Skip to content

Agent Authority

Define what agents may do, what evidence they need and when a person must approve. Run hosted coding tasks or connect the agents you already use.

Start free View pricing

Free records your policy decisions. Paid plans enable enforcement on supported paths. You connect your model provider and pay its usage charges separately.

Run agents. Protect the tools they use.

Both use your policies and decision records. Coverage depends on the integration.

Atellagent Code

Work on GitHub repositories in isolated workspaces, with enforced limits on file, network and credential access.

Explore Atellagent Code

Enterprise MCP

Protect MCP servers you operate. Keep native authentication and require central authorization before direct clients execute a tool.

Protect your MCP servers

Write policies in plain language.

The assistant proposes changes from supported templates. You review and validate them before saving.

  1. Describe the rule

    Explain what should be allowed or blocked and where the rule applies.

  2. Review the draft

    Check where each rule applies, what it changes, and any assumptions. Edit the draft as needed.

  3. Validate changes

    Check that the draft uses supported rules and available targets.

  4. Save reviewed changes

    Saving applies the changes. Creating and validating a draft leave existing controls unchanged.

Choose the controls you need.

Combine action rules, risk review and data protection. Deterministic policy makes the final decision.

Set action rules

Allow, block or limit an action using identity, current task state and required prior results. Eligible hosted actions can pause for approval.

Review action risks

Semantic Review uses AI to assess concerns you describe. In Enforce mode, policy can block matching external actions or pause eligible hosted work for review.

Check data before sharing

Detect personal information and secrets, or bring your own classifier. Observe records findings; Enforce applies controls on configured data-sharing paths.

See data protection

See a stateful policy in action.

The same action can be allowed or blocked based on recorded results from earlier in the task.

Policy example

Production database maintenance

Requested action
Delete archived records
Policy requirement

Backup creation and restore verification must succeed during this task.

Required evidence
  • Create backupSucceeded
  • Verify restoreMissing successful result
Decision
Blocked

Missing evidence: successful restore verification.

Atellagent checks recorded tool outcomes whenever the protected action is requested.

Manage controls and investigate decisions.

Policies

Create rules, configure data checks and AI risk review, and inspect policy decisions.

Visibility

Investigate alerts, action history, audit records and recorded results.

Organization

Manage team access, agent connections and repositories.

Choose how to get started.

Start free with Atellagent Code or connect the agents you already use.

Start free View integrations