3 projects
firstwatch
First line of defence for AI agents: spots attacks hidden in pages, files and tool results, and records each one as evidence in an open, OpenTelemetry-ready format with MITRE ATLAS and OWASP IDs. No dependencies.
doceval
Eval harness for LLM extraction pipelines — field-level accuracy, failure taxonomy, cost tracking. Documents, emails, HTML, transcripts, feeds.
tally-aiagent
Safe writes to Tally Prime for AI agents — refuses any name Tally does not already hold, and reads every voucher back to prove what was stored.