Two query standards, one engine
Run openCypher, the language Neo4j made popular, and ISO GQL, the official international graph standard, on the same executor.
‹ qOS
Agent Graph· Traverse
A graph database agents can reason in.
Traverse is an in-memory graph database written from scratch in Rust that consistently beats the incumbents across reads, traversals, aggregates, and writes, often by several times and by 100x or more on deep analytics. It speaks the languages your tools already know, ships forty graph-science algorithms callable from a single query, and includes a built-in interface for agents to explore graphs directly. The same engine runs as a server, embedded in your app, or entirely inside a browser tab.
7x faster than Neo4j, our benchmarks
375K/s reads at peak concurrency
40 graph-science algorithms built in
Why Traverse
Graph databases made you choose speed or openness.
The open standard is built on the JVM and often slows to a crawl on deep traversals and analytics. The fast engines lock you into their own tooling. None were designed for an AI agent to query directly, and most sell graph data science as a separate, heavyweight product.
Traverse is a clean-room rebuild that is fast on every workload, speaks the open query standards, runs anywhere from a server to a browser tab, and exposes a native agent interface out of the box.
Capabilities
What it can do.
Forty graph-science algorithms built in
PageRank, community detection, centrality, shortest paths, similarity, link prediction, and embeddings, all callable from one line of a query and streamable or written back as properties.
A built-in agent interface
A native Model Context Protocol server ships inside the database, so an agent can query, inspect the schema, search nodes, and find paths directly, with no glue code or extra service.
Runs in a browser tab
The whole engine compiles to WebAssembly and runs client-side, multi-threaded, with data persisted in the browser and no server required.
Embed it in five languages
Drop the engine in-process into Python, Java, Node.js, Go, and .NET at function-call speed, with no server to deploy.
Drop-in wire compatibility
Implements the Neo4j Bolt protocol, so existing drivers, tools, and dashboards connect unchanged, with TLS and role-based access.
Single-file, crash-safe storage
Each database is one portable file, the SQLite model for graphs, with a write-ahead log and per-page checksums.
A visual Studio
A built-in web console for visualization, query editing, and graph data science.
Built different
Why it is not just another graph.
-
Faster on essentially every workload
Against both incumbents on the standard benchmark, Traverse wins the large majority of queries and is 100x or more on deep aggregates and multi-hop traversals, while using a fraction of the memory.
-
Lock-free reads that never block writers
Up to sixty-four readers walk the graph with zero locks while writes proceed, using far less memory than the copy-on-write approach rivals use. Databases hot-swap in and out of memory live.
-
A cache-engineered hot path
A compact layout, binary property encoding, and multi-core parallelism keep the hot path inside the CPU's fastest cache, the same techniques the fastest analytical databases use.
-
Agent-first and standards-complete
The agent interface and graph science are first-class built-ins, not bolt-ons, and Traverse is among the first to claim full conformance on the audited GQL feature set.
-
One engine, everywhere
The identical engine powers the server, the embedded libraries, and the in-browser build, with no feature forks between them.
By the numbers
By the numbers.
Measured on a single desktop against current Neo4j and Memgraph, reproducible from our benchmark suite.
7xgeomean vs Neo4j
100x+on deep analytics
Sub-mspoint read and write latency
10 to 20xless memory on scans
Agent-first
The reasoning substrate for an agent.
Entities, relationships, and multi-hop paths are exactly what an agent needs to answer how this connects to that, or what the blast radius of a change is. Traverse ships a Model Context Protocol server inside the database itself, so an agent connects over a single endpoint and gets a typed toolset, with no integration and no separate service. It can run any query, or call purpose-built, bounded tools that return structured graph results.
- Six built-in agent tools: query, read schema, list databases, search nodes, find paths, walk neighbors.
- Read-only views and per-query timeouts give an agent a safe, bounded window.
- The in-browser build even ships an assistant that knows the full algorithm catalog.
Run it yourself
Traverse is also released as a standalone product by Truespar, our research arm. Run it on your own infrastructure, the getting-started guide takes it from there.
Put Agent Graph to work.
See HQ running in your own Slack or Teams, on the operating system we built for agents.