Fallen-8
An in-memory graph database in C# (.NET 10), built for raw speed on heavy graph algorithms. It has no query language; queries are C#.
Queries are C#: small delegate fragments compiled at runtime, or precompiled stored queries. That is a deliberate choice for the era of code-generating agents: an agent emits a C# fragment, the engine compiles and runs it in-process at full speed, with no query-language layer in between.
Key features
Section titled “Key features”Graph modelA directed property graph; typed properties on vertices and edges, all mutation through a serialized transaction queue.
Delegates, not a query languageFilters and cost functions are runtime-compiled C# fragments; the defining design decision.
Path findingShortest and weighted paths with delegate filter and cost functions (BLS, Dijkstra).
SubgraphsExtract a pattern-matched subset as a standalone graph, recalculate it when the source changes, nest and persist it.
Graph analyticsPageRank, connected components, communities, degree centrality, triangle counting, with optional property write-back.
Stored queriesRegister a vetted, compiled query once and invoke it by name; no dynamic code at call time.
IndexesDictionary, range, fulltext, spatial R-Tree, and vector kNN, all as plugins.
Vector searchExact k-nearest-neighbour over float[] embeddings (cosine, dot product, L2).
Semantic traversalEmbeddings as element state; a code-free semantic block steers paths and subgraphs by similarity.
Semantic layerDocuments in, graph out: PDFs, Office files and markdown become Document, Chunk and deduplicated Entity vertices, found by fused semantic plus exact-token search and traversable like everything else.
Bulk import/exportStream whole graphs as newline-delimited JSON that round-trips exactly.
Live change feedCommitted mutations as Server-Sent Events, in commit order, with in-band resync.
Save gamesCheckpoints tracked by a registry that drives startup, on top of a write-ahead log.
NamespacesMany isolated graphs in one Fallen-8, addressable under /ns/{name}/.
ObservabilityOpt-in Prometheus and OTLP metrics and traces, a graph-shape snapshot, and health probes, plus a multi-tenant consumer stack.
REST APIA versioned HTTP surface with an OpenAPI document and an interactive Scalar reference.
PluginsIndices, algorithms, and services are all discovered plugins.
Plugin registrationAdd runtime algorithm and graph-function plugins by authoring C# source instead of uploading a DLL.
F8 StudioA browser UI to browse, query, visualize, and author the C# delegates, with a natural-language assist.
Standalone F8 StudioDeploy the browser UI as its own container, decoupled from the data plane and pointed at any Fallen-8 REST endpoint at container start.
BenchmarkMeasure raw edge-traversal throughput over whatever graph is loaded: generate a graph or point it at a sample, then run timed passes that follow every out-edge.
MCP serverA Model Context Protocol surface so AI agents call Fallen-8 as typed tools; read-only by default, tiered writes, three auth modes.
NL assist and fine-tuningDraft C# fragments from a sentence, and an offline pipeline to train, evaluate and publish your own model.
Use as a libraryReference the engine in-process via the Fallen-8 NuGet package, with no HTTP and no server to operate.
Capacity and performanceMeasured bytes per vertex and edge, write throughput, and what a save game costs the writer.
SecurityOptional all-or-nothing API key; dynamic code execution is always on, so set a key before exposing the service off-box.
New here? Run it, open F8 Studio, and load a sample graph.