Following up on our 2 skills we published, we want to test them and built a IDA Chat plugin (powered by Claude Code agent SDK)
It’s basically a chat panel inside IDA that lets you talk to Claude about your binary.
You hit Ctrl+Shift+C, ask stuff like “what does this function do” or “find me the crypto functions”, and it actually runs scripts against your database to figure it out. Not just generic AI answers — it iterates, looks at the output, tries again if needed.
You should be able to install it using :
hcli plugin install ida-chat
Give it a shot if you’re curious: GitHub - HexRaysSA/ida-chat-plugin
Feedback welcome!
2 Likes
Why the focus on Claude with these?
Partly I ask because while I’ve heard good things about it, I currently use Grok, Gemini, and ChatGPT.
I know they all have strength and weakness.
Scores high for agentic use?
@Special-K I think this is strictly some sort of selection bias and/or personal preference. Some of us use Agentic AI for various tasks, and organically (rather than by force/decree) tend to use Claude Code.
From my perspective,
- Claude Opus (and Sonnet) do well in an agentic setting, by creating and following plans related to technical tasks (writing code, reading documentation, summarizing assembly/pseudocode).
- the SDK and API for Claude are reasonable, documented, somewhat complete, etc., so its easy to build plugins like the one @pal_hexrays posted above.
- the Pro/Max plans with non-API pricing are very convenient today (though of course this may not last if token subsidies evaporate).
I also use other models for various tasks (especially OpenAI models via Codex, and Gemini 3 Flash via API), paying special attention to the contracts and guarantees made by the providers. For example, Z.ai GLM-4.7 seems to be quite good and cheap, though I would only ever put widely known public information into it, based on my and Hex-Rays’s data policies.
Incidentally, I’ve been really enjoying the Pi agent harness (https://shittycodingagent.ai lol, or https://buildwithpi.ai), and can recommend it as an agentic AI frontend for OpenRouter- and LM Studio-hosted models. Example session: Session Preview – buildwithpi.ai
We’re more than happy to collaborate, learn, and develop solutions for other LLM providers 
Thank you for the detailed and educational reply.
I really have yet a lot to explore in the area myself.
Already the AI explanations, commenting, etc., by AI has been great so far in my experiences.
Mostly good, some bad/halucinations.