A context layer for every agent

Why this conversation

An agent can only do good work with the right context, and a team’s context is something its agents should be able to share. As a database company, YugabyteDB has started to work on the data problem for agents. Hari Krishna Sunder, who is on the engineering side there, showed the host what they are building: Meko, an agent context layer. The two spoke at the YugabyteDB booth in San Jose on Friday 23 October, the second day of AGNTCon + MCPCon North America. More from that Friday is in AGNTCon + MCPCon North America 2026: day two.

Memory, knowledge and sharing

The name is short for the two things Meko holds: the first two letters stand for memory, the last two for knowledge. It is built to help agents with the context problem and with the problem of sharing that context.

The host asked whether it is simply a RAG system. RAG is in there, Hari says, and graph too, already plugged together, so that people can spend their attention on their data instead of on the plumbing.

What makes it different, in his view, is the sharing. Context moves between a team’s agents, and across the team and the wider organization. Most tools have dealt with one piece of this, memory alone for example. Meko is meant to bring the whole picture into one ready-to-use product: conversations, audit trails, artifacts and knowledge.

Managed, your own cloud, or open source

The host asked whether it is a SaaS offering or something to install on his own premises. The answer is both. There is a cloud version that is fully managed, where you can simply sign up. There is also BYOC, bring your own cloud, which lets you deploy Meko on your own hardware, wherever you want, including air-gapped. And an open source version is planned for release soon, something you can run on a laptop and experiment with.

Onboarding begins with seeding the initial context, which Hari calls the most important part. You upload your files, PDFs or plain text for instance; text files are covered today. Meko takes care of chunking and embedding, and after a few minutes everything is ready. From then on the agents do the work, and the mechanics stay out of sight.

Which agents does it support? Practically all of them, he says. “It’s all MCP native, which means you just connect your agent through the MCP call, and they’ll all just automatically work like magic.”

How YugabyteDB uses it

Hari can speak best for software development, and there Meko is used across the whole cycle, from the PRDs through code review to the release notes. “So we have the entire context on what we are building in there.” Other teams rely on it as well: sales for account management, marketing for generating content. His point is a general one: everything needs context, and agents that have it do better than agents that do not.

Who it is for

The database targets the entire market, he says, while Meko is aimed at AI-native companies and at companies that are willing to use AI. It does not have to be exposed to the internet. A company can deploy it on its own hardware, run it air-gapped and use its own models. “So none of the data is going to leave your system, right?” Restrictions on where data may go are therefore no obstacle to using it.

To get started, you create a free account at cloud.mekodata.ai and try it out. Hari adds that the team would love to hear feedback on how people are using it, and the host said he was looking forward to trying it himself.

When it comes out

The episode is edited and in the release order. Like every episode it goes out on YouTube first, then on Apple Podcasts, Spotify and Amazon Music.