2 min read

Why we invested in Airweave

Dave Yen
Dave Yen
//
Cover Image for Why we invested in Airweave

The founders of Airweave are solving one of the most urgent infra challenges in AI that's not well yet understood - but will be soon.

Fun fact: they also won the YC X25 Product Showcase, and got to sit down and chat w/ Sam Altman (pictured above).

Here's what they're building and why it matters:

Reliable tool calling (whether MCP, A2A, or direct API calls) where agents take real actions in apps still breaks down.

To make it work, every agent needs a backend that looks like… a search engine.

That's where Airweave comes in, or another way to think about it: "RAG for tools."

One of my favorite AI investors + builders, Yoko Li (Partner at a16z), recently published an MCP market map and deep dive.

She called out that this "RAG for tools" layer was still a missing piece - in her words:

"A common question we've heard from the developer community is how to think about tool selection when building an MCP client: Does everyone need to implement their own RAG for tools, or is there a layer waiting to be standardized?"

Yoko Li, a16z

MCP Market Map Source: a16z - A deep dive into MCP and the future of AI tooling

Why now?

Just last week, OpenAI launched support for MCP remote servers inside ChatGPT.

Although this RAG for tools layer is abstracted away for developers, if you read the docs closely you'll see: "Your MCP server should resemble a search engine."

Followed by instructions for how to set up semantic search for tool calling.

This is a strong signal, and exactly the infra Airweave ships—already live, open-source, used by 2.5k+ devs, with 100+ pre-built connectors.

What's next?

We think agent search is just the start for Airweave, and they're uniquely positioned to solve another emerging issue:

The agent security layer for RBAC, fine-grained permissions, and guardrails against prompt injection

The right place to enforce role-based access and sanitize inputs is at the sync layer, before the data ever reaches the agent.

In fact, Vercel just wrote a great post about the agent prompt injection issue today:

In summary

  1. Agents won’t go mainstream without reliable and secure tool calling.

  2. Airweave is building the layer every serious AI dev will need.

  3. It's putting a new category on the map of the MCP + agent tools ecosystem.

  4. And it could own the search and security infra that sits between data and the agent.

We’re proud to back them at Orange Collective.

More here → Read our investment memo on Airweave


© 2025. Orange Collecive LLC is an independent group of Y Combinator alumni and is not affiliated, associated, authorized, endorsed by, or officially connected with Y Combinator.