under RLS · 17 scheduled automations · 8 MCP tools
Governance
Rules the machine cannot break.
The money law and the audit law live in the database and the action registry, not in a prompt. A bug in a route or a misbehaving agent cannot route around them.
The price, before it runs
Every metered action declares its estimate, the full price table is published inside the product, and an action without a price cannot run. An override can only ever make a job cheaper.
Reserve → settle → release
Credits are reserved when a job is queued, settled at the actual cost and released in full when a run fails. Settlement is clamped at the reserve: an overrun is absorbed by the platform, never charged.
An audit trail nobody can edit
Every invocation — from a button, the manager or an external agent — lands as one identical row. Update and delete are revoked at the database grant layer, including for the service role.
Numbers need receipts
A manager memo that states a number the system did not supply, or cites a memory that does not exist, is rejected by a blocking linter. Reports carry a bibliography, and a deterministic verifier checks citations, arithmetic, contradictions, placeholders and dates.
Zero AI where money moves
The ledger is a Postgres trigger and three procedures. The n8n execution workflows contain no AI nodes, and CI fails the build if one ever appears. No language model sits in the path of a credit.
AI included, on in-house keys
Every intelligence call runs on New Renaissance Labs’ own keys — Anthropic, OpenAI, Google, xAI, DeepSeek, Qwen, Moonshot, Z.ai — metered back to the calling company in credits. No provider accounts to manage. No fallback models: an error is reported as an error.
For whom
Five kinds of operator. One harness.
Each page walks one reader through what changes on day one, with real captures. Partners run the same harness for their clients under their own brand.
01 — Solution
Private equity operators
Operate the portfolio, not just the deal.
Read the page →
02 — Solution
First-time founders
Your first company, with an operating company around you.
Read the page →
03 — Solution
Business owners
A research department and an ops floor for the company you already run.
Read the page →
04 — Solution
First-time small business owners
Everything to set up and run a small business, in one place.
Read the page →
05 — Solution
Existing businesses adopting AI
Adopt the harness without replacing your stack.
Read the page →
Partner programme — /white-label
Run the harness under your own brand.
You bring the relationships; we bring the harness. Brand tokens, a plan builder, template-based client provisioning with starter credits, and a wholesale statement at Tier 1, 2 and 3 — 50% / 45% / 40% of retail.
Private equity firms · PE & market intelligence firms · Law firms · Accounting & CFO advisory · Consultants · Marketing agencies · Communities of businesses.
Read the partner programme →
Consulting — /consulting
The harness, and the hands to deploy it.
Harness deployment and portfolio onboarding, a fractional Chief AI Officer, the DILIGENCE ENGINE™ research desk, lead generation and outreach systems, Company Memory, legal automation, agent integration. A diagnostic, then a deployment, then a standing operate mandate.
Book a scoping call →
The control plane
The operator’s seat, as it is.
Root Admin is the front office of the portfolio: tenants, pricing, gifting, claimable accounts, broadcast, self-heal. Every change is audited. Today the operator reads the book here and works inside one company at a time.
Next step
See the harness on your own portfolio.
A private demo is a real workspace, provisioned for you with credits and an access password — not a slide deck.