AgentVision360 · Early access
Know your AI spend, and who holds the keys
See AI spend from seven providers by team and model, catch idle seats and unowned keys, and govern every key and agent behind it, all read-only unless you turn on approved actions.
- 7 providers
- one view of AI spend
- month-end
- spend forecast per team
- every key
- with an owner & last use
What AgentVision360 covers
Spend
- Daily cost from OpenAI, Anthropic, Azure OpenAI / Foundry, Amazon Bedrock, Google Vertex AI, Microsoft 365 Copilot and GitHub Copilot
- Explore by provider, account, model, project, key or team, save the view and export to CSV
- Month to date and a month-end forecast at the recent pace
- Unusual days flagged per project, with the key and model behind them
Budgets and chargeback
- Budgets for all AI, a provider, an account or a team, with alert thresholds
- Warnings when the forecast will pass a budget, before it does
- Teams by rules on projects, keys, models or whole accounts; unallocated spend shown
Paid seats
- Microsoft 365 Copilot and GitHub Copilot seats bought, assigned and used
- Idle seats and what they cost each month
- Concealed Microsoft 365 report names stay concealed
Keys and access
- Every API key, admin key and service credential in one register
- Keys whose owner has left, unused keys, admin keys, unrestricted keys and keys without an expiry
- Last use and 30-day spend per key where the provider reports it
- Optional approved actions, off by default: deactivate or delete a key, remove an idle seat, after other people approve
Agent and identity governance
- Non-human identity posture: every key, service credential and agent scored as traceable or shadow, with standing privilege, no-expiry and per-provider posture, as a governance score and grade
- Autonomy control tiers: a recommended tier — Restricted, Approval or Standard — for each identity, from its blast radius, standing privilege and traceability
- Identity lifecycle: the owner, scope, expiry and usage an identity should have been created with, plus an idle, stale and deprovision-due worklist
- Revocation impact: what a key or seat is used for — recent spend, models and projects — checked before a kill-switch revokes it
- Recommended actions ranked by severity, every change under approval and recorded in the audit trail
For service providers
- Customer organizations, each with its own people, connections, spend and audit trail
- Add a customer and their first administrator; the one-time password is shown once
- Open a customer to help, after an authenticator check, with every visit and change in their audit trail
- Provider staff can't manage a customer's people or request or approve changes at their AI providers
Security and audit
- Reads only, unless approved actions are turned on; credentials encrypted and never shown again
- Mandatory MFA, roles for owners, admins, FinOps, security and viewers
- Hash-chained audit trail that shows whether any record was changed
How it works
Connect
Add each provider account with a read-only role or an admin key. The first read brings in up to 90 days of history.
Allocate
Map projects, keys and accounts to teams, and set budgets with alert thresholds.
Watch
AgentVision360 re-reads every few hours, forecasts the month, flags unusual days and idle seats, and checks every key.
Act
Alerts say what changed and where to fix it at the provider. If your organization turns on actions, a key or seat can be changed from AgentVision360 after other people approve it.
Works with
- OpenAI
- Anthropic
- Azure OpenAI / Foundry
- Amazon Bedrock
- Google Vertex AI
- Microsoft 365 Copilot
- GitHub Copilot
- Microsoft Entra ID
Deployment
- Docker Compose on a single Linux VM, alongside Cloud governance and ITOps360
- PostgreSQL with row-level security per organization
On the roadmap
- Project spend limits
Check where you stand
Free, about five minutes, scored in your browser. Nothing is sent unless you ask for the report.
AI Governance Assessment
How well your organization governs its use of AI. 8 questions, about 5 minutes.
AI Agent Risk Assessment
Risk from autonomous agents and AI-driven automation. 7 questions, about 4 minutes.
AI Security Assessment
Protection against prompt injection, data leakage and tool abuse. 8 questions, about 5 minutes.
Questions people ask
Can we use AgentVision360 today?
Yes, in early access. Book a live demo, or ask for a 30-day trial on your own AI accounts. Plans are on the pricing page.
Can we offer AgentVision360 to our own customers?
Yes. Service providers add each customer as its own organization from the Customers page. The customer's administrator connects their AI accounts and invites their people; you can open the customer to help, and each visit is recorded in their audit trail.
Does AgentVision360 see our prompts or responses?
No. It reads billing, usage counts, keys and seat assignments. It never receives prompt or response content, and it sends nothing to an AI model.
Can AgentVision360 change anything in our AI accounts?
Not by default: it only reads. Actions such as deactivating a key or removing an idle seat are off unless your organization turns them on and adds a separate write credential, and each one runs only after other people approve it. OpenAI and Anthropic offer no read-only admin key, so use a dedicated one for AgentVision360; credentials are encrypted at rest and never shown again.
How do you work out spend per key?
From the provider's own cost reports. Where a provider reports cost per workspace and model but not per key, AgentVision360 splits it by each key's token share and marks those figures as estimated.
See it on your own estate
Walk through your clouds, identities and devices with a specialist in 30 minutes, or get a scored report from a free assessment.