An open-source project called Fast-Anything is proposing a narrower alternative to rewriting an entire application in Rust: identify the small portion of code responsible for most execution time, port only those functions, and check the replacements before they are allowed to run.
The project’s repository says the tool can profile applications, record calls to eligible functions and prepare a task for a coding agent. The agent can be Claude, Codex or another command that accepts a prompt through standard input. Developers can also take the generated task file and complete the port through a separate workflow. The resulting Rust library exposes a C entry point and exchanges data as JSON, an approach intended to work across several languages without bespoke bindings.
Fast-Anything does not treat generated code as correct merely because it compiles. It replays recorded calls against the original and rewritten functions, then adds derived cases that probe values such as empty inputs, non-ASCII text, larger inputs, infinities and negative zero where relevant. A replacement is enabled only when its results and error behavior match the original under the project’s comparison rules. The tool also rejects a port when the cost of crossing into Rust makes it slower than the function it replaces.
That focus shapes which code can be moved. The repository describes support for top-level deterministic functions operating on plain data, rather than arbitrary methods with hidden state or complex runtime dependencies. Its scanner explains why a candidate cannot yet be ported. The project covers Python, Node.js and Ruby in its central pitch, while its documentation also describes paths for Java, C#, PHP, Go, C and C++. Developers still need a Rust toolchain to compile replacements.
The project also emphasizes reversibility. Setting an environment variable forces the application to use its original implementation, providing an escape hatch if a deployed port causes trouble. Its examples include reproducible benchmark material, but those project-supplied results should be read as demonstrations rather than independent performance validation. Actual gains will depend on the workload, the selected function and the overhead of moving data across the language boundary.
Fast-Anything arrives as coding agents increasingly move beyond code completion into multi-step maintenance tasks. Its design puts a verification layer between generated native code and activation. That does not prove broad correctness, because recorded and generated cases cannot cover every production input, but it gives developers a concrete review and rollback process for a tightly bounded optimization. The practical question is whether teams can find hotspots that are both expensive enough to justify a native port and simple enough to reproduce faithfully through the project’s plain-data interface.



