What is Interven?
Inline policy + risk firewall for AI agent tool calls.
Interven sits between your AI agents and the upstream APIs they call (Slack, GitHub, Salesforce, Jira, Workday, etc.). Every governed call is normalized, classified for secrets and PII, run through policy and risk signals, and gets one of four outcomes:
| Decision | Meaning |
|---|---|
| ALLOW | Forward to the real API. |
| DENY | Block with a structured error; raise an incident if needed. |
| REQUIRE_APPROVAL | Pause, surface to a human analyst in the Console, then continue in the same conversation turn once approved. |
| SANITIZE | Redact sensitive fields, then forward (or return the redacted body). |
Why teams use Interven
- Drop-in for any framework. LangChain, CrewAI, OpenAI Assistants, OpenClaw, or
the raw
/v1/scanHTTP API. Five lines of code per integration. - Human-in-the-loop on writes. When the agent wants to do something sensitive, it pauses; an analyst clicks Approve in the Console; the agent finishes — same turn, no retry needed.
- Built-in DLP. PII, SECRETS, and INTERNAL data classifiers fire before the payload leaves your network.
- Per-tenant credential vault. Agents call Slack, GitHub, Drive, Jira through
Interven using only an
iv_live_*API key. Real OAuth tokens never reach the agent runtime. - Policy you can actually write. Match by tool, operation, data class, body content, or external principal. Edit in a wizard or as raw JSON.
Get started
Beta access
Interven is in private beta. To request access, email founder@intervensecurity.com or join the community Slack.