Favicon of Reverify

Reverify

A local AI verification toolkit checks binary and code claims with evidence. MIT-licensed, it connects to coding assistants through MCP.

Reverify is a local verification toolkit for developers and reverse engineers who use AI to analyze binaries or rewrite code. It checks a model's claims against the actual file or executed behavior, then returns a verdict with evidence. The Python project is open source under the MIT license and runs on Linux, Windows and macOS as a CLI or MCP server.

Its binary analysis covers PE, ELF and Mach-O files, with disassembly for x86, x64, ARM and ARM64. It can inspect bytes, imports and structured data, scan patterns, and emulate code to check claimed results. The core uses pure Python; optional engines include Capstone, Unicorn, LIEF and Z3. With angr, it can assess function calls, references and reachability, while labeling those findings as analysis-derived rather than fully verified.

For source code, Reverify compares Python or C implementations over shared inputs. When a candidate disagrees with the reference, it reports the failing input and both outputs. Verdicts also identify the binary's hash and the engines used, so another analyst can repeat a check.

A local ledger preserves established facts and rejected claims across context resets. Session handoffs keep verified evidence separate from the model's unverified notes. Rollover support works with Claude Code, Codex, Gemini CLI and OpenCode; Cursor can access verification tools through MCP. The verification toolkit and ledger run locally, while model access depends on the connected agent or endpoint.

Similar to Reverify