Version 0 · October 2026
Nucipal is the memory infrastructure coding agents call, so a product team’s agents perform better and cost less.
01
Product companies with a live product and a stable engineering team.
The product has already been in development for six months to a few years. People change; the product and its knowledge stay. The first partners are smaller startups with at least 4–5 engineers. The next are larger startups. The commercial customer is a company with 50–100 engineers.
Agencies and outsourcing firms sit outside this. Their work is built, handed off, or reduced to maintenance.
A team of a few people, who already know who does what, and whose whole product fits in one agent context, is too small.
02
Programmers use it. The CTO buys it.
In the first startups the CTO is usually a user as well, and can approve the tool directly. In a larger company the buyer is an engineering manager or the head of engineering. Spend under 10,000 a month clears without a procurement team.
Programmers who feel the pain carry it upward. The first note still goes to the CTO when that is the person we know.
03
Coding agents miss company knowledge, so their work is weaker and more expensive.
The models were trained on the public internet. The way this company builds lives in Slack, in agreements, and in people’s heads. An agent with Slack open still cannot pull the fact a task needs, so the task runs on a guess.
That shows up on every agent task that depends on local knowledge. It is urgent when the gain is on the order of half the cost or twice the performance.
Finishing the task on the first try, without another prompt and without being pointed at the source, is the gain on top of that.
04
Nucipal makes a product team’s coding agents perform better and cost less, because they use how that company actually works, and nobody has to maintain a wiki.
05
Automatic infrastructure for coding agents, sold to product teams below enterprise-search contracts.
| Alternative | Nucipal |
|---|---|
| Glean, enterprise search pointed at agents, contracts around 60,000 euro | The same job for a smaller product team, at a price that team can approve. |
| Support-style chatbot agents, with a script for one assistant | Agents that do the engineering work with the rest of the team. |
| Notion, docs, slides, and company knowledge inside ChatGPT | Infrastructure other tools call. It updates from the work itself, including Slack, and runs with Codex, Copilot, or whichever agent the team uses. |
| Re-prompting, asking a teammate, or pasting context by hand | The answer is already in the agent when the task starts. |
One axis is manual versus automatic. The other is application versus infrastructure. Nucipal is automatic infrastructure. The wedge is that the knowledge maintains itself and any agent can call it.
06
Founder outreach to people we already know, on one path: email or LinkedIn.
The list is personal: founders and CTOs in product companies, plus introductions through Digitalna inicijativa Srbije, Tanja Kuzman, Daniela, their angel portfolio, and Unlockit. The first target is 15–20 conversations.
07
A short note, a 20-minute call, a design partnership, then a paid pilot.
08
Twenty messages. One variable. The replies decide the next change.
From each company we keep the task, the workaround, the cost, and whether they take the next step. The segment stays while product teams describe the same performance and cost problem. It changes when the companies are small enough that one agent context holds the whole product.
A batch that misses — few replies from twenty notes — changes the segment, the wording, or the channel, and only one of those.
Customer profile
Business model
| Version 0 | |
|---|---|
| Customer | Product startups with 4–5 or more engineers, then companies with 50–100. |
| Offer | Better agent performance and lower agent cost, with no wiki to maintain. |
| Channel | Founder outreach by email or LinkedIn, starting in our own network. |
| Relationship | The founders onboard the team and sit in a weekly session through the pilot. |
| Revenue | Design partnership is free. Payment starts after the pilot shows a gain in performance or cost. |
| Cost | Model usage and infrastructure for ingesting company context and serving it to agents. |
Design partners
Who. Luka Filipović, CEO. Miloš Medić leads engineering.
Why them. A Belgrade product company, building since 2023, with its own agent in market: Stitch. We know the team. The company is about 4–9 people, on the line of the 4–5 engineer minimum.
What we learn. How their agents get Dynamic Mockups’ own knowledge today, what they already tried, what it costs, and who besides Luka would use a fix.
Product companies we can write to directly. A live product, at least 4–5 engineers, AI already in the work, and the same gap between their agents and how the company actually builds.
Outreach
Zdravo Luka, video sam launch Stitch-a. Svaka čast, ozbiljna stvar. Vi ste produktna firma i proizvod vam već živi neko vreme. Kod timova koji tako grade, agent često nema znanje firme: model nije treniran na tome kako vi radite, a dogovor iz Slacka mu ne uđe u kontekst kad treba da odradi zadatak. Onda performanse i trošak agenta postanu problem. Kako je to kod vas? Da li ste već pokušavali da agent vidi kako Dynamic Mockups zapravo radi, i šta je falilo? Igor
Hvala, to mi znači. Radim na Nucipalu. Agent dobija znanje firme bez wiki-ja koji neko mora da održava, i radi u alatu koji tim već koristi. Imaš li 20 minuta ove ili sledeće nedelje? Hoću da čujem kako vi to radite danas. Igor