GraphForge vs. FalkorDB
How do matrix-based graph operations inside Redis compare with an embedded analytical engine?
FalkorDB stores adjacency as sparse matrices and evaluates Cypher with linear algebra (GraphBLAS), running as a Redis module. Its 6.0 release replaced the C engine with one written in Rust. It targets low-latency graph queries and GraphRAG, with HNSW vector search and ten built-in algo.* procedures. GraphForge also has a Rust core, but runs inside your process instead of Redis, persists to Parquet, and ships 94 algorithms called as API verbs.
On this page
Should I choose FalkorDB or GraphForge?
FalkorDB to serve graph queries from Redis; GraphForge to analyze a graph inside your code with a broad algorithm library.
Choose GraphForge when
- You want a graph library inside Python, Node.js, or Rust, not a Redis module to deploy.
- You need a broad algorithm catalog, including Louvain and Leiden, in the current release.
- Your work needs checkpoints, provenance, and recorded competing hypotheses.
Choose FalkorDB when
- You already operate Redis and want graph queries served next to it from any Redis client.
- Your GraphRAG pipeline needs approximate (HNSW) vector search close to the graph.
- Many readers query one graph at once and writes can be serialized per graph.
What can each one do?
19 of 26 capabilities differ. For 10, FalkorDB's docs did not confirm support, so they are marked Not established rather than No.
Each capability has a status and a source. “Not established” means we could not confirm the feature from FalkorDB's primary documentation. It does not mean FalkorDB lacks it.
| Capability | GraphForge | FalkorDB |
|---|---|---|
| Storage | ||
| Property graph model | Shipped Labeled property graph with typed properties. 1 | Yes Labelled property graph following the openCypher property graph model. 18 |
| Document model | Not offered No document-store model; properties accept nested, heterogeneous values. 1 | Not established Maps and lists are property values; no separate document model documented. |
| Relational tables | Not offered Not a relational database; results arrive as Arrow tables. 1 | Not established No relational/SQL model documented; SQL import is a migration path only. |
| Key-value access | Not offered No key-value API; data is a property graph. 1 | Partial Hosts on Redis, so Redis keyspace coexists; each graph is one Redis key. 23 |
| Columnar storage | Shipped Durable projects store graph data as Arrow and Parquet. 4 | Not established No columnar storage documented. |
| Vector search | Shipped Exact cosine retrieval over embeddings you supply; no ANN index. 5 | Yes HNSW vector index, cosine or Euclidean, 1-4096 dimensions. 21 |
| Full-text search | Shipped Tantivy-backed BM25; hybrid text and vector retrieval through find(). 5 | Yes Full-text indexes backed by RediSearch (8.6.1 in 6.0). 17 |
| Geospatial | Not offered No spatial types or indexes are documented. | Partial Point type (latitude/longitude) with distance() in metres; no geometry types documented. 27 |
| Execution | ||
| Transactional workloads (OLTP) | Partial Local analytical scope; durable atomic writes with queued and optimistic writer modes. 6 | Yes Low-latency per-query reads/writes; writes serialized per graph. 22 |
| Analytical workloads (OLAP) | Shipped Analytical verbs and Cypher over a local columnar project. 7 | Partial Matrix-based algorithm procedures run inside the server; no separate analytics engine. 19 |
| Graph traversal | Shipped Cypher MATCH patterns plus BFS, DFS, Dijkstra, A*, and reachability. 8 | Yes Cypher pattern matching executed via linear algebra over sparse matrices. 18 |
| Variable-length paths | Shipped Bounded and unbounded patterns and shortestPath(). 8 | Yes Variable-length patterns like ()-[*]->(), plus shortestPath() and allShortestPaths(). 27 |
| Graph algorithms | Shipped 94 native algorithms behind analysis verbs, not Cypher CALL procedures. 37 | Yes Built-in algo.* procedures (PageRank, WCC, BFS, betweenness, max flow, etc.). 1920 |
| Incremental computation | Not offered Algorithms recompute on request; no incremental mode is documented. 7 | Not established Algorithms recompute on call; no incremental analytics documented (CCH index self-maintains but unreleased). |
| Deployment | ||
| Embedded / in-process | Shipped In-process library; no server, daemon, or connection string. 9 | Partial FalkorDBLite (Python/TypeScript) spawns bundled Redis+module child process; aimed at dev/CI. 24 |
| Server mode | Planned Optional remote peer extension planned after 0.6.0. 11 | Yes Redis module serving GRAPH.* commands; Docker images bundle FalkorDB Browser. 17 |
| WebAssembly | Not planned A browser WebAssembly engine was declined in ADR 0027. 12 | Not established No WebAssembly build documented. |
| Distributed cluster | Not planned Distributed clusters and HA replication are out of scope. 9 | Partial Redis Cluster shards graphs across masters, but each graph lives wholly on one shard. 23 |
| Managed cloud | Not planned No hosted compute. The graphforge.sh Hub distributes repositories; execution stays local. 13 | Yes FalkorDB Cloud managed service. 24 |
| Investigation | ||
| Branching | Experimental Research versions, branches, slices, proposals, and forks ship in 0.6.0 release candidates. 14 | Not established No graph branching documented; GRAPH.COPY duplicates a whole graph. |
| Snapshots / time travel | Shipped Named checkpoints you can diff and revert, plus transaction-time snapshots. 3 | Not established No point-in-time query documented. |
| Immutable history | Shipped Knowledge ledger records immutable assertions and evidence lineage. 16 | Not established No append-only history documented. |
| Provenance | Shipped Provenance and knowledge ledger with evidence lineage. 16 | Not established No engine-level provenance; GraphRAG SDK tracks MENTIONED_IN source-chunk edges at app level. |
| Competing hypotheses | Shipped Immutable hypothesis groups with membership and selection events. 3 | Not established |
| Operations | ||
| ACID transactions | Partial Acknowledged-durable atomic publication; docs do not claim general ACID or serializable isolation. 6 | Partial Each query atomic, writes serialized; no native multi-query transactions (Redis MULTI/EXEC workaround). 22 |
| Replication | Not planned HA replication is out of scope. 9 | Yes Asynchronous single-primary Redis replication; 6.0 replicates effects only. 1723 |
Showing 26 of 26 capabilities.
How is each one built?
FalkorDB: property graph on sparse matrices. GraphForge: an embedded property graph library in Rust, persisted as Parquet.
GraphForge
GraphForge is a Rust library that runs inside your process: no daemon, server, or connection string. Projects persist to a local directory as Arrow and Parquet data plus JSON metadata, and every query and analysis verb returns an Apache Arrow table. Algorithms run in the same native engine as Cypher and are called as API verbs (rank, cluster, paths, analyze, similar), not as Cypher procedures.
FalkorDB
FalkorDB runs inside Redis as the graph module. Each graph is one Redis key held as sparse adjacency matrices, queries compile to matrix operations, readers see a consistent snapshot, and writes to the same graph run one at a time. In 6.0, vector and full-text indexes run on RediSearch.
Both engines are built for fast graph computation, but they are deployed differently. FalkorDB is a service shared through Redis. GraphForge is a library whose project directory you can copy, version, and open anywhere your code runs.
What the documentation says about FalkorDB
- FalkorDB represents graph adjacency as sparse matrices and evaluates queries using linear algebra (GraphBLAS).18
- FalkorDB 6.0 (2026-09-29) replaced the 4.x C engine with a new engine written in Rust while keeping the same GRAPH.* commands and Cypher.17
- FalkorDB runs as a Redis module named `graph` and can be called from any Redis client.17
- Vector indexes use HNSW with cosine or Euclidean similarity on vectors of 1 to 4096 dimensions.21
- In 6.0, full-text and vector indexes run on RediSearch 8.6.1, and replicas apply write effects rather than re-running queries.17
- Readers see a consistent snapshot while write queries to the same graph execute one at a time in FIFO order.22
- In a Redis Cluster deployment each graph is a single key, so a graph is placed entirely on the shard owning its hash slot.23
- FalkorDBLite bundles Redis plus the FalkorDB module into an embedded runtime for Python (pip) and TypeScript (npm).24
- The 6.x engine registers ten algo.* procedures: pageRank, WCC, betweenness, BFS, labelPropagation, MSF, SPpaths, SSpaths, HarmonicCentrality and maxFlow.20
Which one is faster?
Not measured yet. No head-to-head run has been published, so this page shows no performance chart.
Charts appear only after a reproducible run exists: pinned versions, disclosed hardware, generated data, raw results, and scripts.
No comparative results yet
The suites below apply to GraphForge and FalkorDB. Each will be reported separately per benchmark class, never merged into one leaderboard.
Ingestion and writes
| Suite | Workloads | Metrics | Status |
|---|---|---|---|
| INGEST | CSV, JSONL, Parquet, graph edge lists | Records/s, MB/s, total elapsed | Not run |
Query, analytics, and system
| Suite | Workloads | Metrics | Status |
|---|---|---|---|
| LOOKUP | ID lookup, indexed property filter | p50/p95/p99 latency | Not run |
| TRAVERSE | BFS, k-hop, variable-length path | Latency, visited edges/s | Not run |
| JOIN | Multi-relation pattern matching | Latency, rows/s | Not run |
| ALGO | PageRank, WCC, SCC, Louvain, SSSP | Time, memory, correctness | Not run |
| HYBRID | Vector + graph + metadata filters | Recall@k, latency, throughput | Not run |
| EMBED | Startup, load, reopen, binary size | ms, MiB, disk footprint | Not run |
| CONCUR | Mixed readers and writers | Throughput, tail latency | Not run |
Which graph algorithms ship?
GraphForge: 94 native algorithms. FalkorDB: 10 named in its docs, delivered as bundled.
Algorithms are listed by name, as documented. Counts of extension modules or third-party libraries are not presented as native coverage.
GraphForge
94 Rust algorithms run in the engine behind five analysis verbs (rank, cluster, paths, analyze, similar) and return Arrow tables.37
A selection of the catalog:
- PageRank
- ArticleRank
- Betweenness (Brandes)
- Closeness
- Harmonic closeness
- Eigenvector
- HITS
- CELF influence maximization
- Triangle count
- K-core decomposition
- Louvain
- Leiden
- Label propagation
- Infomap
- Walktrap
- Girvan-Newman
- Weakly connected components
- Strongly connected components
- Dijkstra
- A*
- Bellman-Ford
- Yen's k-shortest paths
- Max flow / min cut
- Steiner tree
- Minimum spanning tree
- Topological sort
- Node2Vec
- GraphSAGE
- Fast random projection
- Node similarity (Jaccard)
FalkorDB
Algorithms ship inside the engine as Cypher procedures invoked with CALL algo.<name>() and implemented with matrix-based computation.192017
Delivered as: bundled
- BFS (algo.BFS)
- Shortest weighted paths (algo.SPpaths)
- Single-source paths (algo.SSpaths)
- PageRank (algo.pageRank)
- Weakly Connected Components (algo.WCC)
- Betweenness Centrality (algo.betweenness)
- Harmonic Centrality (algo.HarmonicCentrality)
- Community Detection Label Propagation (algo.labelPropagation)
- Minimum Spanning Forest (algo.MSF)
- Maximum Flow (algo.maxFlow)
Where can each one run?
GraphForge runs in-process only. FalkorDB: embedded partial, server yes, cluster partial.
| Mode | GraphForge | FalkorDB |
|---|---|---|
| Embedded / in-process | Shipped In-process library; no server, daemon, or connection string. 9 | Partial FalkorDBLite (Python/TypeScript) spawns bundled Redis+module child process; aimed at dev/CI. 24 |
| Server mode | Planned Optional remote peer extension planned after 0.6.0. 11 | Yes Redis module serving GRAPH.* commands; Docker images bundle FalkorDB Browser. 17 |
| WebAssembly | Not planned A browser WebAssembly engine was declined in ADR 0027. 12 | Not established No WebAssembly build documented. |
| Distributed cluster | Not planned Distributed clusters and HA replication are out of scope. 9 | Partial Redis Cluster shards graphs across masters, but each graph lives wholly on one shard. 23 |
| Managed cloud | Not planned No hosted compute. The graphforge.sh Hub distributes repositories; execution stays local. 13 | Yes FalkorDB Cloud managed service. 24 |
| Replication | Not planned HA replication is out of scope. 9 | Yes Asynchronous single-primary Redis replication; 6.0 replicates effects only. 1723 |
What will my team write?
GraphForge: openCypher from Python, Node.js, Rust. FalkorDB: openCypher (with FalkorDB extensions).
FalkorDB is called through Redis GRAPH.* commands or one of seven client SDKs, with openCypher that omits some features such as label expressions and the regex operator. GraphForge passes the full openCypher 2024.3 TCK and exposes algorithms and search as typed API verbs.
- Query languages
- GraphForge openCypher
- FalkorDB openCypher (with FalkorDB extensions)
- SDKs and bindings
- GraphForge Python, Node.js, Rust
- FalkorDB Python, TypeScript/Node.js, Java, Rust, Go, PHP, C#
- Written in
- GraphForge Rust
- FalkorDB Rust (6.x engine); C (4.x engine)
- Surfaces
- GraphForge Python, Node.js, Rust crates, CLI, VS Code extension, Notebooks, Agent skills
What are the trade-offs?
FalkorDB has ten algorithms and SSPL licensing. GraphForge's vector search is exact, with no ANN index or server.
A fair assessment cuts both ways.
GraphForge
- Vector search is exact, without an HNSW or other ANN index.
- No server or Redis-compatible protocol; clients must run in-process.
- Pre-1.0, with no backward-compatibility guarantee yet.
FalkorDB
- FalkorDB does not natively support multi-query transactions; the documented workaround is Redis MULTI/EXEC, which serializes commands across all graphs.22
- The core is licensed under SSPLv1, which requires releasing service source code when FalkorDB is offered as a service unless a commercial licence is obtained.25
- A*, Customizable Contraction Hierarchies and Leiden/Louvain were built for the C engine after 4.20 but have not shipped in any release; they are being ported to 6.0.17
- 4.x and 6.0 nodes cannot replicate to each other, so every node in a replication group must be upgraded together.17
- FalkorDB's Cypher lacks label expressions, the regex operator, temporal arithmetic functions and hex/octal literals.26
How would I migrate?
Load FalkorDB data through Arrow, pandas, or Polars. Standard openCypher carries over; procedure calls become verb calls.
Into GraphForge
- GraphForge 0.5.x loads data through Arrow, pandas, and Polars bulk calls or Cypher scripts; it has no native CSV loader.15
- Algorithms are API verbs (rank, cluster, paths, analyze, similar), not Cypher CALL procedures, so procedure calls become verb calls.7
- GraphForge passes the full openCypher 2024.3 TCK, so standard openCypher queries are the portable part of a migration.10
Moving from FalkorDB
- The official Neo4j migration path exports nodes, edges, constraints and indexes to CSV with a Python extractor, then loads the CSV into FalkorDB.28
- Queries carry over from 4.x to 6.0 unchanged, but replication sets cannot mix engine versions.17
- Cypher written for other engines may need rewrites where FalkorDB omits openCypher features such as label expressions and regex matching.26
What else do people ask about FalkorDB?
Is GraphForge a FalkorDB alternative for GraphRAG? For local experiments, yes: Cypher, BM25, exact vectors, and algorithms in one process. FalkorDB fits ANN at scale.
Is GraphForge a FalkorDB alternative for GraphRAG?
For local GraphRAG experiments, it can be: GraphForge combines Cypher, BM25 full-text, exact vector retrieval, and graph algorithms in one process. For approximate vector search at scale or a shared service, FalkorDB fits better.
Do FalkorDB queries run on GraphForge?
Standard openCypher generally does. FalkorDB-specific commands (GRAPH.*) and algo.* procedure calls do not, because GraphForge exposes algorithms as API verbs rather than procedures.
Which algorithms does FalkorDB ship?
The 6.x engine registers ten procedures: PageRank, WCC, betweenness, BFS, label propagation, minimum spanning forest, two path searches, harmonic centrality, and max flow. Leiden, Louvain, and A* exist for the C engine but had not shipped in a release when we checked.
How was this researched?
From 28 primary sources, accessed 2026-10-19, each cited where it is used.
Capabilities were checked against primary documentation, repositories, and release notes, accessed 2026-10-19. Each cell records two things: the capability’s state and the evidence behind it. Every GraphForge cell carries a release status. Shipped means stable 0.5.2; experimental means 0.6.0 release candidates only. GraphForge is pre-1.0 and makes no backward-compatibility guarantee yet.
- Yes documented natively
- Partial limited or caveated
- Extension official extension or module
- Paid edition paid or enterprise edition only
- Removed removed, with the version cited
- Not established not confirmed by our research; not a “no”
- Shipped in stable 0.5.2
- Partial shipped with documented limits
- Experimental in 0.6.0 release candidates only
- Planned on the roadmap
- Not offered outside the product's model
- Not planned explicitly out of scope
Benchmark classes
- Equivalent workloads. The same logical operation with the same correctness requirement on every engine.
- Native optimized workloads. Each engine's best supported implementation, including projections, indexes, extensions, and native algorithms.
- End-to-end application workloads. Time from raw data to a usable result: import, indexing, projection, computation, and output retrieval.
Sources · accessed 2026-10-19
- GraphForge repository README — GitHub
- GraphForge releases — GitHub
- GraphForge 0.5.0 changelog — GitHub
- Storage architecture — GraphForge Book
- Analytics integration: search and retrieval — GraphForge Guide
- ADR 0018: acknowledged durability and isolation — GitHub
- Algorithm verbs — GraphForge Book
- Cypher guide — GraphForge Guide
- openCypher compatibility — GraphForge Reference
- openCypher TCK compliance — GraphForge Reference
- Remote authority peer extension (#215) — GitHub
- ADR 0027: native runtime boundary — GitHub
- Release scope and roadmap — GraphForge Docs
- Keep research history — GraphForge Guide
- Graph construction — GraphForge Guide
- Knowledge public API v1 — GraphForge Book
- Release v6.0.0 (and v6.0.2 / v4.22.1 release list) — FalkorDB GitHub
- FalkorDB README — GitHub
- Algorithms overview — FalkorDB Docs
- graph/src/runtime/functions/algo_procedures.rs — FalkorDB GitHub
- Vector Indexing — FalkorDB Docs
- Atomicity and Concurrency Control in FalkorDB — FalkorDB Docs
- FalkorDB Cluster Deployment — FalkorDB Docs
- FalkorDBLite: Embedded Graph Database for Python and TypeScript — FalkorDB Docs
- FalkorDB License: Server Side Public License (SSPLv1) — FalkorDB Docs
- Cypher support — FalkorDB Docs
- Cypher functions — FalkorDB Docs
- Migrate from Neo4j to FalkorDB — FalkorDB Docs
Try GraphForge on your own graph.
Install the package, open a project, and run the same notebook we measure with.