Onyx AIOps | see it, govern it, prove it
Every AI call your organisation makes — monitored, policy-enforced, and receipted in a ledger you can hand to an auditor. Works with the providers you already use and the hardware you already own.
One console for every model you run
OpenAI, Anthropic, Gemini, open-weight models on your own hardware — one view. Usage, cost, latency, failures, and which agent made which call, by name.
- Per-provider, per-model, per-user and per-agent metrics
- Cost and token accounting that matches the invoice
- Fleet health for self-hosted models — DGX Spark, Mac, GPU workstation
- Alerts on drift, spend, and failure rates
Policy enforced at the call, not in a PDF
Identity-bound rules decide who may call which model, with what data, through which tools. Enforcement happens in the request path — before the incident, not after.
- Identity-aware routing and allow/deny per user, team, or agent
- Tool- and MCP-call capture with approval gates
- Data-boundary rules: what may leave the building, and to whom
- Model promotion gated on benchmark evidence, not vibes
A ledger you can hand to the auditor
Every call is receipted: what was asked, what was answered, under which policy, from which evidence, by which model and version. Hash-chained, signed, exportable.
- Ed25519-signed receipts, Merkle-batched for independent verification
- Decision audit — who approved, under which policy, with what evidence
- Replayable history for disputes, regulators, and post-mortems
- Export to your GRC platform or a plain, signed archive
Your data stays where you decide
Same product, three postures. Start hosted, move the collector on-prem when compliance asks, go fully sovereign when the mandate does.
Hosted
DefaultWe run the console and the collector. Point your apps at one endpoint. Minutes to first receipt.
Split-plane
RegulatedHosted console; the collector runs on your hardware or VPC. Prompts and documents never leave your network — only hashes, metadata, and receipts do.
Appliance
SovereignEverything on a box you own — DGX Spark or Mac Studio — including local models. Air-gapped if you need it. Delivered with partners.
People who have to show their work
Regulated operators
Financial services, GRC, healthcare, energy. You already answer to a regulator; now your AI can too.
Teams shipping copilots
Your product cites codes, regulations, or standards. When a customer or inspector asks why, you have the receipt.
Multi-model shops
Three providers, a dozen agents, no single view. One console, one policy, one ledger.
What is Onyx AIOps?
Onyx AIOps is an AI assurance platform: monitoring, governance, and a tamper-evident audit ledger for every AI call your organisation makes — across cloud providers and self-hosted models. See it, govern it, prove it.
Do I have to run models on my own hardware?
No. Most customers call OpenAI, Anthropic, or Gemini and simply want the audit trail and policy enforcement. Self-hosted models are supported, not required.
Does my data go through Onyx?
In hosted mode, requests transit our collector and are not retained beyond what the receipt requires. In split-plane and appliance modes, the collector runs on your side and the hosted console only ever sees hashes, metadata, and receipts.
Is this a Tailscale product?
No. Onyx AIOps is built and operated by Onyx AI Labs and does not depend on any VPN or identity vendor.
What does EARLY ACCESS mean?
It runs our own fleet today and is being deployed with design partners. If you want to be one, start a working session.
Early access is open to a handful of design partners.
Bring a real workload. We will bring the receipts.