Protect any SaaS AI agent (Salesforce Agentforce, HubSpot Breeze, Zendesk AI, Intercom Fin) without modifying agent code.

Protect SaaS AI agents you don't control

Most enterprise AI agents — Salesforce Agentforce, HubSpot Breeze, Zendesk AI, Intercom Fin — run inside vendor clouds. You can't inject middleware into the agent's runtime. You can't add an SDK. You can't intercept its tool calls.

But you can control where it sends its API calls.

How destination-side proxy works

SaaS Agent (Salesforce, HubSpot, etc.)
    │
    │  configured to call Interven URL
    ▼
┌──────────────────────────────────────┐
│  Interven Inbound Gateway            │
│  https://api.intervensecurity.com/   │
│         inbound/<your-prefix>/...    │
│                                      │
│  1. Match path to route              │
│  2. Identify calling agent           │
│  3. Run full policy + risk pipeline  │
│  4. Decide ALLOW/DENY/SANITIZE/      │
│       REQUIRE_APPROVAL               │
│  5. Forward to real upstream with    │
│     stored credentials               │
└──────────────────────────────────────┘
    │
    │  with real Bearer token
    ▼
Real Slack / GitHub / DB / API

The agent never sees the real credential. Interven holds it. The agent calls Interven; Interven scans, then forwards.

Setup (3 steps)

1. Create an inbound route

In the Console → Inbound Routes:

Path prefix — short identifier, e.g. sf-slack
Upstream URL — the real API base, e.g. https://slack.com/api
Upstream headers — credentials Interven adds when forwarding:
```
Authorization: Bearer xoxb-your-real-slack-token
```
Tool name (optional) — slack, github, etc., for policy attribution
Agent identification — how Interven attributes the call to an agent

2. Configure your SaaS agent

Point the SaaS agent at the Interven URL instead of the real API. Specifics per platform:

Salesforce Agentforce — Named Credentials + External Services
HubSpot Breeze — Agent Tool actionUrl
Zendesk AI — Integration Builder
Intercom Fin — Data Connector URL

3. Test and monitor

Every call appears in Activity with full decision evidence: matched policies, risk score, classifications. Set policies in Policies to control what each agent can do.

Decision behavior

Decision	What the SaaS agent sees
ALLOW	Real upstream's response, transparently
DENY	HTTP 403 with `reason_codes`
SANITIZE	Real upstream's response (Interven sent redacted body)
REQUIRE_APPROVAL	HTTP 202 with `approval_id`. Retry within 10 min after analyst approves — auto-allowed via `RECENT_APPROVAL_GRANT`.

Headers you'll see

Every response from the inbound proxy includes:

X-Interven-Decision — ALLOW / DENY / SANITIZE / REQUIRE_APPROVAL
X-Interven-Trace-Id — UUID for telemetry correlation
X-Interven-Reasons — comma-separated reason codes (if any)

Coverage matrix

Platform	Method	Coverage
Salesforce Agentforce	Named Credentials → Interven URL	Full
HubSpot Breeze	Agent Tool actionUrl → Interven URL	Full
Zendesk AI	Integration Builder URL → Interven URL	Full
Intercom Fin	Data Connector URL → Interven URL	Full
Sierra	Custom API integration → Interven URL	Full
Decagon	API endpoint config → Interven URL	Full
Microsoft Copilot (custom agents)	MCP server URL → Interven URL	Custom agents only
Google Gemini (Vertex agents)	Custom agent endpoint → Interven URL	Custom agents only

For platforms that don't expose an API endpoint config, use OAuth scope restriction + audit log monitoring as a fallback.

SaaS Agent Protection