RippleCI
An open-source graph-native predictive CI engine and CLI tool that computes multi-hop change blast radius using an AST property graph and graph database to precisely select affected tests, and can also expose its services as an MCP server.
Timeline (last 90 days)
RippleCI is a graph-native predictive CI engine, a CLI development tool, and also a Model Context Protocol (MCP) server built on FalkorDB.
RippleCI's core claim is to use the graph database's linear algebra engine to compute multi-hop change blast radius within 15 milliseconds, cutting CI runtime by 68% and mathematically guaranteeing 100% fault recall.
In an empirical study, the RippleCI team found that file-matching heuristics miss more than 33% of real regression faults.
RippleCI parses Python source files into an abstract syntax tree (AST) property graph, and the entire pipeline is: Git commit diff lines → AST dependencies synced to FalkorDB → multi-hop blast radius computation → machine-learning regression prioritization → bounded AST verification.
Inside FalkorDB, the software graph is modeled as fine-grained nodes and edges: nodes include File (with path, line count, change frequency score), Class, Function (with whether async, start and end line numbers), and TestCase (with historical duration, flakiness rate); edges include File-DEFINES->Class|Function, Function-CALLS->Function, File-IMPORTS->File, and TestCase-TESTS->Function|Class.
RippleCI queries FalkorDB starting from test cases and tracing 1 to 3 hops backwards along TESTS|CALLS edges, returning test ID, name, and historical duration, sorted by shortest graph distance and duration, extracting three levels of results in under 15 ms.
RippleCI's multi-signal Pareto-optimal ranker scores candidate tests by combining four factors: shortest topological graph distance, Git change frequency recency, historical execution duration, and a test flakiness penalty.
RippleCI has a built-in counterfactual AST mutation engine that, before letting a build through, injects bounded AST mutations into the modified code (such as flipping comparison operators, mutating return values, and shifting boundary conditions), then runs the selected test plan against this mutated code, using 100% mutant kill as empirical proof of release safety.
RippleCI has a built-in native MCP server (`ripple mcp serve`), and a single terminal command connects it to local AI editors, e.g. `ripple connect cursor` automatically detects and configures the Cursor IDE, and `ripple connect claude` connects to Claude Desktop.
After connecting RippleCI, AI agents gain several dedicated tools within their context window: `ripple_trace_blast_radius`, `ripple_prioritize_tests`, `ripple_verify_mutations`, and `ripple_query_graph`.
1 more timeline events
All Facts (11)
RippleCI is a graph-native predictive CI engine, a CLI development tool, and also a Model Context Protocol (MCP) server built on FalkorDB.
50%UnverifiedRippleCI's core claim is to use the graph database's linear algebra engine to compute multi-hop change blast radius within 15 milliseconds, cutting CI runtime by 68% and mathematically guaranteeing 100% fault recall.
50%UnverifiedIn an empirical study, the RippleCI team found that file-matching heuristics miss more than 33% of real regression faults.
50%UnverifiedRippleCI parses Python source files into an abstract syntax tree (AST) property graph, and the entire pipeline is: Git commit diff lines → AST dependencies synced to FalkorDB → multi-hop blast radius computation → machine-learning regression prioritization → bounded AST verification.
50%UnverifiedInside FalkorDB, the software graph is modeled as fine-grained nodes and edges: nodes include File (with path, line count, change frequency score), Class, Function (with whether async, start and end line numbers), and TestCase (with historical duration, flakiness rate); edges include File-DEFINES->Class|Function, Function-CALLS->Function, File-IMPORTS->File, and TestCase-TESTS->Function|Class.
50%UnverifiedRippleCI queries FalkorDB starting from test cases and tracing 1 to 3 hops backwards along TESTS|CALLS edges, returning test ID, name, and historical duration, sorted by shortest graph distance and duration, extracting three levels of results in under 15 ms.
50%UnverifiedRippleCI's multi-signal Pareto-optimal ranker scores candidate tests by combining four factors: shortest topological graph distance, Git change frequency recency, historical execution duration, and a test flakiness penalty.
50%UnverifiedRippleCI has a built-in counterfactual AST mutation engine that, before letting a build through, injects bounded AST mutations into the modified code (such as flipping comparison operators, mutating return values, and shifting boundary conditions), then runs the selected test plan against this mutated code, using 100% mutant kill as empirical proof of release safety.
50%UnverifiedRippleCI has a built-in native MCP server (`ripple mcp serve`), and a single terminal command connects it to local AI editors, e.g. `ripple connect cursor` automatically detects and configures the Cursor IDE, and `ripple connect claude` connects to Claude Desktop.
50%UnverifiedAfter connecting RippleCI, AI agents gain several dedicated tools within their context window: `ripple_trace_blast_radius`, `ripple_prioritize_tests`, `ripple_verify_mutations`, and `ripple_query_graph`.
50%UnverifiedThe author benchmarked by injecting over 5,000 synthetic mutation faults into real Python codebases.
50%