Built because AI never learned.
Three years of infrastructure. One week to make it intelligent. The pre-model reasoning layer grāmatr became was not a product concept — it was a production solution to a problem every AI deployment shares.
From digital agency to context engineering pipeline.
Gra Matr
Brian Handrigan founded Gra Matr as a digital agency. Brand engagement, digital media strategy, competitive analysis, campaign deployment — the work that every brand needs and few do well: turning research into strategy and strategy into measurable results.
The domain you're reading this on — gramatr.com — has been owned continuously since 2007.
ChatGPT changes everything
November 2022. ChatGPT launched publicly. The potential was obvious. So was the problem.
Every session started from zero. Every conversation forgot what came before it. Architecture decisions, preferences, codebase context — re-explained every morning to an AI that had no record of the day before. The tools were powerful and fundamentally broken at the same time.
Building context infrastructure
Before turnkey RAG systems existed, the grāmatr infrastructure was being built from scratch: vector memory, embeddings, similarity search. Not because it was trendy — because a live production platform required it.
The NEXT90 cross-media attribution platform was under simultaneous development. Every AI session reset. Agents forgot architecture, forgot decisions, forgot preferences. More time was spent re-explaining the codebase than building new features.
This is when grāmatr's founding team developed hands-on infrastructure knowledge at the layer companies like Mem0 would later productize. The pattern was understood because it was built before those products existed.
The intelligence breakthrough
By early 2026, grāmatr was running as a context foundation: knowledge graph, vector search, MCP tools for Claude Code. It worked. But the CLAUDE.md file — the instruction set that tells the AI how to operate — had bloated without limit. Every rule, every preference, every pattern, crammed in because the system could not learn them on its own.
In one week — March 21 to 28, 2026 — the routing engine was built. A decision router with trained classification models. Seven effort levels. Twenty-five capability categories. Intelligence packets that replaced the bloated instruction file with the precise context each request needs. Within that week the new architecture was routing production traffic, with learning corrections recorded live.
The instruction file collapsed to a fraction of its size — and performance was better after than it had been before. Better results, fewer tokens.
That wasn't compression. That was a system that had learned.
Infrastructure. Intelligence. Enterprise.
grāmatr's mission is to become the context engineering layer for enterprise AI — starting from the same operational problem every AI deployment shares, solved in production before grāmatr had a name.
Infrastructure
Knowledge graph, vector search, MCP tooling, classification pipeline. The platform layer that enterprise AI deployments need — built before turnkey alternatives existed, proven in production before grāmatr had a name.
Intelligence
Pre-classification routing that delivers the right context to every AI request. Seven effort levels. Twenty-five capability categories. A patent-pending intelligence packet replaces the context that AI tools have always had to be re-told.
Enterprise
Organizational intelligence with governance controls at every level. Compliance-ready deployment. Admins control what patterns are shared, what stays private, and how the platform integrates with existing infrastructure.
Not another AI wrapper. A context engineering layer — model-agnostic, infrastructure-grade, built so the intelligence an organization accumulates travels across every AI tool and survives model changes, vendor changes, and capability shifts.
Brian Handrigan
Thirty-two years in technology. Six issued patents. Seven companies founded — four of them in data and data-adjacent domains: grāmatr, NEXT90, Advocado, and Traaqr. Forbes Agency Council member. MRC Working Group contributor on measurement standards, alongside ABC, Nielsen, Disney, FOX, and Oracle.
The career arc is a straight line. From discovering that TV ads drove web traffic spikes in 2000 with no way to measure the connection, to co-founding Recursive Labs and inventing co-browsing technology (US9256691B2, US10067729B2, US10067730B2), to co-founding Advocado and building cross-media attribution technology (US12045853B2 for conversion tracking, US11790394B2 for call routing, US11544748B2 for advertising coordination), to founding grāmatr.
The thread is the same problem at every stage: data exists in silos, context gets lost between systems, and the people doing the work spend too much time re-explaining what should already be known. grāmatr is the latest answer to that problem — built in production, measured in production, and now available to every organization running AI at scale.
grāmatr + NEXT90
grāmatr and NEXT90 are independent entities. Brian co-founded NEXT90 with Randy Cairns in 2022 as an insights and data engine — ClickHouse, DBT, Superset, cross-media attribution.
NEXT90 is grāmatr's first enterprise customer. The entire NEXT90 platform was built using the grāmatr intelligence pipeline before grāmatr had a name. Every routing decision, every classification, every quality gate on that platform ran through the same architecture being sold today. Not a controlled demo environment. A live enterprise platform, in production, under real operational load.
grāmatr and NEXT90 are not parent and subsidiary. They are affiliated companies with a shared founder and a real integration. NEXT90 is proof that grāmatr works — not in a pitch deck, not on a benchmark, but powering a live enterprise platform. That is the dogfood story made specific.
Standing on shoulders
grāmatr's development was informed by the broader open-source AI ecosystem, including Fabric and PAI (Personal AI Infrastructure) by Daniel Miessler, both released under the MIT License. These projects validated patterns grāmatr was already building toward.
Credit where it's due. Open source makes this possible. grāmatr builds on what others share, and is transparent about its influences.
Start the conversation.
Want to understand the intelligence pipeline? See how it works. Want the technical proof? Read the science.
Ready to start the conversation?