From defining the mandate to the printable pack on Monday morning. Every item is part of the application today, not a roadmap.
Size box, cheque, geography, sectors and the weight of each scoring dimension.
AI discovery runs work the open web and the registers, and remember what they have covered.
Candidates are enriched, scored, deep-dived and argued against — with sources.
Fit × timing ranks the list; registry events and cadence keep it current.
Who is investing and where they hunt. Everything downstream reads from here.
WHO is investing: each profile is a complete mandate — size box, equity cheque, geography, sector focus, scoring weights, engine choices and its own company universe. Switch profiles and the whole app follows; presets (including the original DNA) can be instantiated, disabled or edited without losing data.
WHERE you hunt within the mandate: named search territories with a niche brief, signal sources, SIC codes and a cross-run coverage ledger so every discovery run continues where the last stopped. Maps with SIC codes can sweep the entire Companies House register and run a succession radar over the results.
The originator skill that steers every run is layered — a global craft shared by all profiles, a parameter-rendered mandate and a per-profile strategy supplement — and every layer is editable in the UI, alongside per-run-type model/effort choices, API keys and toggles for free primary data sources (FCA Register, Contracts Finder & Find a Tender, CQC, UKAS and more). Prompts are trimmed per run type so each run carries only the guidance it needs.
Companies House for UK profiles, officers, PSC ownership, filing history and iXBRL accounts. Norway, France, Finland, Czechia and Ireland read directly — Norway and France down to officers with dates of birth and filed turnover, which is what makes succession signals work outside the UK.
Contracts Finder and Find a Tender for UK awards, TED across the EU, The Gazette for insolvency and restructuring, the FCA and sector regulators for authorisation, and the GLEIF index for direct and ultimate parents.
A register that does not publish directors is reported as unknown, never as a company with none. Countries with no open register say so, and name what was used instead. No number arrives without its provenance.
The demo is the live application on a sample set of companies.
Turning an empty universe into a researched target list.
Autonomous Claude-powered runs — Discovery (hunt a niche via trade bodies, awards, registers), Enrichment (populate every field with sourced evidence), Scoring (8-dimension fit rubric with per-dimension rationale), Deep-dive (Top-20 dossiers including the anti-confirmation-bias 'why NOT' section and a personalised founder outreach hook) and Thesis synthesis. Every run has token/search budgets, a live streamed activity log and cost tracking.
The screened target list: full-text search, filters for status, ownership, country, score and EBITDA ranges, sortable columns, follow-up flags, note counts, and quick presets (Top 50, Top 20, succession watch, follow-ups due). Every company links straight into Companies House.
Discovery candidates wait here with their evidence and sources until accepted into the universe — the safety valve that keeps weak or unverified finds out of the target list.
Run the research engine on your own hardware. Point PE-CC at an Ollama server and its models join every picker beside the hosted ones, assignable job by job — discovery on a hosted model, letter drafting locally, or the whole pipeline local. Local runs search the web through the app itself rather than the provider, so research still works; they cost nothing and no company data leaves your machine. Each model is only offered for the jobs it can actually do.
Deciding which of them is worth a partner's time, and checking that the judgement holds up.
Top 50 targets ranked with IC-style rationales, a Top-20 origination-priority list with deep-dive dossiers, and an evidence-driven thesis report answering: where should origination time actually go?
An indicative price and return on every company with an EBITDA to work from: an EV range built from a sector multiple band, the senior debt a lower-mid-market deal would carry, the equity cheque that leaves and whether it lands in the profile's box, plus a five-year return sketch. The basis of the EBITDA — reported, estimated or derived — is stated on the face of it, and no sketch is shown at all when there is nothing solid to build on.
One click parses a company's actual iXBRL filings into a multi-year table of turnover, profits, net assets, cash and headcount — reported figures straight from the register, clearly separated from desk-research estimates.
Whether the rubric predicts anything: conversion by score band, which dimensions separated the pursuits that reached a meeting from those that did not, and the recorded reasons the rest ended. Conversion is measured only against companies actually contacted, and the page says plainly when the sample is too small to conclude from.
The timing half of origination: signals that say now rather than someday.
Every UK-registered company is fingerprinted against Companies House. Automatic 12-hourly sweeps raise alerts for newly-filed accounts, officer appointments/resignations and PSC (ownership) changes — the classic signal that a transaction is underway.
Director ages and tenure are computed from Companies House officer records. A 58+ founder with 10+ years at the helm is origination gold — the succession-watch filter surfaces them instantly.
A company already owned by private equity is not a dead end, it is a schedule. The PSC record shows which acquisition vehicle took control and when, and the hold clock turns that date into timing: quiet under three years, most likely to sell between four and six, worth a call beyond that. It flips the sign of the timing score rather than penalising ownership flatly.
Observations kept rather than read once: public contract awards from Contracts Finder and TED, plus open roles, review counts and site numbers recorded by research runs. Enough of them make a series, and a material quarter-on-quarter move raises an alert — “open roles have doubled since March” is a reason to call, and it is invisible in any single look.
Eight European countries publish no open company register at all. Rather than shrug, runs reach for what does exist: the global LEI index for legal identity and parents, and EU-wide procurement records for evidence of real trading with named public customers, dates and values.
From first contact to a meeting, and the reporting that keeps a partnership informed.
Contacts, a typed interaction log (calls, emails, meetings, letters), next actions with due dates and overdue alerts, and days-since-last-touch aging — so warm founder relationships never go cold.
Stage tracking from Outreach sent through Responded, Meeting held and In dialogue to In process or Lost, with pipeline views and funnel counts on the dashboard.
The pipeline as a measured system: stage-to-stage conversion, owner conversations per week (the leading indicator of future deal flow), each market map's yield and cost per accepted candidate, and precedent transactions observed by the research runs. Every metric clicks through to the underlying list.
Proprietary deal flow also comes through people: accountants, lawyers, corporate-finance boutiques and non-execs who see deals early. Track them with regions, sectors, relationship strength and intro paths.
A printable partner-meeting pack: funnel counts, the live pipeline with aging, follow-ups due, the week's registry events, new companies and one-pagers for priority targets.
The last 24 hours on one page: new registry alerts, candidates found overnight, follow-ups falling due and the highest-timing companies in the universe — the morning read before anything else.
Watching the machine itself: what it costs, what it did, and how to plug other tools into it.
The origination picture on one page, arranged how you want it: funnel conversion, weekly intake, registry-event trend, AI spend and cost per accepted company, next actions, succession radar, score and propensity distributions, research coverage, market-map yield and more. Every widget can be reordered, hidden or added back, and the layout is saved to your account rather than the browser, so it follows you between devices.
Expected cost is shown before every launch, expensive actions warn above a configurable threshold, multi-company scoring is grouped into a single run, and prompt caching plus cheaper per-run-type engines cut spend. The Usage page charts spend over time by run type and model, cache hit rates and cost per accepted company.
Activity records every meaningful user action (launches, accepts, profile switches, edits, sweeps) and everything the system does on its own (worker runs, registry sweeps, automation) — each entry clicking through to the run, company or section it concerns.
Open the whole database to an AI assistant of your choice. Thirteen tools cover companies, history, runs and their full transcripts, alerts, market maps, profiles, advisers, funnel and spend — available over stdio locally or as a remote connector URL generated in Settings. Read-only by construction: nothing can be written and no run can be launched, so a connected assistant can never spend AI budget.
An interactive walkthrough of the whole loop over the real app and your real data — resumable, with completed topics ticked off and click-to-jump navigation.