Aegisora | Runtime Security for AI Agents

Example actions: payments.transfer, db.drop_table, email.send, crm.read_ticket, files.export, code.execute, browser.navigate, vendors.export_bank_details

FEATURES

Agents act. Aegisora
decides first.

Policies decide, people approve the risky ones, and every decision lands in a log you can verify.

Policy

Rules that decide per action

Conditions on the agent, the tool, the target and the amount. Versioned and enforced on every call.

Approval

A person decides the risky ones

Sensitive actions wait in an approval queue. Execution resumes only after a decision.

Evidence

A record you can verify

Each decision is stored in a SHA-256 hash-chained log that you can check independently.

Future of runtime security

aegisora.com

agent.tspolicy.json

// ask first, then act
const decision = await checkAction({ agent, action, context });

switch (decision.verdict) {
  case "ALLOW": return runTool(action);
  case "ESCALATE": return queueForApproval(decision.id);
  default: // BLOCK or unknown: fail closed
    return stop(decision.reason);
}

ESCALATEdecision.id = dec_0002

AegisoraIN BRIEF

3

Verdicts: ALLOW, ESCALATE or BLOCK

BLOCK

Default when no policy matches

SHA-256

Hash-chained evidence log

Aegisora answers before the tool runs: the action, the agent, the context and the policy go in, and a verdict with a reason comes out. When nothing matches, nothing runs.

Design principleDecide first, then record

THE PROBLEM

Your agent controls
aren't decisive

Permissions are often granted once, at setup. A sensitive action needs a decision at the moment it is requested.

Without a decision point

Access is granted once, logs are written after the action ran, and no record a reviewer can verify.

With Aegisora

A verdict for every sensitive action before it runs, people approve what needs it, evidence is chained.

Platform

One decision layer for every agent action.

Policies decide, people approve the risky ones, and every decision lands in a log you can verify.

Get Started

4

Steps for every decision

3

Verdicts per action

1

Default when nothing matches: BLOCK

SHA-256

Hash chain across records

Agent actions happen across tools, APIs, databases and inboxes every minute. Aegisora captures each request, decides before it runs, and records proof you can verify, so control keeps up with your agents.

HOW IT WORKS

How Aegisora
decides first

The same four steps apply whether an action reads a ticket or moves money.

STEP 01

The agent proposes an action

Before a sensitive tool runs, your application sends the action with its context: the tool, the target and any amount involved.

Read the docs

DECISION VIEWS

Decisions you can
read at a glance.

The verdicts come from a small, illustrative policy set. They are not read from a live workspace.

Allow

Within scope, it runs

The tool runs and the decision is recorded with its reason.

Escalate

Above the limit, a person decides

Execution waits in the approval queue until someone decides.

Block

Denied, and still recorded

The tool never runs, and the attempt stays in the evidence log.

COMPARE

Better than access
checks alone

A general comparison of approaches to controlling agents, not of specific products.

Control capabilities	Aegisora	Access roles	Activity logs	Prompt guardrails
Decision before the action runs	Yes	Partly	No	Partly
Judges the request and its context	Yes	No	No	Partly
Human approval for sensitive actions	Yes	No	No	No
Tamper-evident decision record	Yes	No	Partly	No
Fails closed when nothing matches	Yes	Yes	No	No
Rules per agent and per tool	Yes	Partly	No	No
A reason recorded for every decision	Yes	No	Partly	No
Works on any tool call	Yes	Partly	Partly	No

Before

A verdict arrives before execution

Human

Approval for sensitive actions

Chained

Every decision on the record

BENEFITS

Control that keeps pace
with your agents

Insight

Decision trends

See how verdicts move across agents and tools over time.

Security

Fail-closed routing

Risky actions stop or wait. Unknown ones are blocked by default.

Scope

Authority per agent

Each agent holds only the authority you grant it, checked on every call.

Delivery

Decisions sent on

Pass decisions to your own systems as they happen.

TOOL CALLS

Sits in front of
every tool call

Works with the actions your agents already take. Send the request, get the verdict, keep your own execution.

Paymentspayments.transferDatabasesdb.drop_tableEmailemail.sendCRMcrm.read_ticketFilesfiles.exportCodecode.executeBrowserbrowser.navigateVendorsvendors.export_bank_details

EVIDENCE

Every decision,
on the record

Each record includes the hash of the record before it. Changing an earlier entry breaks verification for everything after it.

Decisions

ALLOW Within scope
ESCALATE Waits for a person
BLOCK Stopped and recorded

Illustrative sample, not live data.

ACCESS

Access, step by step,
as you are ready

Pricing, hosting and service levels are not published yet. Start with a test workspace.

The workspace includes:
Policies
Approval queue
Evidence log
API keys

Test workspace

Evaluate a first agent action.

Early accessno paid plan today

Test Aegisora

Create a test workspace
Send a request and read the verdict
Write policies per action
Review decisions in the evidence log
Default for no match: BLOCK

Start here

Design partner pilot

Scoped around one real use case.

By applicationagreed in writing

Apply for a pilot

Scoped and agreed in writing
Before customer data is connected
Approval queue for sensitive actions
Hash-chained evidence for review
Scope, data and timeline agreed first

Production

Not yet available.

Not publishedhosting and service levels not verified

Talk to Enterprise

Pricing is not published yet
Hosting is not verified yet
Service levels are not published yet
Talk to us about timing
Nothing here is promised ahead of proof

Decisions without
blind spots

Every sensitive action gets a verdict, whichever agent or tool sends it.

Get Started

DOCS

Guides & documentation

QuickstartSend your first request

DocsAegisora

Send your first request

Create a test workspace, send one request and read the verdict.

Policies, approvals and evidence

How authority, the approval queue and the evidence log fit together.

Design partner pilots

How a pilot is scoped and agreed in writing before customer data is connected.

What teams ask first

Illustrative scenarios, not customer quotes.

A finance agent asks to move $48,000. The amount is above the limit, so a person decides before anything runs.

PA

PaymentsScenario

ESCALATE

A code agent tries to drop the users table. Destructive SQL is denied by policy and the attempt is recorded.

DA

DatabasesScenario

BLOCK

A support agent reads a ticket inside its granted scope. The read is allowed and logged with its reason.

SU

SupportScenario

ALLOW

A scheduled job exports vendor bank details. No policy matches, so the default applies and the export is blocked.

OP

OperationsScenario

BLOCK

FAQ

Common Questions Answered

Find answers about Aegisora decisions, approvals, evidence and early access.

Yes. Create a test workspace to evaluate a first agent action. There is no paid plan and no online checkout today.

Not yet. Pricing, hosting and service levels are not published. Any paid pilot is scoped and agreed in writing before customer data or production systems are connected.

Between the agent and the tools it calls. Each sensitive action is sent for a decision before the tool call runs.

The action is blocked. Aegisora fails closed by default.

Your application. Aegisora returns a verdict; your code acts on ALLOW, holds on ESCALATE and stops on BLOCK.

The action waits in an approval queue. Execution resumes only after a person decides.

Talk to our team anytime. Just send us a message.

Apply for a pilot Contact us

START TODAY

Test one decision in your own workspace.

Create a test workspace, send one request and read the verdict. Without a matching policy, the default is BLOCK.

Apply for a pilot Test Aegisora

Your agents act.
Aegisora decides first.

Agents act. Aegisora
decides first.

Rules that decide per action

A person decides the risky ones

A record you can verify

Your agent controls
aren't decisive

Without a decision point

With Aegisora

One decision layer for every agent action.

How Aegisora
decides first

The agent proposes an action

Decisions you can
read at a glance.

Within scope, it runs

Above the limit, a person decides

Denied, and still recorded

Better than access
checks alone

Control that keeps pace
with your agents

Decision trends

Fail-closed routing

Authority per agent

Decisions sent on

Sits in front of
every tool call

Every decision,
on the record

Access, step by step,
as you are ready

Test workspace

Design partner pilot

Production

Put a decision in front of every tool call

Decisions without
blind spots

Guides & documentation

Send your first request

Policies, approvals and evidence

Design partner pilots

What teams ask first

Common Questions Answered

Test one decision in your own workspace.

Your agents act.Aegisora decides first.

Rules that decide per action

A person decides the risky ones

A record you can verify

Without a decision point

With Aegisora

One decision layer for every agent action.

The agent proposes an action

Within scope, it runs

Above the limit, a person decides

Denied, and still recorded

Decision trends

Fail-closed routing

Authority per agent

Decisions sent on

Test workspace

Design partner pilot

Production

Put a decision in front of every tool call

Decisions withoutblind spots

Send your first request

Policies, approvals and evidence

Design partner pilots

Can I try Aegisora today?

Is pricing published?

Where does Aegisora sit in my stack?

What happens when no policy matches?

Who controls execution?

What does ESCALATE do?

Test one decision in your own workspace.

Your agents act.
Aegisora decides first.

Decisions without
blind spots