Approach comparison
Halo vs AI gateways
An AI gateway governs the traffic routed through it. Halo includes model and tool gateways and adds what a gateway cannot see on its own, with discovery on browsers and endpoints, a sandboxed agent runtime, policy as code, budgets, and evidence.
The short version
- An AI gateway sits in the request path between applications or agents and AI services, and it controls the traffic that passes through it.
- Halo includes model, MCP and tool, and API gateways, and the Halo AI Gateway extends the same controls to Claude Desktop, Claude Code, ChatGPT Desktop, and Codex CLI.
- Halo also finds AI use that never reaches a gateway, runs agents in a sandbox, and records decisions, approvals, and spend in one place.
What AI gateways do
An AI gateway mediates traffic between applications or agents and AI services. Depending on the product, it can provide a single endpoint, routing and fallback, key management, rate limits, logging, and selected guardrails.
MCP gateways apply the same idea to tool calls. Gateways vary widely, with some focused on routing and cost for application developers and others adding policy, approvals, and inspection for agent traffic.
How Halo uses gateways
Halo runs a model gateway, an MCP and tool gateway, and an API and dispatch gateway as separate components. They apply identity, policy, approvals, budgets, and evidence to agent traffic, and provider and app credentials stay bound at the gateway.
The Halo AI Gateway extends the same controls to Claude Desktop, Claude Code, ChatGPT Desktop, and Codex CLI, and access for your own applications is in Preview. Halo adds discovery on browsers and endpoints, a secure agent runtime, and exportable audit logs around the gateways.
Side by side
How Halo and AI gateways compare
| Dimension | Halo | AI gateways |
|---|---|---|
| Primary job | Halo Available nowDiscover AI in use, enforce policy where it runs, and keep evidence across browsers, endpoints, gateways, and agent runtimes. | AI gateways Mediate requests between applications or agents and AI services, with routing, keys, and selected controls. |
| What it sees | Halo Available nowBrowser and endpoint discovery see AI use that never crosses a gateway, and gateways see the traffic routed through them. | AI gateways The traffic that is routed through the gateway. |
| Desktop AI apps and coding agents | Halo Available nowThe Halo AI Gateway connects Claude Desktop, Claude Code, ChatGPT Desktop, and Codex CLI to models and tools that Halo authorizes, meters, and records. On the roadmap Policy for the local tool calls that coding agents make on a developer’s machine. | AI gateways The applications and agents configured to send traffic through the gateway. |
| Agent execution | Halo Available nowA secure agent runtime with sandboxing, default-deny network egress, credential brokering, and quarantine and release controls. | AI gateways Gateways typically mediate requests, and the runtime that executes the agent is a separate layer. |
| Policy and approvals | Halo Available nowVersioned policy bundles with audit and enforce modes, simulation with a decision trace, decision explanations, and human approval for higher-impact actions. | AI gateways Varies by product, from key management and rate limits to guardrails and approvals. |
| Cost and routing | Halo Available nowAccount, agent, and per-user budgets with automatic actions, cost analytics by agent and model, Halo Auto, and route profiles with fallback paths. | AI gateways Routing, fallback, caching, and spend tracking are common gateway features. |
| Evidence | Halo Available nowDecisions, approvals, denials, costs, and configuration changes are recorded and export as CSV and JSON. Coming soon SIEM export, signed webhooks, and API tokens for your own automation. | AI gateways Request logs are common, and the depth of evidence varies by product. |
| Your own applications | Halo PreviewThe AI gateway for your own applications applies the same identity, policy, budgets, and evidence as the other Halo gateways. | AI gateways Many gateways are built for developers who route their own application traffic to model providers. |
| Model providers | Halo Available nowHalo-managed models and Halo Auto, plus bring your own key for OpenAI, Anthropic, Google Gemini, Cerebras, GMI Cloud, and 50+ providers your admins connect themselves. 100+ providers are available through guided setup. | AI gateways Many gateways list support for a wide set of model providers behind one endpoint. |
Why Halo
Why choose Halo over AI gateways
Simple, secure, and efficient AI, governed where it runs.
Discovery before traffic
Halo Browser and Halo Endpoint find AI websites, extensions, apps, and local runtimes that never send traffic through a gateway, with metadata-first collection.
Execution boundaries for agents
Agents that Halo runs use a sandboxed runtime with default-deny network egress, explicit credential boundaries, and quarantine and release controls.
One record for policy, budgets, and evidence
Decisions, approvals, denials, costs, and configuration changes are recorded together, and audit logs export as CSV and JSON.
Credentials agents never see
Provider and app credentials stay bound at the Halo gateway. Agents receive scoped capabilities and never hold raw keys.
Budgets that act
Account, agent, and per-user budgets can notify owners, pause new operations, or stop active deployments, and Halo Auto picks a model for each request from a curated pool.
The AI apps your people already use
The Halo AI Gateway connects Claude Desktop, Claude Code, ChatGPT Desktop, and Codex CLI to models and tools that Halo authorizes, meters, and records.
Roadmap
What Halo is building next
Each of these extends the controls Halo runs today.
Capabilities marked coming soon or on the roadmap reflect current plans and may change.
- Coming soon
SIEM export and webhooks
Send decisions, approvals, and inventory events to your SIEM and alerting tools, with API tokens for your own automation.
- Coming soon
Tamper-evident evidence
Hash-chained decision and audit records with retention, legal hold, and a verifier your auditors can run.
- On the roadmap
Local tool coverage for coding agents
Extend AI Gateway policy to the local tool calls that coding agents such as Claude Code and Codex make on a developer’s machine.
- On the roadmap
Executive value reporting
Spend, adoption, and routing savings by team, cost center, model, and AI system, in views built for finance and executives.
Buyer checklist
Questions to ask in your evaluation
Bring these to each vendor you evaluate. Here is how Halo answers them.
Which AI traffic reaches the gateway, and how do you find the traffic that does not?
How Halo answers
Halo gateways cover agent and AI app traffic, and Halo Browser and Halo Endpoint find AI use that never reaches a gateway.
Where do provider keys and app credentials live, and can an agent ever read them?
How Halo answers
Credentials stay bound at the Halo gateway. Agents receive scoped capabilities and never see raw keys.
Can an agent route around the controls?
How Halo answers
Agents that Halo runs have default-deny network egress, so their model and tool traffic goes through the Halo gateways.
Which evidence can reviewers export, and in what formats?
How Halo answers
Audit logs and gateway decisions export as CSV and JSON, and each decision records the policy, the outcome, and the reason.
Can budgets stop spend, or only report it?
How Halo answers
Halo budgets can notify owners, pause new operations, or stop active deployments at the account, agent, and user level.
Sources and review
Last reviewed October 19, 2026. Information about AI gateways comes from public materials and may change. Product names belong to their owners. Capabilities marked coming soon or on the roadmap reflect current plans and may change. If you spot something out of date, please let us know.
Put a halo around the AI your business runs.
See discovery, runtime policy, approvals, and evidence working together on your own use cases.