Every office has a person who knows where the useful file is buried. Ask about a discontinued component, a warranty claim from 2019 or the latest customer deck, and this colleague performs a small miracle involving an old inbox, a network drive and institutional muscle memory. amber is trying to turn that colleague into infrastructure.
The Aachen company connects the places where a business already keeps its knowledge - Microsoft 365, Confluence, Jira, email, document systems, databases and local drives - and gives employees one place to search, chat and act on it. The interface is the easy part. Underneath sits amber’s increasingly important product: a permission-aware data layer that organizes versions, relationships, ownership and context before a large language model sees anything.
That distinction explains why a company once called amberSearch dropped the second half of its name. Search was the entry point. The ambition now is a European business-AI platform where assistants answer from company evidence and agents can carry out work across systems. In August 2026, Ventech and NRW.Venture put €7 million behind that transition.
The origin
First, make the filing cabinet answer back
amber’s history has a few starting lines. The project traces back to ambeRoad work in 2017. The team says it was focused full-time on access to internal information by 2020. The present company lists 2021 as its founding year, when Philipp Reißel, Bastian Maiworm and Igli Manaj built amberSearch out of the RWTH Aachen orbit. Reißel now runs the company as CEO, Maiworm leads revenue as CRO and Manaj leads technology as CTO.
Their original observation was practical, not mystical. Data piles up faster than employees learn where it lives. Experienced workers retire. New hires ask people instead of systems. Conventional search is often trapped inside one application and depends on remembering the right keyword. The founders built a layer across those silos, preserving each source system’s access rights and returning results with document links and highlighted evidence.
Today the catalog has four visible faces. amberSearch retrieves knowledge. amberAI chats, writes and researches against it. amberAgents packages repeatable assistants and actions. amberProjects keeps a task’s conversations, sources and outputs together. More than 70 standard connectors support the platform, while desktop, mobile, Teams, Outlook and browser extensions put it inside existing routines. APIs, widgets and MCP connections let developers trigger processes or update records.
The turning point
The search box refused to stay small
The revealing customer is Zentis, the 125-year-old food company. Its international R&D knowledge was split across countries, languages, Microsoft 365, a document system and network drives. Zentis already had experience with enterprise search, but the existing system left room for improvement. The plan was to test amberSearch narrowly on German systems and selected English material with a group of key users.
Then the pilot changed shape. Search quality, speed and an internet-like interface persuaded Zentis to make amber a central part of its SharePoint R&D portal. Staff nicknamed it “Umbrella Search.” The next phase connected more international sources and forced amber to solve what had failed first: employees could find German material, but many could not understand it. Cross-language search and machine translation turned access into actual reuse.
“The competition for models is more or less over. The quality of responses depends on the context you ingest.”Philipp Reißel, co-founder and CEO
R-Biopharm supplies the less polished sequel. Its sales and service teams could spend tens of minutes hunting through Teams, Outlook and network drives for certificates, guidelines and device knowledge. amber connected the first systems with internal IT in half a day. The early software was still visibly young, but fast feedback created enough “aha” moments for R-Biopharm to expand it. Only then did the harder problems surface: continental rollout, more languages and multiple sub-permission systems.
This is the company’s repeatable motion. Start with a painful retrieval job and a controlled user group. Connect rather than migrate. Prove speed and relevance. Expand to more sources, languages and departments. Only after that foundation works do assistants and agents become credible. It is less cinematic than releasing a general chatbot, but it matches how risk-conscious IT departments buy.
The economics
What the tidy layer costs
amber now publishes pricing, which makes the business unusually legible. Starter costs €9.99 per user each month and includes amberAI and amberAgents. Professional adds amberSearch and up to 10 connected systems for €13.99 per user, with a 50-seat minimum. Business costs €17.99 per user, also with a 50-seat minimum, and adds up to 20 systems, OCR, broader API access and MCP. Enterprise begins at 500 seats and is quoted individually. Retrieval tuning, rollout planning, integrations and training can arrive as paid onboarding.
For a 50-person team, the list-price floor is therefore €699.50 a month for Professional or €899.50 for Business, before onboarding and tax. Automated API and MCP work uses prepaid monthly token budgets of €100, €250 or €500 on top of an eligible seat plan. amber says it adds 5 percent to the underlying model provider’s token price.
The infrastructure pitch includes a second economy. amber claims intelligent retrieval and prompt engineering can send 35 to 40 percent fewer tokens per request by supplying models with selected, structured context instead of a document dump. That percentage is a company claim, not a universal law. The sound idea beneath it is reusable preparation: clean and govern knowledge once, then let every search, assistant and agent draw from the same layer.
The market
Not another chatbot, but still a crowded room
amber sits between enterprise search, knowledge management and agent infrastructure. Glean is the closest international comparison. Microsoft 365 Copilot, Google Agentspace, Elastic, Coveo, Sinequa, Guru and Langdock approach pieces of the same problem. The other competitor is the internal build: a company’s own connectors, vector database, retrieval pipeline, access-control logic and favored model.
amber’s angle is European mid-market pragmatism. Development and default hosting stay in Germany and Europe. The company says customer data is not used to train models, existing permissions remain in force, and the platform is GDPR compliant and ISO 27001 certified. It supports cloud and on-premise sources, multiple model providers and self-hosted enterprise deployments. That package matters to a manufacturer with decades of technical files and a small AI team more than a leaderboard score does.
The €2.1 million seed round led by Ventech in March 2025 funded the jump from search toward assistants and automation. The 2026 Series A, co-led by Ventech and NRW.Venture, is meant to deepen the data layer, add integrations, grow the team and expand into Benelux. A 2024 distribution partnership with document-software company BCT already offers a route through more than 75 European channel partners.
The stealable playbook
Five things another operator can copy
- Choose one frequent, measurable information hunt instead of launching “AI for everyone.”
- Run a pilot with key users who know what a good answer looks like.
- Preserve source permissions and citations from day one. Trust is a product feature.
- Meet people in Teams, Outlook or the intranet instead of demanding a new daily habit.
- Expand only after retrieval works - first more sources, then languages, then actions.
The catch
A context layer cannot rescue rotten context
There are conditions under which this does not work. If documents are stale, ownership is unclear and permissions are already chaotic, amber can index the disorder more elegantly but cannot declare which specification is true. If the valuable workflow happens twice a year, the integration may never repay its cost. If nobody owns adoption, even a good answer box becomes another tab. And if a company lives entirely inside one well-governed software suite, the cross-system advantage shrinks.
Failure conditions
Bad source data, unmanaged access rights, weak rollout ownership, rare workflows and no baseline for measuring time saved. Add autonomous actions before those basics are stable, and the blast radius grows faster than the value.
The more interesting limit is cultural. amber’s product depends on employees asking the system rather than the veteran at the next desk, and on experts allowing their knowledge to become shared infrastructure. The company can make that behavior easier. It cannot make a guarded organization generous.
Still, amber has found a sensible place in the market. Its customers are not buying an oracle. They are buying fewer scavenger hunts, faster onboarding and a controlled path from answers to actions. The Series A wager is that Europe’s mid-sized companies will want agents, but will first pay someone to make those agents less clueless. The boring layer may be exactly where the useful work begins.
Go deeper