Hosted
Governed coding workspaces
Atellagent owns the execution boundary, so teams can start coding work with policy and evidence already on the path and no runtime integration step.
Agent Authority
One control plane for agents you host and agents you already run. Atellagent brings policy, action controls, data-release protection, and decision evidence together for governed AI execution.
Execution models
Use Agent Authority where the work runs.
Start in a hosted workspace or integrate a customer-operated runtime. Both paths use the same policy lifecycle and preserve evidence with the resulting decision.
External
Customer-operated runtimes
Atellagent Client integrates your own runtime through supported SDK, bridge, adapter, and host-hook paths. It makes the action surface visible before a team adopts a strict central control posture; each integration documents its exact enforcement boundary.
Shared
One policy lifecycle
Define, review, observe, and enforce policy in one system while retaining decision evidence across the participating paths in your environment.
Natural-language policy authoring
Turn intent into a reviewable policy proposal.
Ask for the outcome you need. The policy authoring assistant uses approved policy templates to propose structured policy changes that a human reviews, validates, and publishes.
Describe the rule
Choose a model and reasoning level, then explain the control in plain language.
Review the proposal
See the chosen template, scope, effect, and assumptions before the policy changes.
Validate and test
Check the proposal against the policy system and expected decision outcomes.
Publish deliberately
A human publishes the final policy; the authoring assistant is not a policy publisher.
Controls at the point of consequence
Govern the action and the release.
Use deterministic controls where an agent attempts a side effect, semantic evidence where context warrants review, and data protection at governed egress boundaries.
Actions
Allow, deny, narrow, or require approval
Evaluate a specific request with identity, context, approval state, and deterministic policy before it proceeds through a governed route.
Semantic Action Review
Escalate meaning and context to accountable people
Assess a proposed action for bounded risk evidence. Policy can require human review for an elevated or incomplete assessment; the assessment does not authorize, deny, or execute the action.
Data release
Inspect data before it leaves
Use built-in PII and secret detection, your own filter or classifier, and observe or enforce modes at governed egress boundaries.
Explore Agent DLPOperate and improve
Make the control system usable by the people who own it.
Policies
Manage rules, data, roles, and decisions
Author and review policy in a dedicated management surface.
Visibility
Investigate control outcomes
Review decision, audit, monitoring, and side-effect information together.
Organization
Manage identities and connections
Operate users, service accounts, provider access, and runtime inventory in one place.
Start with a governed workspace.
Use the hosted path for immediate value, then extend policy coverage to the supported external agents in your environment.