Layer 1
Instant Firewall
Gonative sidecar proxy. Submillisecond regex DPI on every prompt and model output.
How It Works
Every action. Intercepted. Evaluated. Proven.
Actsurance Shield sits between AI agents and the tools they call. It enforces deterministic policy, seals secrets, and produces cryptographic receipts an auditor can verify offline.
Live policy demo
Watch a risky refund request get blocked
The demo runs automatically, then you can reset it or enter your own tool call to watch the two-tier firewall emit a signed receipt.
The request lifecycle
An AI agent generates a tool call — a request to read from a CRM, issue a refund, write to a database, or call an external API. At this point, the agent has no credentials and no network access to the target system.
The request passes through our instant anomaly detection firewall, followed by strict policy evaluation and AI-assisted risk scoring. No AI hallucination sits in the final enforcement path.
If the policy returns ALLOW, Actsurance injects credentials and executes the call. If DENY or ESCALATE, the action is blocked or queued for human approval. Either way, a mathematically verifiable audit receipt is issued immediately.
How It Works Questions
Our friendly team is always here to help you with quick, clear, and reliable answers whenever needed.
Contact SalesHow does Actsurance Shield evaluate OpenAI function calls?
Actsurance Shield is designed to work with OpenAI tool/function calling workflows. It evaluates structured tool-call requests before execution using deterministic policy enforcement.
Where does Actsurance Shield sit in the agent stack?
Actsurance Shield sits at the execution layer, not the prompt layer. Any agent framework that makes structured tool calls can route those calls through the Shield without modifying the agent logic itself.
What happens when policy detects a risky action?
Does AI make the final Actsurance Shield decision?
3tier fallback: regex, Ollama (Llama 3.1), NVIDIA NIM (Nemotron120B). AI *informs* risk but never decides.
How does Actsurance Shield prove what happened?
Ed25519signed, SHA256 hashchained receipt for every decision.
The four-layer model, explained
Each layer has a single responsibility. None shares state with the agent.
Layer 2
Policy Evaluator
YAMLbased, defaultdeny engine. No AI in the enforcement path.
Layer 3
Zero-Trust Broker
Agent never sees raw credentials. Secrets are injected at execution time inside a restricted broker.
Layer 4
Audit Engine
Ed25519signed, SHA256 hashchained receipt for every decision.
See the full feature set
Every capability that makes up Actsurance Shield, with status.
View all features