Meko FAQ

Meko stores “context,” enables sharing across agents and harnesses. The exact contents of “agent context” are continually evolving. Meko currently stores and appropriately compresses and indexes information to enable users to:

• Save AI conversations as both searchable text and graph memory, so that previously asked questions don’t have to be asked by your LLM again.
• Upload documents into a hosted knowledge base, so that you don’t have to write complex data pre-processing pipelines.
• Allow resuming your personal context over time or switching between LLMs and IDEs, so you can pick up where you left off if you run out of tokens, an LLM goes down for a while, or you simply go on vacation and then need to get going again.
• Share agent memory and knowledge base documents with other agents or human teammates selectively, so all agents that work together learn from each other.
• Store agent traces in a visual portal, so you can see what the agent thought or did during the conversation
• Gain access through a remote MCP endpoint, so that any MCP-capable agent (Claude Desktop, Claude Code, Cursor, Codex, and others) can use it.

In a few ways:

1. Remember shared context across agents, not just within one chat session.
2. Share knowledge of what agents have said and done across a team, so people and agents stop rediscovering the same things independently.
3. Reduce token spending by reusing what’s already been figured out, rather than reprocessing it from scratch every time.
4. Decision trace all the way from the human-issued prompt to the LLM’s chain of thought.

Built-in memory lives inside one vendor’s product and one agent’s context. Meko sits outside any single model or harness and is shared and selective: you decide what gets shared, with which agents, and with which teammates. That cross-agent,cross-vendor sharing (and the control over it) is the part that the built-in memory stores don’t do. Meko does not just persist memories, but the entire shared context (conversations, knowledge base, and decision traces), all encapsulated in a datapack that you can carry with you, regardless of the agentic SDK you use to build your agents, or if you just use a daily driver AI agent for your work life.

Meko uses mem0 under the hood for memory extraction, so you get that capability, but Meko is the hosted, multi-tenant platform around it: graph memory, a shared knowledge base, conversation persistence, selective sharing, and observability. All this is built on YugabyteDB and exposed through a single MCP endpoint. You’re not running and scaling the infrastructure yourself.

The data primitives are built around how agents actually work. Agentic conversations, memory, and the actions they took are all context rather than a sprawl of many specialized databases (vector, graph, relational) bolted together to make that happen. That’s the distinction: the constructs (datapacks, memory, knowledge, conversations, decision traces) are shaped for agent workflows and the way humans interact with them.

Meko reduces repeated work. When one agent or teammate has already done something (and spent tokens to do it), the next one can reuse it by asking Meko for the information instead of reprocessing the same context. Lookups can be offloaded to our context platform instead of traversing everything through a frontier model every time. If you have your agent query Meko before it needs to search, it significantly reduces spend. You don’t have to create your own RAG in order for it to work; Meko’s infrastructure works out of the box.

Meko is a remote MCP server, so you point your MCP client at the Meko endpoint and authenticate. We have connection steps for Claude Desktop, Claude Code, Cursor, and Codex; VS Code support is in progress. See the docs (Integrations → Connect to an AI agent) for the current setup per client.

New accounts start on a Basic (free) tier with generous limits. There is plenty of room to build and test. If you need more for a proof-of-concept or production use, reach out to the team, and we’ll move you up.

Today, the flow is human-driven for adding to knowledge. Fuller agent access (agents writing to knowledge directly) is on the roadmap.

Meko stores what you send it. We don’t anonymize or strip names automatically. If you’re working with sensitive data, scope what you put into a datapack accordingly, and note that you can run a local model with Meko as long as it’s in a harness or chat app that can connect to a remote MCP server (most can).

Meko is currently available only as a service running on our own infrastructure, providing you with an individual account hosted on mekodata.ai. You don’t have to worry about the process of hooking up your own pieces to use Meko, just connect your Meko account via our remote MCP server and go. Local and BYOC deployments are on our roadmap for enterprise customers, so reach out if that is something you want to explore.

Open source is planned. Currently, Meko is running as a cloud-hosted SaaS that runs in Yugabyte’s infrastructure, accessed by an MCP server endpoint used by an AI agent you connect to it.