Home › Changelog

Changelog

What's new in the guardrail, the agent API and the tooling.

June 2026

Pro tooling for agent fleets

Expanded detection: payment cards (Luhn-validated), SSNs, IBANs and more leaked-key patterns — surfaced as machine reasonCodes.
Safe rewrite: when an email is risky only because of removable content, the API returns a cleaned safeBody to send instead.
Per-agent keys (named, scoped verify/sandbox), per-agent stats, per-request policy override, real-time action.verified webhook, and a tamper-evident audit chain.
Published accuracy: a reproducible benchmark — and a status page.
API robustness: Idempotency-Key header, webhook retries + delivery log, cursor pagination.

May 2026

The decision API: send / request_human_confirmation / do_not_send before every agent email.
MCP server + OpenAI / Anthropic tool schemas + zero-dependency JS & Python SDKs.
Semantic LLM judge on top of the rule engine, workspace policies, signed webhooks, credit packs and auto-recharge.
Bilingual site & dashboard (FR / EN), docs, and the live demo.