Welcome to Sector8

Pre-execution control for AI agents. Every proposed action returns ALLOW or DENY before it executes.

Pre-execution control for AI agents

AI agents propose actions. Sector8 decides whether those actions can proceed before the proposed action executes.

Sector8 gives your business an independent control point between an agent and a real-world side effect: refund creation, customer data export, shell execution, inventory reorder, deployment, file access, API call, or model handoff.

Every evaluated action returns an ALLOW or DENY decision with an audit-grade evidence record: reason_code, decision_trace_id, evidence_hash, and policy_version_id.

The agent does not approve itself. Proposed actions pass through an independent decision gate. Enforcement holds where the runtime routes proposed actions through Sector8 and fails closed on gate unavailability.

---

What each buyer should understand

Security leaders

Sector8 moves control to the execution boundary. You can block prompt injection, unsafe tool use, PII exposure, and policy breaches before the action runs.

Engineering teams

Sector8 is model-agnostic. Integrate at the tool or action boundary through SDKs, REST evaluate, or MCP-connected runtimes.

Operators and compliance teams

Each decision creates evidence with a reason code, trace ID, evidence hash, and policy version so reviewers can see what happened and why.

Business owners

Versioned policy sets can be bound per tenant to control what agents are allowed to do, such as refund limits, inventory reorder limits, customer data export, and human review for high-risk actions.

---

What you can do with Sector8

Decide before execution

Stop prompt injection, shell abuse, PII exfiltration, and business-policy breaches before the proposed action executes.

Govern agent tool calls

Intercept proposed actions from SDKs, REST evaluate, or MCP-connected runtimes, including Claude Code tool calls, before operating systems or business systems act.

Prove every decision

Return an audit-grade decision record with decision, reason code, evidence hash, policy version, and decision trace ID.

Drop into existing apps

Keep your model and application architecture. Add Sector8 at the point where the app would otherwise dispatch a tool call, API call, or agent action.

---

How it works

The agent or model proposes an action. Sector8 evaluates the proposed action against baseline safety checks and the active policy set bound to your tenant. Your runtime then honors the returned decision.

Proposed AI action
(tool call, API call, model call, or handoff)
    |
    v
Sector8 evaluate
    |
    +-- Pre-filter
    |   Fast path for obvious unsafe patterns
    |
    +-- Admission and policy
        Baseline safety plus tenant-bound and agent-scoped policy
    |
    +-- ALLOW  -> runtime may execute the action
    +-- DENY   -> runtime blocks the action

Unlike output filters, SIEM, or audit logs alone, Sector8 owns the execution decision before the side effect starts. When correctly integrated, there is no authorized path around the gate.

Sector8 does not replace content security, SIEM, DLP, or customer governance. It owns the pre-execution decision for routed actions and emits evidence that downstream systems can use.

Policy model

Sector8 policy is modeled as versioned policy sets, bound per tenant. A tenant binding selects the policy version used by evaluate for that environment. Agent-scoped policy can add narrower controls for a specific agent or action.

Operators bind, suspend, resume, and review versioned policy sets through the control plane. The runtime gate enforces the active binding for routed actions and still applies baseline safety checks.

---

Why pre-execution matters

Enforcement point	Timing	Data / system exposure
Sector8 gate	Before the proposed action executes	None from that action because the side effect has not started
Output filters / model safety	After the model generates content	The model may already have seen sensitive input
SIEM / DLP	After network or tool activity	Data or action may already have moved
Audit logs alone	After execution	Detective only

---

Canonical decision artifact

Every evaluate path should converge on the same fields:

Field	Meaning
decision	ALLOW or DENY
reason_code	The machine-readable reason for the decision
decision_trace_id	Unique audit identifier for the evaluated action
evidence_hash	Fingerprint of the decision evidence for correlation and integrity checks
policy_version_id	The policy version or runtime policy identifier applied

These fields are what you show in Decision Feed and Evidence. They are not a model-written explanation.

---

Get started

Quickstart

Choose MCP, SDK, or REST and get your first decision.

Python SDK

Call evaluate from your application runtime.

Guard MCP Server

Govern Claude Code tool calls routed through the Sector8 MCP connector.

---

Check the gate

Use the endpoint issued for your environment.

curl -s https://sdkapi.sector8.ai/api/health

Expect a healthy status with active invariants and gate capabilities before sending routed actions through the gate.