Agent Safety Control Tower

Runtime health, eval gates, prompt rollout, tool governance, action approvals, and audit-ready evidence.

AI autonomy policyrecommend_and_prepare mode keeps risky actions approval-gated. Human approval is required for destructive, access-changing, remediation, broadcast, and license-reclaim actions.
Production eval history and governance analytics0 eval runs

Prompt drift monitoring

0

prompt drift signals

Model comparison

0

provider/model histories

Budget burn

$0.00

0% of local budget

Tool risk analytics

0 high risk

Policy simulation

No policy simulations recorded yet.

AgentPromptModelPass rateGateReadinessFinished
No eval history found.
AI governance evidence package

Package state

not set

not set

Audit events

0

local evidence samples

AI action audit evidence export

AI actions

0

approval queue records

Tool calls

0

trace step evidence

Approvals

0

approved actions

Denials

0

rejected actions

AI runtime/governance 5.0 acceptance evidenceScore candidate

Target score

5

domain target

Tasks

97, 98, 99, 100

completed locally

Routes

0

acceptance surface

Agent registry and eval gates0 agents
Autonomy policy

Allowed autonomous actions

Human approval required

Provider configuration

No tenant AI provider integration found

AI action approval queue0 records
Eval datasets0 datasets
Prompt registry0 prompts
Recent eval runs and prompt rollouts0 records
TypeAgentPromptGate/statusRecorded
No eval or rollout records yet.
MCP tool governance0 tools