yggy: Product, team and how it was built
yggy (Self-building knowledge graph from company documents, for people and AI agents, Germany): Free and paid plans live after a #20 Product Hunt launch, November 2026. Evidence grade E: no revenue figure was checked. Team: 3 co-founders (CEO, CTO, CRO); headcount not published. Launched 20 October 2026; still running.Added 6 November 2026
- Outcome
- Free and paid plans live after a #20 Product Hunt launch, November 2026
- Evidence
- E no revenue figure was checked
- Launch
- 20 October 2026
- Status
- Still running
- Team
- 3 co-founders (CEO, CTO, CRO); headcount not published
- Outside funding
- No outside funding found in the sources; run by Yggdrasil UG, a German limited company
- Main channel
- Not stated
What is not verified
No revenue figure was found. We looked for a payment-processor profile, a founder interview with numbers, a company filing and a funding announcement, and found none. The pricing data contains live price identifiers for the paid plan, but nothing shows how many customers pay. A reader could look for the company's annual accounts in the German company register (Unternehmensregister) or ask the founders. This story is grade E: the business is real and its product, price and customers are described from public pages, but no revenue figure was checked. Aiolum shows it as an idea worth studying, not as proof that it makes money.
yggy builds a knowledge graph, a database that stores things (parts, suppliers, contracts, people) and the relations between them so that both people and software can query them, from a company's scattered documents. A language model reads each document and extracts triples, facts of the form subject, relation, object, inside a fixed ontology, a formal list of the allowed kinds of things and relations. Each fact keeps its source and a confidence value, and agents query the graph through MCP (Model Context Protocol, a standard way for AI agents to call outside tools).
Three founders built it: Carlos Gienow (CEO), Dominik Müller (CTO) and Max (Maximilian) Pfestorf (CRO). Their company, Yggdrasil UG, develops it in Germany and hosts it in the EU. It launched on Product Hunt on 20 October 2026 and placed 20th for the day. The pricing page offers a free plan and a Pro plan that the company lists at 20 euros per member a month.
No revenue figure, customer count or user count was found. The story is grade E: it shows a live product with published prices and a dated launch, not a business result.

How much revenue does yggy make?
Each row shows the evidence grade and the source numbers from the list at the end of this page. The grades are explained in How Aiolum verifies.
| Fact | Value | Grade | How we know | Sources |
|---|---|---|---|---|
| Status, November 2026 | live web product with a free plan and a paid Pro plan | C | Pricing data read on 6 November 2026: plan "pro", "monthlyPrice":2000, "annualPrice":20000, "currency":"eur" (euro cents), with metered prices for credits and storage. | [2] |
| Product Hunt launch | 20 October 2026, rank 20 for the day | B | Product Hunt page read on 6 November 2026: featured 20 October 2026, "# 20 Day Rank", 81 points and 86 followers. The MilestoneWins story of 21 October 2026 gives 79 launch-day upvotes. | [3, 4] |
| Founders | Carlos Gienow (CEO), Dominik Müller (CTO), Max Pfestorf (CRO) | C | The company's press page. Previous employers it lists: verticus, Audi, WHU (Gienow); Bosch, Mercedes-Benz (Müller); Kyndryl, Blockbrain (Pfestorf). | [1, 4] |
| Company | Yggdrasil UG, Germany | C | Press text: "Yggdrasil UG develops yggy in Germany." A UG is a German limited company with low minimum capital. No founding date was found. | [1] |
| Team size | headcount not published | C | The press page lists only the three founders as the team. MilestoneWins: "3 Founders behind yggy". | [1, 4] |
| Outside funding | none found | C | No funding round is mentioned on the company pages, the Product Hunt page or the MilestoneWins story. | [1, 3, 4] |
| Pro price | 20 euros per member a month, about US$22.33 | C | Converted at 1 euro = US$1.1163 (Federal Reserve H.10, 30 October 2026). | [2, 5] |
| Hosting | managed EU hosting, private cloud or on-premises | C | Press text and pricing page. Self-hosted models only on the Enterprise plan. | [1, 2] |
| Ways in | web app, REST API, MCP, CLI | C | Press text: "Developers reach the graph through MCP, a CLI and APIs in yggy core." | [1] |
| Time from launch to the status fact | 1 month | calc | Calculated: October 2026 to November 2026. | [3] |
How to read this table: a grade C value is a statement by the company on its own pages or by its founders on Product Hunt. A grade B value comes from an independent page that shows a count it measured itself (Product Hunt's own vote counter). A "calc" value was computed on this page. No revenue figure exists.
| Plan | Per month, billed yearly (US$) | Per month, billed monthly (US$) | Limit |
|---|---|---|---|
| Free | 0 | 0 | up to 10 members, 100 credits (about 5,000 facts), 1 GB, 100 agent calls a day |
| Pro | 18.61 | 22.33 | per member; 1,000 credits per workspace, 5 GB per member, SSO, audit log |
| Extra credit (Pro) | 0.03 | per credit above the allowance | |
| Extra storage (Pro) | 0.67 | per GB-month above the allowance | |
| Enterprise | custom | custom | SLAs, per-case pricing, private cloud or on-premises, self-hosted models |
The company states Pro at 20 euros per member a month, or 200 euros per member a year (two months free), excluding VAT. Pro includes 1,000 extraction credits per workspace a month (the page says about 50,000 facts) and 5 GB stored per member. Above the allowance, Pro bills 0.03 euros per credit and 0.60 euros per GB-month, unless the owner sets it to stop. Agents and API keys never take a seat. A separate result fee for "Care" (conflicts resolved, duplicates merged) is capped by a monthly budget; results reverted within 14 days are not charged. The Product Hunt launch offered 25% off for 3 months.
When did yggy launch, and how did it grow?
From the first articles on the company blog in July 2026 to a Product Hunt launch in October 2026. Each entry carries its grade.
- July 2026CEarliest dates in the site map for articles on the company blog (the "Grove"), on why AI agents fail without maintained company knowledge. These are last-changed dates, not proven publication dates. [1]
- 17 October 2026CDominik Müller posts "Almost there" in the Product Hunt forum and writes that the team "broke a lot of graphs and merged things that should never have been merged", as MilestoneWins quotes him. [4]
- 20 October 2026BLaunch on Product Hunt with the tagline "The AI knowledge graph that builds itself" and 25% off for 3 months. It ends the day at rank 20. [3, 4]
- 21 October 2026CMilestoneWins publishes a founder story on the launch. [4]
- November 2026CFree and Pro plans live; the company blog keeps adding articles on knowledge graphs and product compliance. [1, 2]
Why did yggy work?
Four causes, in the order they mattered. Each one is a reusable lesson.
Sell the part that makes graph projects fail
Extraction is the easy part of a knowledge graph. Entity matching (deciding that two names mean one supplier), contradictions between sources and updates when documents change are the slow work. yggy makes that work the product, and charges per accepted result for its automatic clean-up.
Every fact carries its source
Each extracted fact keeps the source document, the supporting text and a confidence value, and every merge can be undone. An answer can be checked before someone signs it, which is what a quality or compliance team needs.
One graph, three users
People use the app, code uses the API, and agents use MCP. Agents do not take a paid seat, so the price grows with the documents read, not with the number of agents.
EU hosting as a selling point
The company hosts in the EU and offers private-cloud and on-premises installations with self-hosted models, for German industrial customers who do not want to depend on a US provider.
How it was built
These details come from the company's own pages and the founders' notes on Product Hunt. They are grade C. [1, 2, 3]
- A language model extracts facts into an RDF-based ontology with a bounded OWL 2 RL profile; tables and key-value blocks go through a cheaper schema-guided path the company calls SGSE.
- The graph is stored in Dgraph, a graph database; facts export as JSON-LD or N-Quads, the ontology as OWL, SKOS and SHACL.
- The founders list Claude Code and OpenAI Codex as their coding tools, and Bun, Vite, Linear, Resend and PostHog in their stack on Product Hunt.
- They write that open models from Hugging Face helped them experiment and fine-tune. The model providers used in production are not named.
- The website is a React single-page app; its text and prices load by script.
Tools named in the sources
Each one opens its page in the Tools tab, with every story that uses it.
Who tried this before yggy, and why did they fail?
Three companies in our database of failed startups tried to connect scattered information into one source of answers and failed. The values below are from that database and are not independently verified.
Omnity
The promiseSemantic search that connected facts across unstructured sources for pharma, academic and legal research.
Why it diedLarger companies with their own data and AI products took the market, and data processing cost more than the revenue it brought.
Niles
The promiseA Slack bot that collected a company's knowledge and answered staff questions.
Why it diedSlack and others added similar features, AI answers became a commodity, and setup and upkeep were a burden for customers.
Singly
The promiseOne API that gathered a user's data from many online services.
Why it diedIt depended on other companies' APIs, which changed under it.
The failed startups come from our database of 1,749 dead startups, retold here in our own words. Its entries are AI-assisted summaries of public sources, so these cards carry no evidence grade.
What changed
Omnity and Niles needed people or fixed rules to turn documents into structured facts, and that work did not pay for itself. A language model now does the first extraction; the pricing page states that a typical page gives 10 to 25 facts and that one credit gives about 50 facts at Standard reading, so the remaining work is checking and correcting, which yggy puts in a review queue.
AI agents also created a new buyer: a team that runs agents on company data needs answers it can trace to a record. Whether that need is large enough for a small company, against the knowledge features of the big platforms, is not yet shown.
Can one person copy yggy now?
Hard for one person.
The product is a data platform with an ontology, a graph database, entity resolution and enterprise hosting options, built by three people with backgrounds at Bosch, Mercedes-Benz and Kyndryl. It is sold to industrial companies. A solo founder could copy the idea in a narrow niche, for example one document type in one industry.
Where it is still open: Narrow, traceable knowledge bases for one industry: product compliance, quality complaints, supplier data, legal clauses.
- 1Pick one regulated document set where answers must be traced to a source (certificates, complaints, supplier records).
- 2Use a language model for extraction into a small fixed ontology, and keep the source and a confidence value on each fact.
- 3Give experts a review queue for unclear merges and make every merge reversible.
- 4Offer MCP access so the customer's AI agents can use the result.
What could go wrong for yggy?
The risks are the mirror image of the causes above.
What is not verified
No revenue figure was found. We looked for a payment-processor profile, a founder interview with numbers, a company filing and a funding announcement, and found none. The pricing data contains live price identifiers for the paid plan, but nothing shows how many customers pay. A reader could look for the company's annual accounts in the German company register (Unternehmensregister) or ask the founders.
Unsourced credentials and case pages
The German About page states "350+ umgesetzte Use Cases" and "80+ KI-Systeme im Einsatz" (350+ use cases delivered; 80+ AI systems in use) (translation) without saying whose work these are, probably the founders' earlier jobs. The site's case pages (automotive, life sciences, legal) were not checked as named customers.
Competition
Large platforms (Microsoft Copilot, Palantir, graph databases with GraphRAG add-ons) sell company knowledge for AI, and the company itself writes comparisons against Notion, Obsidian and Palantir. On this site, Recall is the nearest story, a personal knowledge base, not a company graph.
Dependence on model providers
Extraction needs a language model for every document. The security page says "Every model provider in your deployment is contractually bound" not to train on customer data, but does not name the providers; only Enterprise customers get self-hosted models. A price or terms change by a provider would hit the cost of every credit.
Sources
The grade next to each source is the best grade that the source can support. "Read in full" means that the text of the page was downloaded and the numbers were checked in it.
- C yggy website (yggy.tech): home, about, press, FAQ, security and sitemap, read 6 November 2026The pages load their text by script. Read the FAQ page's structured data and the English and German page text in the site's translation file, which the pages load. Source for the press text, the founders, the company name, the hosting and the technical approach.
- C yggy core pricing page and its plan data (yggy.tech/pricing and yggy.tech/api/plans), read 6 November 2026Read the plan data in full and the pricing text in the translation file. Source for the prices and limits.
- B Product Hunt: yggy, The AI knowledge graph that builds itselfRead in full on 6 November 2026. Source for the launch date, day rank, points and followers (Product Hunt's own counters), and the founders' notes on the tools they built with.
- C MilestoneWins, 21 October 2026: The three builders automating the part of knowledge graphs that makes teams give upRead in full. A founder story listed with the founders' names as authors; quotes the pre-launch forum post and gives the launch-day figures.
- A US Federal Reserve H.10, euro exchange rate historyRead the row for 30 October 2026: 1.1163 US dollars per euro.
- C Database of failed startups, local exportLocal export. AI-written summaries, not independently verified. Source for the failed predecessors.
Run this company? Send corrections or a verified figure to aiolum@livyden.com