Changelog

Material changes to the product. Behind-the-scenes refactors aren't listed.

v0.6.0 — June 2026

Multi-channel alerts — Discord, Microsoft Teams, Telegram in addition to Slack. SIEM export now supports CEF + OCSF natively.
Per-API-key IP allowlists — bind a key to a CIDR range; out-of-range calls blocked at the edge before policy runs.
Ephemeral keys — short-lived, scope-restricted iv_eph_* keys (TTL + max_uses + tool/verb whitelist). See Ephemeral Keys.
Inbound Routes — destination-side proxy for SaaS agents you don't control (Salesforce Agentforce, HubSpot Breeze, Zendesk AI, Intercom Fin, GitHub Copilot, Microsoft Copilot Studio, Sierra, Decagon). See Inbound Routes.
PHI patterns in classifier — MRN, NPI, DEA, ICD-10, CPT, MBI, DOB. Bundled in the new healthcare-hipaa.yaml starter pack.
AWS tag-aware policies — match on resource tags (e.g. only apply approval to Environment=prod resources).
Auditor role — read-only role with sensitive headers redacted; for SOC 2 / PCI / HIPAA review windows. See Team & Roles.
Compliance evidence export — single button at Settings → Audit Log → Export.

interven Python v0.6.x — adds wait_for_approval() + ephemeral-key support + new tool / operation overrides on guard and gate.
@interven/sdk-js v0.5.x — guard() decorator + gate() one-liner + waitForApproval() + TypeScript types exported.
interven-langchain v0.4.x — adds [langgraph] + [langsmith] extras with native interven_tool_node, guard_state_graph, scan_tool_call, and LangSmith span emission.
interven-openai-assistants v0.3.x — scan_tool_calls with url_resolver override for custom tool names.
@interven/mcp-guard v0.4.x — proxy mode to wrap an existing MCP server.

New endpoint POST /v1/scan/response for response-body classification (read → write exfil correlation).
New endpoint POST /v1/a2a/scan for agent-to-agent message scanning.
New endpoint POST /v1/policies/apply for idempotent policy upserts (used by @interven/policy-cli).
New endpoint POST /v1/keys/ephemeral for ephemeral-key minting.

New pages: Inbound Routes, Tool Credentials, Compliance evidence export, Environments, Auditor-role view of every page.
Setup wizard for first-time signups.

7 starter policy packs shipped under packages/policy-packs/: sre-starter, vibe-coder-starter, hobbyist-starter, browser-agent-starter, enterprise-oem-starter, healthcare-hipaa, iso-42001.
daily_budget and numeric body-value conditions in the DSL.
Per-environment policy targeting (env_names: ["production"]).

New: LangGraph, LangSmith, GitHub Copilot Coding Agent, Claude Code (PreToolUse hook), Cursor, Microsoft Copilot Studio.

LangChain + CrewAI + OpenAI Assistants first-party SDKs.
MCP guard (@interven/mcp-guard) for Claude Desktop / Cursor / Cline.
Threat intel from 6 free feeds (urlhaus, urlabuse, openphish, phishtank, abuseipdb-free, alienvault-otx).
Initial OpenClaw plugin.
Browser Use / Computer Use shim (interven-computer-use).
Trust scoring per agent (PAM-style two-actor approval pattern).

For a daily-or-less notification when something material ships: