#preflight

memory → preflight → action Preflight validation before every AI agent action. - Block unsafe memory before agents act on it - Stop dangerous drift propagation across agent fleets - Z3 formal verification (overridable: false) - F1=1.000 across 1,190+ adversarial corpus cases (R1-R11) - Co-benchmarked with @grok — Round 11 F1=0.979 on compound attacks - 4 independent detection layers: timestamp, identity, consensus, provenance Demo key: sg_demo_playground — no signup needed

Compliance evidence infrastructure for regulated tokenized markets. 27 MCP tools with ES256K-signed responses covering MiCA preflight (PASS/WARN/BLOCK), peg monitoring, reserve quality, custody risk, evidence grading (A-F across 61 RWA protocols + 105+ stablecoins), KYA agent identity, tamper-proof audit trail, macro risk (86 FRED + 20 ECB indicators), and signed PDF reports with XRPL blockchain anchoring. Part of a 3-server ecosystem: Compliance MCP (27 tools), Macro Intelligence (13 tools), Stablecoin Risk MCP (13 tools) — 53 tools total. New in v4.2: Know Your Agent (KYA) identity + trust scoring, tamper-proof SHA256 chain-linked audit trail, Evidence Lifecycle (6 states), OAuth 2.0 with Dynamic Client Registration, Enterprise Trust Layer (14 components), Unit-Based Billing, Trust Policy v1.0, 8 reference workflows. OAuth 2.0 + StreamableHTTP + SSE. Free tier: 300 units/day, no API key required. MiCA enforcement July 2026. Connect: claude mcp add --transport http feedoracle https://feedoracle.io/mcp/ Trust proof: https://feedoracle.io/trust Docs: https://feedoracle.io/docs.html

Local-first memory firewall and MCP server for AI agents. Audrey gives coding agents recall, memory capsules, preflight checks, reflexes, validation, tool-trace learning, contradiction handling, and SQLite/sqlite-vec storage.