Rikugan

buzzer-re/Rikugan is ranked 4,175 among these 8 peers; devagrawal09/jev-review is ranked 4,176.

This project was vibecoded together with my friend, Claude Code. Documentation | Architecture | Issues Install Auto-detects IDA Pro, Binary Ninja, or both. **Linux / macOS:** **Windows (PowerShell):** ***Windows users should also install the latest Microsoft Visual C++ Redistributable 2015-2022 x64 before launching IDA. Rikugan uses IDA's bundled PySide6/Shiboken Qt bindings, and an outdated VC++ Runtime can cause it to crash during startup. *** For host-specific install, manual setup, and configuration, see the docs. No, Rikugan is an ***agent*** built

to live inside your RE host. It does not consume an MCP server to interact with the host database; it has its own agentic loop, context management, role prompt (source), and an in-process tool orchestration layer. The agent loop is a generator-based turn cycle: each user message kicks off a stream→execute→repeat pipeline where the LLM response is streamed token-by-token and tool calls are intercepted and dispatched. The agent really ***lives*** and ***breathes*** reversing. The

agent always asks permission before running scripts and will never execute the target binary. The orchestrator maps the binary (imports, exports, strings, key functions), then spawns isolated subagents to analyze in parallel. Each reports back, and the orchestrator synthesizes a complete picture. Rikugan explores the binary, builds context, and applies the patches. Reuse skills and MCP servers from Claude Code and Codex. Profiles Profiles let you customize the agent to fit your analysis needs. They give you granular control over

which data the LLM can read, restrict which tools it can use, and let you define custom rules to filter data. Recommend Claude Pro/Max plan with OAuth over API. Both Anthropic models use prompt caching. 5 / Highspeed** | On par with Opus in local tests. Hallucinates more than Anthropic/MiniMax. 5** | Strong coding, but lacks rigor for complex RE tasks. | | **LLAMA 70B / GPT 120B OSS** | Interesting but not production-ready for RE. | Also supports any OpenAI-compatible endpoint and Ollama

for local models. 0+ with Hex-Rays decompiler or Binary Ninja (UI mode) Python 3. 10+ At least one LLM provider Windows, macOS, or Linux Windows: latest Microsoft Visual C++ Redistributable 2015-2022 x64 **IDA Pro + Python >= 3. Rikugan includes a workaround, but Python 3. " But this year I completely changed my mind when I saw what was achievable. AI is not the ChatGPT from 2023 anymore; it's something entirely different. For that reason, I decided to invest this year

in researching this topic. Rikugan is just one of many projects I've built in the last three months. The first version was built in a single night. Within two days it already supported both IDA and Binary Ninja. Within three days, it was essentially what you see here, with only minor tweaks since. This is a work in progress with many areas for improvement. I took care to ensure this wouldn't be another AI slop project, but I'm certain there

is still room to grow. If you find bugs, have suggestions, or want quality-of-life improvements, please open an