Skip to content

Policies & Enforcement Cloud / Team

Configure how Intutic governs your AI coding agents using a two-layer protection architecture.

Two-Layer Protection

Intutic protects your workspace using two complementary enforcement layers:

  1. Layer 1 (Client/Rule Layer) — Compiles and injects natural-language instructions (SOPs) directly into harness configuration files that agents read before starting
  2. Layer 2 (Network/Packet Layer) — Intercepts raw LLM API calls via the proxy to block policy violations, redact sensitive data, and break execution loops

Data Loss Prevention (DLP)

The DLP gate scans request payloads and tool outputs for sensitive data before forwarding them to LLM providers.

What's Detected

  • API keys and access tokens
  • AWS credentials and service account keys
  • Social Security Numbers (SSNs)
  • Personally Identifiable Information (PII)
  • Database connection strings
  • Private keys and certificates

Enforcement Modes

Configure DLP behavior using the INTUTIC_DLP_MODE environment variable:

ModeActionDescription
BlockReject requestThe request is rejected entirely with a DLP error
RedactMask & forwardSensitive data is replaced with masks (e.g., sk-proj-****) and the request continues
LogObserve onlyThe incident is logged in the dashboard but the request proceeds unchanged

TIP

Start with log mode to understand your DLP exposure, then switch to redact or block once you've reviewed the findings.


Enforcement Actions (PCAS)

The Policy Compliance and Action System evaluates every tool use request against your safety guidelines:

Risk Categories

Each tool execution is classified by risk level:

CategoryExamples
NoneRead-only operations, search, local project compilation
Credential AccessReading .env files, accessing keyrings, querying credential managers
DestructiveRunning rm -rf, deleting cloud infrastructure, executing destructive database queries

Four Enforcement Actions

ActionWhat HappensWhen It's Used
BypassTool call passes through unmodifiedFully compliant with all policies
EnhanceSafety prompts or warnings are injected into the agent's contextMinor risk but generally safe
HijackTool outcome is replaced with a mock response or overridePolicy violation that can be safely redirected
KillTool call is blocked immediatelySerious policy violation or budget breach

Intervention Modes

How the enforcement action is communicated:

ModeBehavior
TransparentDeveloper sees a clear explanation of the policy breach
OpaqueAgent receives a generic error and tries an alternative approach
Silent LogRequest proceeds but is flagged in the dashboard for review

SOP Synchronization

When you run intutic connect, the sync daemon keeps your workspace aligned with the control plane:

  1. Detects all active coding harnesses in your workspace
  2. Formats SOPs into harness-native structures (Markdown for Claude Code, JSON for Antigravity, YAML for Aider, etc.)
  3. Writes governance rules atomically, preventing configuration drift or corruption
  4. Monitors for manual edits and rewrites configurations based on your bypass enforcement tier

See How It Works for the full sync daemon architecture.


The active firewall for AI agents