Halo AI governance platform
Govern AI where the work happens.
Halo finds the AI your people and agents use, applies your policy live on laptops, in browsers, at the model and tool gateway, inside agent runtimes, and records each decision so reviewers can see what happened and why.
Discover, control, prove
One platform from inventory to evidence.
Halo connects what your teams use with the policy that governs it and the record that shows it worked.
Know the AI in use.
Halo Endpoint and Halo Browser inventory AI apps, extensions, developer tools, local runtimes, and model artifacts on macOS and Windows devices and in Chrome and Edge, plus Safari on macOS. Each finding records its source and confidence, and the inventory exports as a CycloneDX AI bill of materials.
Apply policy where AI runs.
Write policies once as code, and Halo compiles them for each enforcement point. Run a policy in audit mode first, then enforce it on model, tool, and MCP traffic. Hold high-impact agent actions for a person to approve, block unapproved AI apps on managed devices, and check prompts, pastes, and uploads on supported AI sites.
Govern model and tool access.
Agents reach models through the Halo model gateway and tools through the MCP and tool gateway. Both share one policy engine, so identity, policy, budgets, and evidence apply to each call, with Halo-managed models or your own keys. Routing each workload to the right approved model creates savings potential.
Run agents inside the guardrails.
Halo agents run as non-root workloads with default-deny networking, and their credentials stay at the gateway instead of inside the agent. Operators can quarantine an agent at any time. Halo supports several agent frameworks and a Python SDK.
Show reviewers what happened.
Each gateway decision records the policy, the outcome, and the reason. Audit logs and gateway decisions export as CSV or JSON, and administrative changes are logged.
Keep AI spending predictable.
Set account and per-agent budgets that the gateway enforces, track cost by agent and model, and see what provider caching saved.
Supported surfaces
Where Halo discovers and enforces today.
Discovery tells you what is in use. Enforcement changes what can happen. This table separates the two for each surface.
| Surface | What Halo discovers | What Halo enforces | Availability |
|---|---|---|---|
| macOS endpoint | AI apps, IDE tools, MCP configuration, local runtimes, and model artifacts | Audits or blocks launches of unapproved AI apps | Guided rollout |
| Windows endpoint | AI apps, IDE tools, MCP configuration, local runtimes, model artifacts, and Office AI add-ins | Audits or blocks launches of unapproved AI apps | Guided rollout |
| Chrome and Edge | Use of catalogued AI sites and installed AI extensions | Checks prompts, pastes, and uploads on supported AI sites | Guided rollout |
| Safari on macOS | Use of catalogued AI sites | Checks prompts, pastes, and uploads on supported AI sites | Guided rollout |
| Model gateway | Model calls by agent, model, and cost | Policy, approvals, and budgets on each model call | Available now |
| Tool and MCP gateway | Tool and MCP calls by agent | Policy, approvals, and gateway-held credentials on each call | Available now |
| Halo agent runtime | Agents, frameworks, versions, and health | Default-deny networking, non-root execution, and quarantine | Available now |
- Guided rollout means our team helps you package and deploy to managed devices.
- Linux devices and Firefox are not supported yet.
FAQ
Questions buyers ask us.
It connects AI inventory, accountability, policies, controls, and evidence so an organization can make and review decisions about AI use.
Governance defines responsibilities and acceptable behavior. Security controls help enforce those decisions and provide evidence about threats and actions.
Halo collects metadata about AI apps, extensions, runtimes, and configuration. Standard collection does not capture prompts, responses, documents, or credentials. The Browser and Endpoint Privacy Notices describe the details.
On macOS and Windows devices, in Chrome and Edge, plus Safari on macOS, at the Halo model and tool gateways, and inside the Halo agent runtime. Linux devices and Firefox are not supported yet.
Review the coverage table above, or bring your environment to a walkthrough so our team can confirm the supported discovery and control paths.
Software supports governance work and evidence. Your organization remains responsible for its decisions, processes, and applicable requirements.
Details on data collection are in the Browser Privacy Notice and the Endpoint Privacy Notice.
Bring your AI stack to a walkthrough.
We will map the AI your teams use to the surfaces Halo covers and show policy, approvals, and evidence working end to end.