Dealflow, notes, and the ops stack a fund already trusts

Funds still run on a weekly partner meeting. The agenda is a list of companies pulled from Affinity, a Slack channel, or a spreadsheet someone updates Sunday night. Each row has a stage, a source, and a partner who said they would take the meeting. Cyriac Zeh and the Nidrosoft team start by sitting in that meeting or watching the recording, because the CRM never shows the side comments that decide a pass. We watch who speaks, who files the note after, and which company dies in the hallway between the Zoom and the record.

Associates carry the intake. They open Gmail with a search for decks that landed over the weekend, download PDFs into a shared Drive folder named for the company, and paste a one-line summary into Slack before anyone has opened the deck. Ops owns the CRM fields that partners ignore: source, geography, check size, and whether a second meeting exists. An executive assistant holds the calendar, the travel, and the pile of founder emails that never become a company record. Nidrosoft maps those three jobs before anyone talks about models.

The stack is familiar even when the logo changes. Affinity or Dealcloud for pipeline. Google Docs for first-meeting notes. A memo template in Drive for investment committee. Carta or a cap-table export for ownership. Excel for the ownership waterfall, the reserve model, and the LP pack. Slack for the fast pass. WhatsApp for the tip that never gets logged. We do not ask the fund to move off those tools. We sit in them and write the missing steps between a message and a field.

Partners ask for fewer lost companies and cleaner IC packets, not a new thesis engine. They want inbound decks turned into a draft company record with a source and a one-paragraph summary. They want meeting notes filed against the right Affinity entry before the next Monday. They want the quarterly LP letter to start from the same workbook the finance lead already maintains. Nidrosoft will draft those artifacts. A partner still decides the pass, the check, and the sentence that goes to limited partners.

Monday morning inbound is a Gmail label, a filter, or a shared inbox that three people check at different hours. A founder sends a deck, a data room link, and a paragraph that repeats the homepage. The associate downloads the PDF, creates a Drive folder if they remember, and writes a Slack blurb so the partner can skim before lunch. Half of those threads never become an Affinity company. The same founder returns in a quarter through a different intro and looks new because the first pass lived only in mail.

Warm intros behave differently. A founder of a portfolio company forwards a note. A limited partner sends a nephew. Another GP texts a name. Those messages skip the inbound label and land on a partner who forwards them with no company name in the subject. The associate guesses the legal name from the deck footer and hopes the domain matches last time. Nidrosoft traces each intro type, because a single intake bot that treats a cold deck like an LP intro will file the wrong source.

Founder updates are a third pile. Portfolio CEOs send monthly emails with burn, runway, hiring, and an ask buried in paragraph six. Someone stars the message. Nobody pastes the metrics into the spreadsheet ops uses for the LP letter. When quarter-end arrives, ops rereads twelve inboxes and rebuilds the numbers by hand. A useful system drafts a structured update from that email and parks it on the company record. A person still confirms the figures before they leave the building.

Duplicates waste more time than empty records. One company exists as a legal name, a product name, and a misspelled domain. Affinity shows three cards. The partner meeting debates a company that already passed in February under the other name. We spend the first week listing how names enter the CRM: deck filename, email domain, LinkedIn company, and whatever the associate typed at 11pm. Then we write match rules the team can see and override.

First-meeting notes start in a Google Doc titled with the company and the date, or they start in an associate's private notebook and never move. The useful version has the problem, the buyer, the ask, the round, and the open questions. The common version is a transcript dump with no owner. Nidrosoft watches where the note is supposed to live after the call. If Affinity has a notes field that nobody opens, we wire the Doc to the company card so the next reader can find it in one click.

Investment committee memos are a collage. Someone copies the deck narrative. Someone pastes comps from a prior memo. The associate spends a day hunting the last clean template and swapping names. NanoBrief is a public Nidrosoft product that turns a short questionnaire into a brief in under two minutes. A fund memo is heavier, and we have not shipped a public IC product. The same drafting pattern still applies. Ask the associate the ten questions they already know, draft the sections, and leave the recommendation line blank for the partner.

Reference calls live in personal notes and phone recaps. An associate talks to a former customer, types fragments into Notes.app, and summarizes in Slack. The next associate on the same company cannot find that call. We treat reference notes as records with a person, a date, a relationship, and a quote the partner can stand behind. The system can draft the summary from a transcript the associate recorded with consent.

Diligence questionnaires arrive as Google Forms, shared Sheets, or a counsel checklist forwarded as a PDF. Answers sit in email attachments named final_v3. Nidrosoft maps who owns each question: product, legal, finance, or the founder. A build can prefill the easy fields from the data room and flag the gaps. Counsel still owns anything that becomes a representation. We will not put a generated paragraph into a closing binder without a named reviewer, and we say that in the scope before anyone writes a prompt.

Quarter-end at a small fund is one person on ops with a master workbook. Tab one is the portfolio list. Tab two is cost basis. Tab three is last reported ARR. Tab four is a narrative graveyard from last quarter. They email founders for updates, chase the ones who ignore them, and paste numbers into cells that still hold last quarter's comments. Nidrosoft sits with that workbook before we talk about a portal. If the letter is honest, it starts from the sheet the finance lead already trusts.

Capital calls and distributions have their own trail. A notice goes to LPs from a mail-merge, a fund-admin portal, or a PDF attached to a thread. Wiring instructions live in a password manager and a printed sheet in a drawer. We do not touch money movement. We will draft the cover note from the call notice and file the sent copy against the close calendar. A human still sends the notice and confirms the wires. If a fund wants an agent that presses send on capital calls, we decline that scope.

Board decks from portfolio companies arrive as attachments the week of the meeting. The partner skims, the associate extracts three metrics, and ops hopes those metrics match the LP workbook. They often do not, because the CEO used a different definition of net revenue. A useful system extracts the figure and the date, then parks a discrepancy next to the cell. Someone at the fund still decides which number goes in the letter. Filing is the work we can take.

LP meetings need a packet: performance, pacing, a few company stories, and a pipeline slide that someone rebuilds from Affinity the night before. Nidrosoft can assemble the first draft from the workbook, the CRM export, and the last letter. We keep the anecdotes in a folder the partner edits, because a generated company story that invents a customer is how you lose a re-up. The packet ships when the partner says the words are theirs.

Nidrosoft has not shipped a public product for venture funds. There is no branded dealflow app on this site, and we will not dress a clinic or a dealer system as if it were a fund. Cyriac Zeh still does this work the same way he does every other industry. Sit with the people who file the records, rank the hours, and build into Affinity, Gmail, Drive, and Excel. If you need a logo on a slide that says we automated a specific fund, you will not find one in the archive.

TokenTra is live at tokentra.io. Teams that already pay OpenAI, Anthropic, Google, Azure, or Bedrock connect those bills and see spend by feature, team, and project. Funds that bought a pile of seats often cannot say which associate burned the month. Public results on this site say teams cut 10 to 30 percent of wasted AI spend after they can see the waste. We will not promise that number to a fund that has not connected a provider.

Nerlude is live at nerlude.com. It is a control center for domains, hosting, databases, API keys, and renewals. A seed fund with a website, a data room vendor, a CRM contract, and a forgotten staging app already has this mess. Portfolio companies have it worse. Keys live in Slack. Domains sit at a registrar the last CTO used. Nerlude is how ops sees renewals 30, 14, and 7 days out instead of learning about an outage from an LP who cannot open the site.

Calendar, travel, and inbox coordination is the other adjacent pattern. Partners live in holds, airport changes, and meeting briefs that an assistant compiles from LinkedIn and the last memo. We have built that class of ops around profiles, conflict checks, and a brief that lands before the call. We have not published a fund-branded assistant. The work still looks like a draft brief in the calendar invite and a follow-up that files the note. The assistant remains the person who sends it.

Embed means we sit with associates, ops, and the assistant, not only the managing partner. We watch a Monday inbox, a partner meeting, and a quarter-close week if the timing allows. We take notes on the workarounds: the private spreadsheet, the Slack emoji that means pass, the Drive folder nobody can find. Nidrosoft does this from San Diego and on the fund's accounts. A fund is another set of systems with a person who already owns them.

Diagnose means we cost the recurring tasks. How many hours go into filing inbound decks. How long an IC memo takes when the template is clean versus when it is not. How many days ops spends on the LP pack. We score each task on hours, feasibility, and whether a person must still approve the output. The heat map is a ranked list, not a vision deck. Partners often expect the first item to be sourcing. The first item is usually filing, notes, or the letter, because those tasks already have a shape.

Scope means the fund buys a result. Inbound decks become draft Affinity records with a source and a summary the associate edits. Meeting notes land on the company card within a day. The LP workbook gains a draft narrative tab that ops accepts or rewrites. We write what runs itself, what a person approves, and what we will not touch. Investment decisions, capital calls, and LP legal notices stay human. If a partner wants a bot that votes at IC, we stop the scope conversation there.

Build and Own happen in the fund's stack. We ship into Gmail, Affinity, Drive, and Excel with someone on ops or an associate building alongside. Training happens on their live deals, not a toy dataset. When we step back, the workflows keep running on their accounts, their policies, and their audit trail. Nidrosoft stays on for what they ask and stays out of the rest. There is nothing to hand over because it was theirs from the first commit.

An AI audit at a fund is two weeks in the work. We read the inbound label, the Affinity pipeline, the memo template, and the LP workbook. We interview the associate who files decks, the ops lead who owns quarter-end, and the assistant who guards the calendar. Cyriac Zeh will tell you in the first call if the request is a thesis workshop. Nidrosoft does not sell a new way to pick winners. We sell a map of where filing, drafting, and reporting eat the week.

We cost tasks in the units the fund already uses. Minutes per inbound deck. Hours per IC memo. Days to close the LP letter after the last founder update arrives. We write the tools next to each task so nobody pretends a chat window will replace Affinity. The output is a ranked roadmap with a first build that a small team can run. If the first build needs six vendors and a data-warehouse project, we say the fund is not ready and we should stop.

Risks go on the same map. A generated summary that invents a customer. A merge that hides a live deal. A draft LP sentence that states a valuation the finance lead has not signed. We mark those as human gates before anyone asks for a demo. Funds that want speed without gates should buy a writing tool and accept the errors. Funds that want Nidrosoft want the gates named in the scope.

The audit is the right first buy when partners disagree about the problem. One partner thinks sourcing is broken. Ops knows the LP letter is the fire. Associates know Affinity is a junk drawer of duplicate companies. Two weeks of sitting in the inbox and the workbook ends that argument with a ranked list. Email cyriac@nidrosoft.com if you want that list. Book the public calendar if you already know whether filing, notes, or the pack should go first.

A first build we will take is inbound to a draft company record. The deck arrives, we extract name, domain, round, and a short summary, and Affinity gets a card in a review queue. The associate accepts, edits, or rejects. Source stays a required field because a fund that cannot say how a deal arrived will lie to itself later. We log the model, the prompt version, and who clicked accept. That log is how you explain a bad summary in a year when the associate has left.

A second build we will take is notes to the company card. After a call, the associate drops a transcript or a rough Doc. The system drafts the structured note and links it in Affinity. The associate has a deadline, usually before the next partner meeting, to mark it filed. Late notes stay visible. Partners who want notes without a human file step will get a graveyard of drafts that nobody trusts. We would rather ship a smaller queue that stays clean.

A third build we will take is a draft LP packet from the workbook, the CRM export, and last quarter's letter. Ops reviews every figure. Partners review every company paragraph. The send stays on their mail and their letterhead. We have seen adjacent drafting work in NanoBrief, where briefs come back in under two minutes against a questionnaire. A fund letter is slower on purpose. Speed that skips the finance lead is how you send the wrong NAV.

Larger funds sometimes want a transformation program: audit, two or three builds, training for associates, then a second team such as platform or finance. We will do that when there is a named owner who can approve access to Gmail, Affinity, and the workbook. A build without access is a slide. Nidrosoft will not start implementation on screenshots. If IT needs a week for permissions, we wait. The calendar is public for the conversation that decides whether we wait or walk away.

Many funds already pay for ChatGPT seats, a few Claude seats, and a custom GPT that one partner likes. The invoice lands on a card. Nobody can say which workflow used the tokens. TokenTra exists for that mess. Connect the providers, break spend down by team or feature, and set alerts at 50, 80, and 100 percent of a budget. Auto-disable is available when a key blows the cap. A fund that only wants more seats should buy the seats. A fund that wants the bill attributed should look at TokenTra or a custom view on top of it.

Infrastructure hygiene shows up when the fund site, the data room, and a side project share a credit card. Domains auto-renew on a personal Gmail. A contractor still has the AWS key. Nerlude is the public product we point at for renewals, encrypted credentials, and time-limited access. Platform teams at funds who help portfolio companies can use the same pattern. We will not pretend Nerlude is a portfolio monitoring suite. It is a control center so a forgotten domain does not become an LP email.

Protectron is live for EU AI Act work: risk classification, document drafts, and checklists by risk level. A fund that is not placing models in the union may not need it. A portfolio company that is shipping an agent into Europe might. We mention it so nobody asks us to invent a compliance opinion in a memo. If the company needs a filing, they need counsel and a product like Protectron, not a paragraph we generated from their homepage.

Corporate AI training for a fund is a half-day on their inbox and their CRM. Associates bring live deals. We write prompts against their memo template and their Affinity fields. We do not run a generic tools tour. If the request is a lunch-and-learn with no access to real threads, we will say a public workshop is the better buy. Training that cannot touch a live deal does not change Monday.

Hire a strategy firm if the board wants a thesis on AI in the portfolio and no one will give us Gmail or Affinity. Nidrosoft is a poor fit for a slide that never becomes a filed note. Cyriac Zeh will say that on the first call. We would rather lose the work than sit in a partner offsite that ends with a framework and an action item nobody owns. The about page states the same rule in plainer words. We build into the systems you already run.

Buy ChatGPT or Claude seats if the only request is access and a one-page acceptable-use policy. We will not wrap seat rollout as an AI transformation. TokenTra can still help once the seats exist and the monthly invoice is ugly. Until someone can name a repeating workflow in Affinity or the LP workbook, there is nothing for Nidrosoft to scope. Associates who already write well in the model do not need us to watch them do it.

Do not book us if the decision maker will not be in the room. An associate who wants a side project, a vendor who wants a logo, or a consultant who wants us as a sub without access to the CRM will waste a week. We work with funds that can grant approved access, name an ops owner, and say which workflow pays back first. Smaller emerging managers fit when one build pays for itself. Larger platforms start in one pod, not across every sector at once.

We are also the wrong seat if you want us to source proprietary dealflow or pretend a model found a company nobody else saw on LinkedIn. Sourcing theater is not a Nidrosoft product. Filing inbound decks, meeting notes, LP packets, token spend, and renewals are. If that list is too operational for the partnership, we will say so on the first call and leave the calendar open for someone whose Monday looks like that list.

Write cyriac@nidrosoft.com with the fund name, the CRM you use, and the workflow that hurts. Dealflow filing, notes, LP pack, or the AI bill. Attach a redacted screenshot if you can. Cyriac Zeh reads that inbox. If the note is a pitch for us to raise a fund or to take a meeting with no owner, we will decline. If the note names a system and a person who can grant access, we will book the audit conversation on the public calendar.

The first live process should be boring on purpose. A review queue in Affinity. A note template that files. A draft tab in the LP workbook. We measure it in hours returned and errors caught, not in companies sourced. Nidrosoft has shipped 125 products and keeps 16 public systems in the archive. None of those 16 is a fund OS. The number that matters for you is whether Monday's inbound is filed before the partner meeting.

You own the workflows on day one. They run in your Google Workspace, your Affinity instance, and your cloud. We use Azure, AWS, or Google Cloud when a new piece of infrastructure is required, and private infrastructure when you already have it. EU hosting is available when a policy requires it. Audit trails stay on. After we leave, ops should be able to explain the queue to a new associate without calling San Diego.

Software companies in the portfolio often need TokenTra and Nerlude before the fund does. Clinics and plants in the portfolio have their own pages. We will not collapse those industries into a fund story. If you want the method in writing, read how we work. If you want the person, read about. If you want the calendar, use contact. The work starts when someone who can approve access is on the thread.