Trust, observability, and cost attribution for AI agents
Prove what your agents did — and whether it was worth it. Signed proof chains + decision-level cost attribution. EU AI Act ready.
Every agent action is cryptographically signed (Ed25519) and hash-chained (SHA-256). Tamper one record and the chain breaks. Verifiable, tamper-evident audit trail.
Decision-level cost attribution. See which agent decisions were worth the money, which were waste, and get actionable recommendations with projected savings.
Every event linked to the previous via SHA-256. Tamper one, break the chain.
Per-decision and per-agent cost breakdown. Know exactly where your tokens went.
Dead-end exploration, retry loops, redundant work. Confidence scores + evidence.
Specific actions with effort level and projected weekly savings. Mark as applied, track ROI.
Checkpoint to Sigstore Rekor for publicly verifiable timestamps.
Article 12 aligned compliance reports. 8-section reports in JSON or Markdown.
Only hashes stored, not content. Configurable redaction levels 0-3.
One command installs Claude Code hooks. Every session auto-traced.
npx agentproofs init # Generate Ed25519 keypair npx agentproofs install-hooks # Auto-capture every tool call # Use Claude Code normally — everything is captured npx agentproofs verify # Verify chain integrity npx agentproofs trace session.jsonl # Cost report npx agentproofs compliance-report # EU AI Act report