A deliberately small product surface that is easier for both people and agents to operate cleanly.
Features
A compact system for building, querying, and exposing local memory.
Clio is intentionally small at the surface. It gives you a mapping agent, a local memory layer, a command interface, and a local MCP server that lets assistants work against that memory directly.
Build structured local memory from your own material instead of relying on flat retrieval alone.
Run clio mcp to make the memory available through tool use, not repeated copy and paste.
Core Capabilities
What Clio can do today.
The system is built around a small number of strong capabilities. Each one is meant to support real work on top of your own material, not abstract experimentation.
Clio Agent reads source material and turns it into a local structure of concepts, facts, and linked situations.
The resulting memory lives inside .clio/ and can remain fully local when paired with a local model backend.
You can ask questions against the memory through the CLI instead of depending on a flat retrieval layer alone.
Clio can run as a local MCP server so assistants can use the memory through tools rather than through pasted context alone.
As source material changes, the memory can be synced again so the structure remains usable over time.
Commands And Workflow
A small command model, intentionally close to Git.
Clio is designed to be easy to operate and easy for agents to call. The command surface is deliberately small, with each command doing one clear job.
clio init
Create the hidden .clio/ layer inside the current folder.
clio sync
Process new or changed files and update the local memory structure.
clio watch
Keep watching the folder and update the memory as material changes.
clio query "..."
Ask a question against the structured memory.
clio status
Inspect the current state of the system, including counts and summary information.
clio mcp
Start the local MCP server so assistants can access Clio through tools.
MCP And Agent Access
Clio is built to be usable through assistants, not just beside them.
The most important access path for many users is not a visual interface — it is the ability to expose a local memory system to an assistant in a way that is structured, inspectable, and native to tool use.
Clio runs as a local MCP server over stdio, making memory available to clients like Claude Desktop and Cursor without a cloud relay.
Assistants can sync, query, and inspect memory through tools tied to the local machine instead of relying on pasted context.
clio mcp starts the server. Point your MCP client at it and the memory becomes available immediately.
Local Memory Assets
What Clio leaves behind is a usable local system of files.
The output of Clio Agent is not an opaque remote index. It is a local memory layer with concrete files that can be inspected, maintained, and queried over time.
.clio/
graph/
facts/
interactions/
.manifest.json
config.yaml
graph/
Concept nodes and the structure that organizes the memory.
facts/
Fact nodes attached to the concepts that contextualize them.
interactions/
Interaction history tied back to the parts of memory a query touched.
.manifest.json
Tracks processed files and drives change detection on sync.
config.yaml
Backend and runtime settings for the local system.