Define
Author policies in natural language
Select a model and reasoning level, describe the intent, and receive a constrained proposal based on approved templates.
Govern every AI action.
Agent Authority is Atellagent’s policy and enforcement platform. Define policy in natural language, review a concrete proposal, and enforce approved controls before agents act.
- Skip the integration and get right to value with hosted runtimes.
- Integrate external runtimes with Atellagent Client.
- Keep decisions, evidence, and resulting outcomes connected.
The control loop
Define, enforce, protect, and prove.
Atellagent makes governance operational: policy is understandable to its owners, decisions happen before execution, and evidence remains usable afterwards.
Enforce
Decide actions before side effects
Deterministic policy remains authoritative. Semantic Action Review can add bounded risk evidence and route an action to accountable human review.
Protect
Control data release
Inspect data at governed egress boundaries with built-in detectors and your own filter or classifier.
Prove
Review decisions with evidence
Connect the request, governing policy, semantic assessment, decision, and outcome for investigation and continuous improvement.
Two governed paths
Start hosted. Extend to the runtimes you already operate.
Hosted workspaces deliver the shortest path to governed coding. Atellagent Client extends the same policy and evidence model into customer-operated runtimes, including supported local agent hooks. Coverage is made explicit per integration.
Hosted workspaces
No runtime integration to get started
Run coding workflows in Atellagent’s hosted environment, with governed execution built into the workspace.
See hosted executionExternal runtimes
Discover, then govern
Use Atellagent Client with customer-operated runtimes, including supported Codex, Claude Code, and Gemini CLI hooks. Make the action surface visible first, then move the configured host boundary to a strict central control posture.
See coverageYour policy
Keep governance consistent as you expand
Use the same policy lifecycle and review model across the participating paths that your environment supports.
Read the architectureStart with a real workload
Put governance around work your engineering team already needs to do.
Code changes
Review high-impact repository actions
Require approval or narrow allowed targets before a coding workflow reaches a governed side effect.
Outbound research
Protect findings before release
Inspect data before it is delivered to a model, tool, channel, or service boundary.
Operations
Make sensitive agent actions reviewable
Attach a policy decision and outcome record to work that would otherwise disappear into disconnected logs.
Policy rollout
Observe before you enforce
Learn a policy’s practical effect, refine it, then turn on enforcement where supported.