Blog
Technical deep-dives, project updates, and lessons learned.
A walkthrough of FM-Agent's compositional reasoning pipeline — from top-down specification generation to natural-language Hoare logic verification — and how it found 522 previously unknown bugs across four large-scale, LLM-generated systems.
Read more →Learn how to set up FM-Agent on your own codebase — from installation to running your first compositional analysis and interpreting the results.
Read more →We generalize classical Hoare-style inference rules to work with natural-language pre/post-conditions, enabling LLMs to perform rigorous program reasoning.
Read more →How we designed FM-Agent's top-down spec generation to analyze large call graphs concurrently while maintaining consistency across function boundaries.
Read more →For every potential bug, FM-Agent generates system-entry test cases. Here's how this automated validation works and why it matters for developer trust.
Read more →