Skip to content
Halo by OmniTrust

Optimize · Halo Agents

Available now· Self-serve plans from $99 per month

Launch secure AI agents with controls built in

Choose a model, connect apps, set policy and approvals, and add a budget in one setup. Halo Agents run in a secure runtime, and Halo Chat gives people a place to work with them.

Book a walkthrough Start with Halo
per month for the Individual plan
$99
self-serve plans before Enterprise
3
apps in the connection catalog
1,000+

Why it matters

Launching a useful agent takes too many steps

Picking a model, wiring up apps, setting permissions, adding approvals, and capping spend usually happen in separate tools. The agent that results is hard to review because nobody configured it as one thing.

  1. 01

    Owners of growing businesses want agents working soon, and they still need guardrails.

  2. 02

    Permissions, approvals, and budgets configured separately drift apart over time.

  3. 03

    People need an easy place to talk with an agent and see what it did.

Capabilities

One setup flow for the whole agent

The setup flow covers the decisions that usually live in separate tools, so the agent you launch is the agent you reviewed.

  • Model choice

    Pick a Halo-managed model with no credential to set up, or bring your own key for OpenAI, Anthropic, Google Gemini, Cerebras, or GMI Cloud.

  • App connections

    Connect business apps from the Halo connection catalog of 1,000+ apps, with available actions that vary by app.

  • Policy and approvals

    Assign a policy bundle and approval rules at launch, so higher-impact actions wait for a person from the first run.

  • Budget at launch

    Set an agent budget with thresholds and automatic actions as part of setup, so spend has a limit from the start.

  • Secure runtime

    Agents run in the Halo secure runtime as sandboxed, non-root workloads with default-deny network egress, and operators can quarantine them.

  • Halo Chat

    Halo Chat gives people a place to talk with an agent and see its work, with policy and approvals applied to the actions the agent takes.

How it works

Four decisions, one setup flow

The flow moves from purpose to launch in order.

  1. 1

    Describe the job

    Name the agent, state its business purpose, and choose an environment and a model mode, either Halo-managed or BYOK.

  2. 2

    Set the guardrails

    Choose an enforcement mode, a risk posture, and an approval policy, and set runtime limits such as run time and concurrency.

  3. 3

    Connect apps and assign policy

    Select the apps the agent may use, bind credentials at the gateway, and assign the policy bundle that governs its actions.

  4. 4

    Review and deploy

    Review the full configuration, deploy, and talk with the agent in Halo Chat while operators watch health, cost, and approvals.

Example agent

A weekly brief with limits built in

For example, an operations team might launch an agent that gathers approved metrics, summarizes the week, and drafts a brief for review. Halo lets that team scope the app connections, require approval before anything leaves the company, and cap spend.

  • Connect only the apps the agent needs
  • Hold external messages for approval
  • Cap spend with an agent budget
  • Review actions in the evidence trail

FAQ

Questions, answered

Halo Agents lets you launch AI agents with model choice, app connections, policy, approvals, and budget configured together. Agents run in the Halo secure runtime, and Halo Chat gives people a place to work with them.

Individual is $99 per month, Professional is $299 per month, and Team is $1,299 per month. Enterprise is custom, with annual contracts from $39,999 per year. The pricing page lists what each plan includes.

Agents can use Halo-managed models, which run through Amazon Bedrock and are billed through Halo credits, or your own provider key. BYOK providers include OpenAI, Anthropic, Google Gemini, Cerebras, and GMI Cloud, and other providers use guided setup.

Through app connections. The Halo connection catalog lists 1,000+ apps, and available actions vary by app. Credentials stay bound at the gateway, agents receive scoped capabilities, and policy and approvals apply to the calls they make.

Yes. Higher-impact actions can pause for a person to approve or reject. You choose which actions need approval, and the decision is recorded with the agent, the action, and the target.

Related capabilities

App Connections1,000+ apps in the connection catalogLearn more about App ConnectionsSecure Agent RuntimeSandboxed agents with default-deny egressLearn more about Secure Agent RuntimeBudgets and Cost ControlsBudgets that notify, pause, or stopLearn more about Budgets and Cost Controls

Put a halo around the AI your business runs.

See discovery, runtime policy, approvals, and evidence working together on your own use cases.

Book a walkthrough Start with Halo