Coordinate Claude, Gemini, ChatGPT, ElevenLabs, and your own self-hosted agents under one auditable approval boundary. Your secrets stay in a local AES-256-GCM vault. No agent ever touches a raw key.
A live multi-agent message bus where 15 actor roles, 9 message intents and 7 status states make collaboration real, not a chat illusion. Every cross-agent handoff is rendered on a canvas with directional arrows and audit-grade timestamps.
A local-first knowledge graph with 24 node types and explicit edge relations. Bring your Obsidian vault, your provider receipts, your subscription dashboard, your travel plans — Operator builds a queryable map you actually own.
Drive gcloud against pre-staged scripts through IAP tunneling with risk-tiered approvals. Inspect, restart, or copy back from a remote VM — without ever exposing a service-account key to an agent or to a log line.
Service-account JWT exchange, gemini-2.5-flash responding through Operator's audited Live Connector.
Hermes 3 Llama 3.1 8B fp16 on vLLM with tool calling, served from an NVIDIA RTX PRO 6000 vGPU.
Real cross-agent dispatch via the Theater Bridge. No mock interactions, no synthetic replies.
OpenAI-compatible proxy installed, firewall-scoped to the founder IP. Awaiting upstream auth handshake.
Read-only Google Drive scope to index existing vault notes as graph nodes.
Manual browser profile orchestration with explicit human-in-the-loop. No credential capture, ever.
Operator is built by one developer in Turkey, dogfooded daily, and not yet open to the public. If you're a solo operator who burns through five-to-ten AI subscriptions and wants the productivity of an agent fleet without the trust assumptions of hosted platforms, get in touch.
hello@operatorhive.com