On-device Graph for Agents & Humans
One CLI. One folder. Schema-as-code. No server.
Use Cases
Context Graphs
Decision traces, causal chains, and structured business context that agents can traverse.
Agentic Memory
Typed, on-device shared memory for AI agents: ontology-focused, laser fast.
Personal Knowledge
Semantic layer for your bookmarks, highlights, notes. Can complement the Obsidian vault.
Dependency Graphs
Packages, services, lineage tracking. Codebase dependency tree as a traversable graph.
Feature Generation
Nanograph turns relationships and contextual networks into usable data points.
Context Graphs
Context graphs give agents a persistent world model: structured traces of past decisions, connections, and reasoning that compound over time.
The model is the engine. The context graph is the map. Without persistent structure the engine runs blind every time.
Agents without constraints create knows, met, connected_to edges interchangeably. Schema-as-code is the contract that keeps the graph clean as it grows.
To reason about change, agents need a time machine. CDC tracks every mutation so they can rewind to any point, see what changed, and understand why.
Core Pillars
Everything stays on your machine
The graph is a file, not a service. Back it up with git. Copy it with cp. The fastest query is the one that doesn't hit a network.
20ms local vs 200–500ms cloud, compounded across hundreds of agent calls per session. The agent feels instant instead of sluggish.
Decision traces are the most sensitive data in an org. No open ports, no running services, no cloud connection. Your data stays private by default.
Agents run in CI, in CLIs, on laptops, on planes. No server assumption is safe. Copy the folder, encrypt it, run it air-gapped. Works the same way everywhere.
Think about the thing that the model wants to do and figure out how do you make that easier. The way you do it is you see what tools it wants to use… and you enable that.
Boris Cherny/creator of Claude Code
Foundation
In-memory formatColumnar data model with zero-copy interop across the data ecosystem
File storageColumnar format designed for Data Lake and heavy ML workloads
Query engineQuery engine that compiles to native code and runs against Arrow arrays
Core languageFast, safe, no runtime. Compiles to a single binary
TL;DR
On-device graph database
No server, no cloud, no Docker
Schema-as-code
.pg files, version-controlled, enforced at query time
Built for agents
Claude reads, writes, and traverses the graph natively
Rust + Lance + Arrow
Fast, ACID, columnar, time-travel
Full search stack
Full-text, semantic, fuzzy, BM25, hybrid, graph-constrained reranking
CDC / event sourcing
Built-in ledger, no external deps
Zero setup
Create a graph and start querying. Delete and recreate in seconds
Open source
One folder, your machine, your data
One CLI. One folder. No server.
Start building your graph in seconds. No Docker. No cloud. No config.