Granite for your RE agent
Give your agent a binary analysis engine.
Delegate function-level investigation to Granite, then inspect the names, descriptions and related code behind the answer.
Private analyst access is being prepared for IDA on macOS. For authorized security research.
What checks happen before entry data is consumed?
Header checks happen in parse_document. The shown entry path checks its length in parse_entries before calling consume_entry.
- Header checks
parse_document → read_headerRejects a short input or an invalid header.- Entry check
parse_entries → consume_entryRejects an entry longer than the remaining input.- Still to review
- Other paths and how the remaining length is computed.
Inspect the illustrative code
// parse_document
if (size < HEADER_BYTES) return INVALID;
if (!read_header(buffer).valid) return INVALID;
return parse_entries(/* remaining input */);
// parse_entries: shown entry path
if (entry.length > remaining) return INVALID;
return consume_entry(entry.data, entry.length);Keep your disassembler.
Give your agent function-level analysis.
One question.
Evidence you can inspect.
Use Granite alongside your own investigation. Review the explanation against the code and decide what to investigate next.
- 1
Set the question.
Choose a function and a concrete question about behavior, input handling or a code path.
- 2
Delegate the analysis.
Your agent uses Granite’s dedicated engine for function-level investigation, with names, descriptions and related code for context.
- 3
Check the answer.
Inspect the supporting code, correct mistakes and keep unresolved questions visible.
REA is free.
The investigation isn’t.
A free connector still leaves model usage, investigation and analyst review to account for. Granite gives your agent a dedicated function-analysis engine. Judge it by the useful answers it produces and the effort needed to verify them.
Why pay if I already use REA or IDA MCP?
Tool access is part of the workflow. The purchase decision is whether Granite produces useful analysis with an acceptable review burden and total cost on your task. Compare it with the tools you already use.
Does Granite replace my agent or disassembler?
Granite is an analysis engine an agent can delegate to. Your disassembler remains where you inspect the code, and you remain responsible for the conclusion.
What should a useful evaluation establish?
Agree a real question and acceptance criteria, then review the output, corrections, unresolved cases, model usage and analyst effort. A useful answer and repeat use matter more than the number of generated reports.
Work directly with the founders
Analyst access first.
Team evaluations alongside it.
We’re preparing vetted, paid analyst access for IDA on macOS. The planned offer includes defined usage and support boundaries, with additional usage available for the current subscription period.
Private access is in preparation. Pricing and usage limits will be stated before purchase.
For teams: one binary, one useful question.
Start with a scoped evaluation on software you’re authorized to assess. Agree the question, handling plan, deliverables and fee before analysis starts.
- A question from your actual analysis workflow.
- A mutually approved binary and handling plan.
- Analyst review of the output and its limitations.
- A readout against criteria agreed with your team.
What are we evaluating?
Whether Granite produces useful, reviewable analysis for your task. Outputs can be incomplete or incorrect. Agree the supported environment and scope before an evaluation.