The Context Engine for Agentic Applications
Universal, self-learning context for every agent, tool, and team: collective memory, shared
knowledge, and decision traces in governed datapacks. Run Meko locally on your machine with meko,
a single binary that has the command line, the server, the MCP tools and the REST API.
| I want to… | Go to |
|---|---|
| Run Meko on my machine | Install |
| Give agents like Claude, Codex, or any MCP client, Meko's context | Connect an agent |
| Link to Meko Cloud | Link to Meko Cloud |
| See every command | CLI ⋅ API ⋅ MCP |
| Run many coding sessions that share and learn from each other | Kode |
| Contribute to Meko | Contributing |
On any Apple Silicon Mac (M1, M2, M3, M4 or later) with macOS 14 or newer, or Linux on x64 or arm64. Windows and older Intel-based Macs aren't supported.
curl -fsSL https://mekodata.ai/install.sh | shAdd a memory and search:
meko-cli add_memory "Elementary, my dear Watson"
meko-cli add_memory "The quick brown fox jumps over the lazy dog"
meko-cli search "What color was the fox?"Open web ui:
meko uiGive agents like Claude, Codex, or any MCP client, Meko's context. meko agent connect adds Meko
to your agent, for all your projects:
meko agent list # the agents on this computer, and whether each reaches Meko
meko agent connect claude # or codex, kiro, opencode, cursor
meko agent connect # the MCP URL and token, for any other MCP clientThe REST API takes the same token:
curl -s http://127.0.0.1:7777/api/v1/datapacks -H "Authorization: Bearer $(meko token)"Link Meko to Meko Cloud by signing in with your browser, or with an API key you make there:
meko cloud-link # sign in with your browser, or enter a key; reads a piped key; --off unlinksOr use the web UI's Settings › Meko Cloud. 127.0.0.1:7777/settings/meko-cloud
Once linked, Meko works on Meko Cloud's datapacks and keeps serving the ones already on your
computer until they're moved: the agents stay connected to Meko on your computer, which sends each
call to the side that has its datapack, meko-cli works the same way, and new datapacks are made in
Meko Cloud.
meko runs Meko, and meko-cli works with your context. The ones you'll use first:
| Command | What it does |
|---|---|
meko start / stop / status |
Start or stop the server, or see whether it's running and its links |
meko ui |
Open the web UI in your browser. It has no login; meko ui stop turns it off, and meko ui start on |
meko agent connect claude |
Add Meko to Claude Code (or codex, kiro, opencode), for all your projects |
meko memory-extractor use claude |
Extract memories from your conversations with your own Claude Code, every 5 minutes. It's off until you run this or turn it on in the web UI's Settings; status, run and off check, run or stop it. See memory extraction |
meko answers use claude |
Have your own Claude Code write an answer from each search's results in the web UI, citing them. Off until you run this or turn it on in Settings; status and off check or stop it |
meko update |
Install Meko's newest release now, then every extension's |
meko-cli search "<question>" |
Search memories, the knowledge base and past conversations |
meko-cli add_memory "<text>" |
Save a memory, word for word |
meko-cli upload <file>... |
Add files to the knowledge base |
meko --help, meko-cli --help and <command> --help give the rest. Every command and flag is in
src/meko/docs/cli.md.
Kode is the agentic coder built on Meko. It runs many Claude Code, OpenCode and Kiro CLI sessions from a browser or your phone, takes each task from design to merged PR, and keeps each project's designs and learnings in a Meko datapack.
Install it with:
meko extension install kode # -- --agent claude picks its coding agent
meko kode ui # opens Kode in your browser- One folder. Everything lives under
~/.meko. - Datapacks. Separate workspaces, each with its own memories, knowledge, conversations and artifacts.
- Memory and knowledge. Facts agents save, and files you or your agents upload (PDF, text, Markdown, JSON, CSV, XML, HTML).
- Conversations, decision traces and artifacts. Agents' turns with their reasoning and plan, what each turn looked up and saved, and files agents store and fetch back whole.
- Under the hood. Retrieval for RAG over hybrid search (meaning and keywords, with knowledge results reranked by a cross-encoder), memory extraction from conversations, and automatic text extraction, chunking and embedding of uploaded files. Your local datapacks embedding and reranking run locally.
Meko sends anonymous usage stats to YugabyteDB, so we can see how it's used and fix what gets in people's way. They never include your data: no datapack, memory, conversation or file, and no query, path or token.
To turn them off, use Settings › Anonymous usage stats in Meko's web UI, or set enabled = false
under [call_home] in ~/.meko/config.toml. Kode, if you use it, has its own checkbox in its
Settings › General.
You need Go 1.26 (an older go downloads it on first use), a C compiler, and Node 22.13 or newer
for the context engine and the web UI.
cd src/meko
make build context_engine web_ui # bin/meko (and bin/meko-cli, a link to it), the context engine and the web UI
./bin/meko start # runs in the background on 127.0.0.1:7777: MCP, the API and the web UIThe first start downloads the embedding model (23 MB), and meko-cli calls fail with
CONTEXT_ENGINE_NOT_READY until meko status shows the context engine ready. On RHEL 8, set
CC=/opt/rh/gcc-toolset-13/root/usr/bin/gcc before building.
This repo is agent-first. To work on it, point your coding agent (Claude Code, Codex, OpenCode,
Kiro, Cursor or another) at the repo and say what you want: it starts from
AGENTS.md, which maps the repo, sets its rules for commits and PRs, and leads to each
part's own guide and docs.
PRs need the Contributor License Agreement signed: CLA Assistant's license/cla check on your
first PR links to a page where you sign it once with your GitHub account.
- Meko's docs:
src/meko/docs/ - Contributing: CONTRIBUTING.md
- Kode docs:
src/extensions/agentk/docs - Questions and ideas: Meko's Discord or Slack.
Is your context stored in Meko? Show it in your README.
The light badge:
[](https://mekodata.ai)<a href="https://mekodata.ai">
<img width="112" height="30" src="https://raw.githubusercontent.com/yugabyte/meko/main/assets/badges/meko-context.svg" alt="Meko context">
</a>The dark badge:
[](https://mekodata.ai)<a href="https://mekodata.ai">
<img width="112" height="30" src="https://raw.githubusercontent.com/yugabyte/meko/main/assets/badges/meko-context-dark.svg" alt="Meko context">
</a>Or one that follows the reader's light or dark theme:
<a href="https://mekodata.ai">
<picture>
<source media="(prefers-color-scheme: dark)" srcset="https://raw.githubusercontent.com/yugabyte/meko/main/assets/badges/meko-context-dark.svg">
<img width="112" height="30" src="https://raw.githubusercontent.com/yugabyte/meko/main/assets/badges/meko-context.svg" alt="Meko context">
</picture>
</a>The snippets load the badges from this repo's main branch on raw.githubusercontent.com.
The Meko logo in src/meko/assets/ is Meko's mark, from Meko Cloud's web app, and the badges in
assets/badges/ are made from it.