AI Operations Disclosure
This page documents the autonomous AI agents we operate in public, the data they touch, and the controls around them. It exists for two reasons: to satisfy our obligations under the EU AI Act’s transparency rules and Meta’s Platform AI Disclosure policy, and — more honestly — because if we’re going to claim our agents run autonomously in production, we owe you the receipts.
Autonomous AI agents
Two of the agents below interact with the public: forger-social-01 on Facebook, and forger-support-veloura on the Veloura demo store. Wherever they act, you’ll see an AI-operated label — on the workshop tiles, on individual exhibit pages, in chat headers, and in social-platform metadata where the platform supports it. If you’re talking to one of our agents, you’ll know.
forger-social-01
Drafts and schedules posts, and replies to comments and direct messages, on a public Facebook page we own. It accesses page posts, comments visible to the page admin, and metadata exposed by the Meta Graph API. It runs on a pinned MiniMax-M2.7 model (served via NVIDIA NIM) — the version is locked and only changes after offline smoke testing with explicit operator approval. Every action it takes is written to a session-level audit log we can review on demand. The agent has no ability to modify the page’s settings, post on behalf of unrelated pages, or contact users outside Facebook. A human kill-switch is wired in: a single command pauses the agent and rolls back any in-flight scheduled action.
forger-support-veloura
Handles customer messages on our Veloura demo store. It accesses conversation transcripts and order metadata for the conversation in front of it — nothing else. It runs on the same pinned MiniMax-M2.7 setup as forger-social-01: version locked, smoke-tested before any update. Every reply it sends is logged. It cannot mutate orders, take payments, or escalate beyond a handoff to email. The kill-switch is the same shape: one command pauses replies and routes new messages to a human queue.
Opt-out and human review
If you’d prefer the agent didn’t reply to your Facebook comments, email hello@forgingapps.com with subject [FB-OPTOUT] and a link to your profile or the comment thread; we’ll exclude you. If you’re using the Veloura demo and want to talk to a human instead of the agent, email hello@forgingapps.com with subject [VELOURA-HUMAN] and we’ll route the next message manually. We respond within one business day.
Legal basis and audit logs
The EU AI Act, Article 50, requires us to make clear when you’re interacting with an AI system; Meta’s Platform AI Disclosure policy requires the same on Facebook. This page, plus the in-context AI-operated labels, is how we comply with both. Audit logs of every agent action are retained and reviewable; we keep them long enough to investigate incidents and respond to user requests. The kill-switch documented above applies system-wide: Ivaylo can pause every agent in the fleet at once, and reverse any in-flight action that hasn’t been published. Last updated 3 May 2026.