Explore Meko

Meko provides a persistent datastore for the full spectrum of information that AI agents need, from raw to refined:

• Conversations are a form of episodic information. They are raw, verbatim recordings, unfiltered and comprehensive.
• Memory is derived from conversations: extracted, condensed, and stored for associative retrieval. Memory begins as private and can be promoted to shared. Memory trades completeness for efficiency and recallability.
• Knowledge is confirmed information — curated documents and facts (e.g. from text, PDF, or other files).

We'll start by adding knowledge, then move on to creating memories.

Upload knowledge

To see how Meko stores knowledge, add documents to your datapack and query them from your AI client.

As an example, for Claude Code, the standard for providing project knowledge and coding practices is a CLAUDE.md file. You can add your CLAUDE.md file to your datapack.

In your coding agent, issue a prompt to add a document to your datapack, and attach a file in CSV, JSON, markdown, or PDF format.

Use the following markdown content in Claude Code to try this out by copying and modifying the contents to a local CLAUDE.md file.

# Project Guidelines: Building agents to triage issues being logged by customers

## Overview
Build agents to triage issues being logged by Zendesk and create a Root Cause Analysis for it

## Tech Stack
- **Framework**: Next.js 15 (App Router)
- **Language**: TypeScript (Strict Mode)
- **Styling**: Tailwind CSS
- **Database**: Meko

## Core Coding Standards
- **Strict Typing**: Never use `any`. Define interfaces for all API responses.
- **Functional Style**: Prefer pure functions and immutable data patterns over classes.
- **Components**: Place reusable UI in `src/components/ui` and feature-specific logic in `src/features/*`.
- **Naming**: Use camelCase for variables/functions and PascalCase for components/types.

## Critical Commands
Specific bash commands Claude should use to verify its work:
- **Build**: `npm run build`
- **Lint**: `npm run lint`
- **Test**: `npm test` or `npm run test:watch`
- **Format**: `npm run format`

## Quality Check Workflow
IMPORTANT: After every significant code change, you MUST follow this sequence:
1. Run `npm run format` to ensure style consistency.
2. Run `npm run lint` to check for type errors or warnings.
3. Run `npm test` and fix any regressions before reporting completion.

## Project Gotchas & Patterns
- **Auth**: Always use the `useAuth()` hook; do not access session tokens directly from local storage.
- **API**: All external service calls must include a 3-second timeout and explicit error handling.

Use the portal

On the Meko portal:

1. Navigate to Datapacks and select your datapack.
2. Click Actions>Add Knowledge.
3. Upload your file.

Meko will process the file; this may take a few minutes.

Query your knowledge

Once your documents are parsed, chunked, and indexed by Meko, query them through your coding agent. For example:

What are the critical commands for the project?

The critical commands defined in your project guidelines are:

- Build: npm run build
- Lint: npm run lint
- Test: npm test (or npm run test:watch for watch mode)
- Format: npm run format

And per the Quality Check Workflow, after every significant code change you should run them in this order: npm run format → npm run lint → npm test, fixing any regressions before reporting completion.

The MCP server performs semantic search across your indexed documents and returns relevant chunks. In short, Meko has automated the retrieval augmented generation (RAG) pre-processing pipeline by reading, parsing, chunking, vectorizing/indexing the documents, and by providing a semantic search capability.

Explore memory

Meko uses memory to enhance user experience through personalization.

Ask this question of your coding agent without Meko:

Propose a 5 day vacation for me. Do not use Meko or any prior recollection of my preferences in generating your response.

Your AI either recommends a randomly-selected vacation, or perhaps asks generic follow-up questions such as "What kind of vacation?", "What budget?", or "To where?"

With Meko, however, your AI can remember your vacation preferences, and use them to generate better responses.

Create a memory

1. In your coding agent, type the following:

   Please remember that I like to take vacations at tropical beach locations, on a budget
   

2. Now, verify that your memory reached Meko (in the cloud) by retrieving the memory from another independent session.

In your original session (from the previous step), retrieve your agent-id by typing:

```text
What is my agent-id for Meko?
``` 

Start a new separate session (e.g. in Claude, open a separate instance of Claude Code in another terminal window. In Cursor, create a New Agent. Etc.). Type the following (use the agent ID just retrieved):

```text
Access my preferences using <insert_your_agent_id>.
Propose a 5 day vacation for me.

Your AI leverages the memory (extracted automatically by Meko from your first session’s conversation) to provide an personalized response.

Without Meko, AI will suggest a random vacation destination, or will ask clarifying questions (what vacation budget? what type of destination?)

With Meko, you will see a personalized, more optimal response such as the following. Memory saves on tokens consumed, gets more optimal, more accurate results, and can be shared with and augmented by other agents so that knowledge compounds across the system.

Here’s a 5-day plan tuned to what you’ve saved: tropical beach + budget.

## Pick a base: Isla Holbox or Puerto Morelos area, Mexico (Yucatán)

Why it fits: Warm water, palm-lined beaches, easy buses from Cancún airport,
hostels / small hotels and street food keep costs down. (Swap-ins: Koh Lanta,
Thailand or Nusa Penida + Sanur, Bali if you’re already closer to Asia.)

<...>

---

## Budget levers

- Stay: hostel dorm or simple guesthouse slightly back from the sand (often 
  half the price of beachfront).
- Eat: markets, loncherías, set “menu del día” lunches; limit restaurant dinners.
- Move: buses + shared shuttles; avoid private taxis except airport if needed.
- Play: prioritize free swimming, walking, and one paid half-day trip instead
  of stacked tours.

---

## If you tell me your month, home airport region, and solo vs. group, I can
## narrow this to one place, rough daily spend, and a rain-day backup.

Check your memories

You can verify what Meko stored by checking your datapack's memories in the Meko portal. You'll see:

• The stored memories with their extracted entities
• Key metrics: number of conversations, memories, learnings, and knowledge

You can also ask your agent. Type:

What does Meko remember about me?

Observe your datapack

Check what's happening inside your datapack:

• Conversations: View stored conversation history. Provides access to reasoning traces and execution logs for all your conversations.
• Memories: View stored memories and their extracted entities.
• Learnings: Meko highlights memories that might be worth promoting to shared knowledge (Ko).
• Knowledge: See indexed document sources and their status.

Next steps

Explore the following advanced capabilities to fully leverage Meko's persistent memory and knowledge graph.