Our first year in review

At the end of each year, I like to write a letter to my investors and friends of the company, sharing my retrospective on the year. In the spirit of building in public, I'm sharing an excerpt of that letter with you all.

The excerpt of the investor letter

Thank you all so much for your continued support. As we got started on Hypermode, we interviewed developers and tech leaders. We wanted to validate our view of the world and see if the problems we saw resonated with them. A few quotes have stuck with me from those early conversations:

We have no idea how many models are actually in prod, let alone what data and apps they're attached to.

Head of AI, Global Bank

We spent a ton of time nailing the typewriter animation to hide how slow our AI Search is.

Unicorn Startup Founder

Local demos are easy, going to production with AI is an entirely different story.

G2000 Principal Data Scientist

The hard problems in modern AI are speed, context, and coordination. Primarily, how can I execute AI functions fast enough for real-time apps? How can I give my models the right context to create the behaviors I expect? How can I track how each model, function, and experiment is performing?

We started Hypermode to solve those problems in a cohesive way.

Too much work is being put on developers to integrate a dozen spot solutions for basic AI. Today, Hypermode is a tightly integrated kit of tools and infrastructure. We made a runtime and framework called Modus (speed), a suite of knowledge graph tools (context), and a hosting/observability platform (coordination) - all to help developers overcome the hardest problems in modern AI development.

Milestones

Funding

In late 2023, Hypermode raised a little over $9M from some of the best execs, investors, and angels in the world, including:

• 10-time Midas List investor, Salil Deshpande
• Ex-Yahoo/Facebook/Box exec-turned-VC, Ethan Batraski
• Long-time friend and investor from Vercel, Reid Christian from CRV
• Angels from Github, Sanity, Confluent, Stripe, Atlassian, and more
• Top-tier funds like SignalFire, 8Bit, Abstract, et al.

Key hires

• Derek Briggs joined us to lead design after helping build beloved developer brands like Clerk, PlanetScale, and Tailwind.
• Will Lyon joined us to lead Developer Experience. We got to work together back at Neo4j. He's unique in his ability to take ultra-complex topics and make them intuitive and approachable.

Acquisitions

We acquired Dgraph Labs and began integrating it in early 2024. I've been following Dgraph since my time at Neo4j. The team built something truly unique: distributed graphs remain one of the most difficult problems in the category. As we consider building knowledge graphs for AGI and agentic workflows, Dgraph's scalability will ensure our users continue to thrive.

Major releases

• Dgraph

  • We recently announced the move to a fully open-source Dgraph, including many of its enterprise features. We want to make Dgraph the global standard for building knowledge graphs.
  • We also added vector search and storage to Dgraph, enabling developers to implement sophisticated natural language use cases. You can store and search dozens of separate vector indexes, making it easy to combine several embedding models with different optimizations in the same app.

• Modus

  • In October, we open-sourced Modus, our internal serverless framework and runtime for building model-native apps. We originally created Modus using WebAssembly to allow our SaaS service to scale context-rich inferences (functions), deploy those functions at the edge, and execute untrusted code securely. Early adopters are finding a ton of value in an ultra-portable format that lets them run AI functions against real-time workloads.

• Hypermode Platform

  • We launched the public preview of Hypermode at AI Engineer World's Fair last June
  • We expanded the platform to include shared model hosting with local access, making open-source models like Llama dramatically less costly
  • We built out Hypermode Model Tracing to include token counts, inputs, and inference duration

Hypermode in Production

Our first design partners went live with a variety of use cases, including:

• Hashnode uses Hypermode for document search and spam prevention.
• PickYourPacker has implemented it for product discovery and recommendations.
• AgroPatterns automates crop management using Hypermode in Latin America.
• Our next wave of customers and users included MIT, DevolverDigital, and Rootly.
• Existing Dgraph users in financial services, travel, and the federal government kicked off knowledge graph initiatives to ready their data for AI.

What's Ahead

Here's a glimpse of what the Hypermode team has planned for the year to come:

• Automate the construction and maintenance of knowledge graphs
• Expand Modus with use-case-specific APIs, agent kits, integrations, languages, and DX improvements
• Continue work on automated agentic tool construction
• Public preview of Serverless Dgraph
• Launch zero-config AI components for the front-end ecosystem (e.g., product search)
• Continue hiring the world's best in developer tools and AI

In our first year, we've made so much progress — but with the size of our ambition, miles to go.

Best,

KVG