Security Teams
Enforcement at runtime. Evidence for every decision.
Guard runs before the action executes — not after. Every allow, approve, and block produces a signed receipt. Your compliance team has answers, not logs to correlate.
Enforcement, not detection.
Guard intercepts every action before it executes and evaluates it against policy. This is not a SIEM or a log analyser. Policy runs at the action layer — before money moves, before the deployment lands, before the secret is read.
CLI hook
Claude Code · Cursor · Copilot · Codex
HTTP proxy
Any SDK with configurable base URL
MCP layer
Any MCP client — Claude Desktop, Cursor, Windsurf, Codex, and more
Signed evidence for every decision.
Every Guard evaluation produces a receipt. The receipt records the agent, action, resource, decision, rule that fired, user, and timestamp. Receipts are SHA-256 hash-chained — altered entries break verification. No pixel editing required to make your audit trail look right.
15 compliance packs. Shipped.
Each pack is a JSON ruleset mapped to a compliance standard. Install once and Guard evaluates every agent action against the pack's rules automatically.
Deploy where your controls need to live.
Guard runs in your cloud, on your infrastructure, or as SaaS. Same policy engine, same audit trail, regardless of deployment model.
- SHIPPEDSaaS (conductai.ai)US-based
- SHIPPEDDocker Compose (self-hosted)
- PREVIEWKubernetesReference templates
- PLANNEDAir-gapped / on-prem
Guard threat coverage
Honest scope. We publish where Guard stops.
Consequential actions guarded by policy
Refunds, deployments, network changes, secret reads — any action routed through Guard is evaluated before execution.
Audit trail integrity
SHA-256 hash chain on every decision. Altered entries break verification.
Compliance pack coverage (SOC2, HIPAA, PCI DSS, EU AI Act, and 11 more)
15 JSON rulesets evaluated against every agent action automatically.
Pre-call prompt injection
Guard includes a prompt-injection detection pack. Attacks that alter model intent before tool selection are upstream of Guard.
Model-layer attacks (adversarial inputs to the LLM)
Pattern-based detectors. ML-based detection is not implemented.
Actions that bypass Guard enforcement surfaces
An agent that does not use the hook, proxy, or MCP layer is invisible to Guard.
Cross-agent context correlation
Guard evaluates each action in isolation. Operations context is planned.