The graph database that lives inside your Python application. Trace dependencies, retrieve connected knowledge for agents, and explore relationships with graph, text and vector search.
pip install okto-grafxPre-alpha · Python 3.11–3.13 · No database server
MATCH (n)-[r]->(m)
RETURN n, r, m01 / CAPABILITIES THAT CONNECT
From agent memory to dependency analysis: the building blocks to turn connected data into useful answers.
Connect entities with typed relationships. Follow paths, match patterns and aggregate connected data with a documented Cypher-style query language.
Requirements, decisions and changes live in separate records.
WHAT YOU CAN BUILDTrace a changed requirement to its tasks, tests and supporting decisions.
Combine BM25 text search, exact or approximate vector search, and weighted rank fusion. Use graph evidence to boost or filter candidates within one snapshot.
Keywords miss related ideas. Similarity alone misses relationships.
WHAT YOU CAN BUILDGive an agent relevant records with connected evidence for its answer.
Independent local processes can read and write the same store. Snapshot isolation, optimistic conflict checks, WAL-backed durability and verified recovery preserve the transaction contract.
Workers update data while other processes are reading it.
WHAT YOU CAN BUILDRead a consistent snapshot and handle write conflicts explicitly in your application.
Explore snapshot projections with BFS, weighted shortest paths, PageRank, connected components, k-core and label propagation.
A growing network is too connected to inspect record by record.
WHAT YOU CAN BUILDDiscover paths, rank connected entities and identify clusters in your data.
Bring in typed CSV and JSONL data. Exchange Arrow batches and NetworkX graphs, or use optional Pandas, Polars and local Parquet integrations.
Useful knowledge is scattered across files and analysis tools.
WHAT YOU CAN BUILDBuild connected datasets from local sources, then take results back to your Python tools.
Run inside your Python process with an in-memory graph or a persistent local directory. Use the Python API and CLI without operating a separate database server.
An embedded tool should not need a remote database to start.
WHAT YOU CAN BUILDBuild local agents, desktop tools and offline data workflows around an application-owned graph.
The qualified development source adds multiple labels, flexible properties, writing subqueries, exact decimals, temporal values and typed collections, alongside history and hybrid search. Native and installed Pulse qualifications are documented; release availability remains separate from published 0.0.5.
02 / CLOSE TO YOUR CODE
Start with an in-memory graph. Use a local directory when your data needs to persist. The engine runs in your application’s process.
from okto_grafx import connect with connect(":memory:") as db: with db.begin("write") as txn: txn.execute( "CREATE NODE TABLE Person(" "id INT64, name STRING, PRIMARY KEY(id))" ) txn.execute( "CREATE (:Person {id: $id, name: $name})", {"id": 1, "name": "Ada"}, ) result = db.execute("MATCH (p:Person) RETURN p.name") print(result.rows) # (("Ada",),)(("Ada",),) Connected03 / THE OKTO ECOSYSTEM
THE GRAPH BEHIND OKTO PULSEOkto Pulse uses Okto Grafx as its graph database. Requirements, decisions, tasks and evidence become connected project knowledge that agents can retrieve.
Discover Okto Pulse
Coordinate your AI agents.
Connect intent, delivery and knowledge.
The embedded graph foundation.
04 / START BUILDING
Bring a graph into your Python application.
pip install okto-grafxWindows · macOS · Linux
Validate API and storage compatibility before upgrading. See the release notes for available features.
Read release notesA FEW DETAILS
No. Grafx runs inside your Python process. Use an in-memory database for temporary data or a local directory for persistence.
Grafx supports a documented Cypher-style subset, including typed schemas, pattern matching, traversals and aggregates. It does not claim full Cypher compatibility. See the supported query language.
Yes. Grafx provides full-text search, exact and approximate vector search, and weighted reciprocal rank fusion for hybrid retrieval with bounded graph evidence. Explore hybrid search.
The source is available under Elastic License 2.0 with the SaaS/Branding Addendum. Read the license for the terms that apply to your use.