DIMAGGI AI — Policy Firewall for AI Agents
DIMAGGI AI sits between your AI systems and the tools they use — intercepting every action,
evaluating it against your governance rules, and producing a tamper-evident audit trail.
Enforce security policies on AI agent tool calls before they execute.
How It Works
- Intercept — Every AI agent tool call passes through the DIMAGGI AI proxy before execution.
- Evaluate — Tool calls are matched against your policies. Rules fire based on tool name, parameters, agent identity, and context.
- Decide — Allow, deny, escalate for human review, or flag as a near-miss. Shadow mode logs without blocking.
- Audit — Every decision is logged with HMAC signing and SHA-256 hash chaining. Tamper-evident and regulator-ready.
Live Demo: 30-Day Security Snapshot
Simulated activity for a financial services company running four AI agents in shadow mode.
- 1,847 tool calls evaluated across 4 AI agents
- 74% policy coverage (26% matched no policy and were allowed by default)
- $87,500 near-miss value flagged in shadow mode
- 12ms average evaluation time per tool call
Active Security Policies
- High-Value Transfer Guard
- Escalates wire transfers and payments over $10,000 for CFO approval
- PII Data Protection
- Blocks email sends, CSV exports, and file uploads that may contain personal data
- External API Rate Limit
- Flags high-frequency calls to external services to prevent data exfiltration or runaway costs
- Security Command Block
- Denies shell command execution and script invocations from all agents unconditionally
Shadow vs Enforcement Modes
Shadow Mode — All tool calls are evaluated and logged, but nothing is blocked.
Near-miss records show exactly what enforcement would have caught. Use this to tune policies before going live.
Enforcement Mode — Policy violations are actively blocked.
Escalations require human approval before the tool call proceeds.
AI Agents Under Governance
- Finance Bot — Wire transfers, payments, financial reporting
- Data Pipeline — Data ingestion, CSV exports, file management
- Support Agent — Customer communications, ticket handling
- Code Review Bot — Security scans, shell diagnostics, code analysis
Audit & Compliance
- Tamper-Evident Audit Trail — Every decision is HMAC-signed and linked via SHA-256 hash chains. Any modification to historical records is detectable.
- Complete Decision History — Full trace of every tool call: who requested it, which policies matched, what the outcome was, and why.
- Offline Verification — Export audit chains as JSONL and verify integrity offline with the CLI tool.