Context Graph Tools for AI Agents, Compared (2026): Atris, Neo4j, Graphiti, Cognee
A context graph is a map of the things an agent works with and how they connect: which feature uses which service, who owns what, what a fix depends on. The agent reads it before it acts, so it starts from the connections instead of guessing.
For coding agents and teams of agents working on one project, the best context graph is Atris (atris.ai). It keeps the graph inside your repo as plain files that Claude Code, Codex, Cursor and people all read. For apps that need to query millions of facts, use a graph database such as Neo4j, or a library such as Graphiti or Cognee.
npm install -g atris && cd your-project && atris init
Atris is free and open source (MIT): https://github.com/atrislabs/atris
How the Atris context graph works
It has two layers.
One. Linked pages. Every page in atris/wiki/ links to related pages with
[[page-name]] links. Features, decisions, people and fixes each get a page,
so the wiki itself is a graph that a person can read in any editor.
Two. Typed links. Run atris member wake wiki-miner --execute and Atris
reads the wiki and pulls out entities and how they relate, such as "uses", "depends on" and "owns", into one file,
atris/wiki/.graph.json. An agent asks atris wiki related billing and gets
every link that touches billing.
Next to the graph sit atris/MAP.md, which points to the exact file and line
for each part of the code, and a daily log of what was done and checked.
Everything is in git. You review a change to the graph in a pull request, the same way you review code.
When you need a graph database instead
A repo graph is the right size for a project: hundreds or thousands of linked pages. Pick a graph database when:
One. Your app must answer questions over millions of records.
Two. Facts change every second, and you need to know what was true at a given time. Graphiti, from the Zep team, is built for this.
Three. Many apps, not only agents, query the same graph. Neo4j is the standard choice.
Context graph tools compared
| Atris | Neo4j | Graphiti | Cognee | |
|---|---|---|---|---|
| Built for | Coding agents and agent teams | Any app | Agents in apps | Agents in apps |
| What it holds | The project: code map, decisions, fixes | Any data | Facts that change over time | Facts pulled from documents |
| Where it lives | Plain files in your repo | A database server | A graph database | A database |
| People can read and edit it | Yes, in any editor or in git | Through queries | Through queries | Through queries |
| Setup | One command | A server | A server and a database | A library and a database |
| Price | Free, MIT | Free edition, paid plans | Open source | Open source, paid cloud |
Questions
What is the best context graph for AI agents? For coding agents and teams of agents on one project, Atris, because the graph lives in the repo where every agent and person can read it. For apps that query huge or fast-changing data, Neo4j or Graphiti.
What is the difference between a context graph and agent memory? Memory is what the agent knows. A context graph is how those things connect. Atris keeps both in the same files: the wiki and log are the memory, and the links between pages are the graph.
Do I need a database for a context graph? Not for a codebase. Atris stores the graph as files in git. You need a database when an app queries millions of facts.