About Amberscript
Amberscript is an EU-based AI company building best-in-class language and communication tools. It runs two product lines: Amber Notes, an AI meeting assistant that joins calls, transcribes them, and delivers structured summaries with action items; and Amberscript Media, an AI subtitling service trusted by major European broadcasters including ZDF, ARD, M6 and Pro7. The company is ISO 27001 and ISO 9001 certified and fully GDPR and NIS2 compliant, with all data processed and stored inside the EU.
AI is at the core of the product. Every summary, action item, and "ask AI about this meeting" answer in Amber Notes is generated by large language models. Because Amberscript serves regulated industries, healthcare, finance, legal, government, that AI has to be both best-in-class and provably compliant, which makes how and where each model call is routed a first-class engineering concern.
The Challenge
As Amber Notes scaled, its AI surface area grew: summaries, custom per-user templates, action items, and conversational chat over transcripts. Each feature called a single model provider directly, and that tight coupling started to hold the team back.
- Provider lock-in. Hard-wiring one LLM API made it slow and risky to adopt newer, better, or cheaper models as the landscape shifted week to week.
- No cost visibility. AI spend arrived as one bundled provider bill, with no breakdown by feature, plan, or model to manage unit economics.
- Reliability exposure. A single provider outage or rate limit could degrade summaries for every customer at once, with no automatic fallback.
- Compliance overhead. Keeping inference routed through appropriate, EU-fit models, and auditable, added engineering work to every model change.
“We didn't want our product roadmap dictated by a single model vendor. We needed to move between models freely, without rebuilding our pipeline every time.”