Line drawing: a call passes through a gate and leaves on one of three lanes, allowed, held or blocked.
Decide before execution
Every tool call is checked against your policy, then allowed, held or blocked.
Govern AI agents
before they act.
Aethis is an AI-agent governance platform. Decide which model and tool calls can proceed, need human approval, or must stop.
Agents act fast. Aethis decides what proceeds. Rules for routine work, a person for the calls that matter, and a record of every decision.
Line drawing: a held call waits in a bay until a single token, the reviewer’s approval, drops into it.
A person when it matters
Held calls wait for a reviewer. Approve once and the call runs.
Line drawing: four stacked plates, request, policy, decision and outcome, joined by one thread, the Action Record.
Proof afterwards
One Action Record links the request, the matched policy, the human decision and the outcome.
Between your agent and its tools.
Every call is checked on the way through.
Point your MCP-compatible agent at an Aethis MCP server URL on mcp.aethis.io. Aethis governs each call and forwards it to your existing MCP server.
Diagram: calls travel from the Agent through Aethis to the MCP server and the system behind it. At Aethis each call is allowed, held until a person approves it, or blocked.
- Allowed Passes straight through.
- Held Waits until a person approves.
- Blocked Stops at Aethis.
MCP client config
"finance": {
-"url": "https://finance.internal/mcp",
+"url": "https://mcp.aethis.io/v1/agents/{agent_id}/mcp",
"headers": {
"Authorization": "Bearer <aethis_agent_token>"
}
}
Controls that stay out of the way.
Until a call needs one.
Rules you can read
A policy rule is one sentence: which agent, which tool, and when a person has to approve.
Schematic app window: rows of pills for the parts of a rule, the first row ending in an amber pill for the approval.
Approvals with a deadline
A held call waits for a reviewer and expires if nobody decides in time.
Schematic app window: a request with a time bar partly filled in amber and two buttons.
Action Records
Each call leaves a record: the policy decision, the human decision, what ran.
Schematic app window: a list of records, each with three small marks; the first row is picked out in green.
Spend you can see
Track each agent against its monthly budget and get alerted before it runs over.
Schematic app window: bars of spend climbing to a dashed budget line; the last bar is amber and crosses it.
Tool changes
When an MCP server adds or changes a tool, Aethis raises a finding for review.
Schematic app window: a grid of tool tiles; one has a dashed amber outline because it is new.
Observe Mode
New agents start in Observe Mode: decisions are recorded, calls are not stopped.
Schematic app window: a switch turned on above a list of calls, each marked as seen and none stopped.
03 / After the requestFollow the action.
Follow the action.
Understand the decision.
One Action Record connects the exact request, matched policy, human review, and execution state, so you can see what was decided and what actually happened.
Inside an Action RecordIllustrative Aethis console with demo data. The Activity timeline lists what agents did. One event is picked out: Billing Agent requested approval for finance.refund.create, SGD 2,480.00. Policy: approval required. Approval: pending. Execution: not executed. Other events were allowed, blocked or failed.
Your tools.
Your workflows.
Your control.
Keep your stack. Point your MCP-compatible agent at Aethis. No SDK required.
Explore connection paths
Companies that publish their own MCP server. This is not a list of supported integrations, customers or endorsements. At launch, Aethis governs MCP-compatible agents and remote MCP servers; direct model routing supports OpenAI.
Give agents room to act. Keep control of what happens.
Bring your first workflow under policy. Build from there.