Govern · Agent Identity
Available nowGive agents an identity of their own
Halo keeps agent identity separate from user identity, credentials, tools, and action scope. Agents act under their own identity, so access can be scoped, decisions can be attributed, and operators can contain an agent without touching a person's account.
- single sign-on for people
- OIDC
- enrollment required before account access
- MFA
Why it matters
Shared accounts hide who acted
When an agent runs under a person's login or a shared API key, the record says a person did something that software actually did. Reviewers cannot tell who acted, and revoking access means changing a person's access as well.
- 01
Agents that borrow a user's credentials inherit access that was never meant for them.
- 02
A shared key gives agents the same scope no matter what each one is meant to do.
- 03
Without a separate identity, tool permissions and audit records cannot point to a single agent.
How it works
From creation to attribution
Identity is set when an agent is created and carried through the stages of its work.
- 1
Create the agent
An agent is created with its own name, purpose, environment, and policy bundle, and gateways attribute traffic to that agent.
- 2
Scope the access
Choose the tools, apps, models, and destinations the agent needs, and bind its credentials at the gateway.
- 3
Attribute the actions
Gateways identify the calling agent on requests that pass through them, so decisions, approvals, costs, and evidence name the agent.
- 4
Contain or retire
Quarantine an agent to suspend its access during an investigation, release it when the issue is resolved, or stop and delete it when the work ends.
Capabilities
Identity that follows the agent
Gateways, policy, approvals, and evidence work from the agent's identity, so limits and records attach to the right actor.
Agent-specific identity
An agent acts under an identity that is separate from the people who built or use it, so records name the agent.
Scoped tools and actions
Policy conditions can reference the agent and its tags, so tool and action permissions attach to the agent itself.
Brokered credentials
Provider and app credentials stay bound at the gateway. Agents receive scoped capabilities and never see raw keys.
People and roles
Account admins and members have different access, and OIDC single sign-on connects Halo to your identity provider.
Entities, groups, and aliases
Model people and services as entities, collect them in groups, and map external identities such as OIDC subjects to them with aliases.
Quarantine and release
Operators can quarantine an agent to suspend its access without changing any person's account, then release it after review.
Clear attribution
Records that name the agent
Because an agent has its own identity, policy decisions, approvals, costs, and evidence point to the agent that acted. Reviewers can see what the agent was allowed to do and which revision was running when it acted.
- Decisions and costs attributed to one agent
- Policy conditions can match agent ID and tags
- Revisions show which configuration was running
- Quarantine leaves people's accounts untouched
FAQ
Questions, answered
A user identity represents a person who signs in to Halo. An agent identity represents the agent itself, so its tool access, credentials, and actions are scoped and recorded separately from the people who created or use it.
Agents act under their own identity with scoped capabilities. Provider and app credentials stay bound at the Halo gateways, and agents never see raw keys. Credential values are write-only, so they cannot be read back after you save them.
People can sign in through OIDC single sign-on with a standards-compliant identity provider, and Halo requires users to complete multi-factor authentication enrollment before they can use account features. Roles separate account admins from members.
Yes. Quarantine suspends one agent's access pending review, and release returns it to active deployment. Each agent runs as its own workload, so containing one agent does not stop another. Both actions are recorded in the audit log.
Audit records and gateway decisions name the agent that acted, the policy that applied, and the outcome. That lets a reviewer answer questions about one agent without sorting through activity from the people around it.
Put a halo around the AI your business runs.
See discovery, runtime policy, approvals, and evidence working together on your own use cases.