Features

How it works — every flow in the open.

Every capability across the three planes — Visibility, Control, Protection — shown as the flows it actually runs. Each diagram is illustrative of the enforcement path; every path is fail-open by design.

Protect it

Detection & enforcement — ST Shield.

One detection engine runs at three enforcement points. Sensitive data is caught before it leaves — and every value is masked in the record.

ST Shield detection and enforcement flowchart Prompt pre-send, command pre-execution, and hub ingest all feed the ST Shield detection engine. A policy decision routes each event to allow, redact, or block, and every event produces a masked audit record exported to SIEM. Promptpre-send Commandpre-execution Hub ingestserver-side ST Shield cards · IDs · keys entropy · custom match? Block Redact & continue Allow (no match) policy: block policy: redact
Allow Redact / flag Block

Whatever the outcome, ST Shield writes a masked audit record — the matched secret is never stored in the clear — and can export it to your SIEM or a webhook.

Govern it

The approval loop — fail-open by design.

Approve an agent's next command from your dashboard or your phone. If the decision can't reach you in time — or our software is down — the agent keeps working. We are in the path, never in the way.

Fail-open approval loop flowchart An agent tool call is checked against policy. If no approval is required it runs. If approval is required, the admin is notified on dashboard and phone; approve runs the call, deny stops it. If the request times out or the software is unavailable, the call fails open and runs. Every path is logged. Agent toolcall approval required? no yes Notify admin dashboard + phone Approve → run Deny → stop No match → run FAIL-OPEN timeout or ST down → agent keeps working
Runs Stopped Fail-open safeguard
Govern it

ST Gate — one governed door for MCP.

Every MCP client — CLIs, desktop apps, Cursor, Windsurf, VS Code — routes through one gateway. Only approved servers are reachable; every tool call is audited; credentials are brokered, never handed to the client. Read the full ST Gate page →

  • Claude Code
  • Cursor
  • Windsurf
  • VS Code
  • Desktop apps
ST Gate MCP gateway flowchart Multiple MCP clients route through ST Gate, which checks an approved-server registry, audits every tool call, and brokers credentials. Approved servers are reached; unapproved servers are blocked. Claude Code (CLI) Desktop apps Cursor / Windsurf VS Code ST Gate approved-server registry tool-call audit credential broker ◆ one governed door Approved MCP servers reachable + audited Unapproved servers blocked at the gate

Integrations — with a purpose, not a logo wall.

Every integration exists to answer a specific question or close a specific gap. We integrate with your stack; we don't ask you to replace it.

IntegrationWhat connectsPurpose
MCP clients (via ST Gate) Claude Code, desktop apps, Cursor, Windsurf, VS Code Route every MCP tool call through one governed door — approved-server registry, tool-call audit, credential brokering.
Vendor compliance APIs OpenAI Compliance Logs Platform, Anthropic Compliance API, Google Vault for Gemini Retrieve prompt/response content for org-managed enterprise desktop accounts — the only sanctioned way to see desktop-app conversations.
SIEM & webhooks Any SIEM or HTTP endpoint Stream ST Shield findings and audit events into the tooling your SOC already watches.
CASB & enterprise browser Netskope, Zscaler, Microsoft Purview Enrich endpoint and web signal for desktop-app coverage — filling the gap where CASB can't see terminal agents.
Identity provider SSO (SAML / OIDC), SCIM Tie RBAC and the identity graph to your directory, and discover shadow-AI logins via OAuth.
Deployment & IaC Docker Compose, Helm, Terraform, offline bundle Install ST Hub the way your platform team already ships infrastructure — including fully air-gapped.
Transparent by design

What leaves each machine — you choose.

Reporting is a ladder, set per policy. It starts at anonymous metrics and only climbs to full content where you deliberately enable it. The active tier is always visible to the employee on their own dashboard.

Privacy reporting tiers Four reporting tiers from least to most data: anonymous metrics, usage and cost, activity metadata, and full-content compliance. The employee always sees the active tier. TIER 0 Anonymous metrics TIER 1 Usage & cost by person/team TIER 2 Activity metadata + audit TIER 3 Full-content compliance mode more data leaves the machine → you decide how far it climbs ▲ the active tier is always shown to the employee — no stealth mode exists

The full feature matrix.

Everything below is included in the single plan — nothing is gated.

PlaneCapabilityWhat it does
See itUnified spendToken spend and API-equivalent cost by hour, day, model, project, cost center.
See itLive sessionsLive context fill and activity per session, every coding agent.
See itSearch & recoveryFull-text search and transcript recovery across every session.
See itBudgets & chargebackDepartment rollups, alerts at 80/100%, showback and chargeback exports.
See itCross-agent compareClaude vs Codex vs Gemini vs Kimi, side by side.
Govern itRemote approvalsApprove or deny a tool call from dashboard or phone — fail-open.
Govern itCentral policyOrg-wide guardrails pushed to every machine, enforced by local fail-open hooks.
Govern itST GateMCP gateway with approved-server registry, tool-call audit, credential brokering.
Protect itST Shield engineCards (Luhn), national IDs, keys, private keys, entropy heuristics, custom patterns.
Protect itThree enforcement pointsPrompt pre-send, command pre-execution, and hub ingest.
Protect itBlock / redact / flagPer-pattern policy actions with a disposition workflow.
Protect itMasked audit & SIEMSensitive values masked everywhere; export to SIEM or webhook.
PlatformIdentity graphCLI, desktop, MCP, endpoint, and IdP signal stitched to one employee.
PlatformDeploymentOn-prem or your own cloud — Docker, Helm, Terraform, or air-gapped bundle.
PlatformAccess & SSORBAC with SSO / SCIM for org, department, and per-user views.

Installation — running in an afternoon.

Deploy the hub on infrastructure you own, enroll machines with a lightweight agent, set policy, connect your integrations. No SaaS account, no data leaving your walls.

1

Deploy ST Hub

Stand up the hub on your own servers or cloud — Docker Compose, Helm, or Terraform. Air-gapped? Load the offline bundle.

2

Enroll machines

Install the stl spoke on each machine. It reads AI session logs read-only and reports to your hub.

3

Set policy

Choose a reporting tier, push org-wide guardrails, and configure ST Shield detection and ST Gate's approved-server registry.

4

Connect integrations

Wire up SSO/SCIM, your SIEM, CASB, and the vendor compliance APIs for desktop-app content.

# 1 — Deploy ST Hub on your own infrastructure (choose one) docker compose up -d # on-prem, single host helm install st-hub sentinel/hub # kubernetes terraform apply # your cloud VPC # 2 — Enroll a machine (reads agent logs read-only) stl enroll --hub https://hub.internal.yourco.com # 3 — Push an org-wide guardrail from the hub stl policy push --rule "deny: curl | bash"

Illustrative — exact commands ship with your licensed bundle.

Deploy anywhere you run it On-prem AWS Google Cloud Azure

See these flows on your own infrastructure.