Select Experience:

Secure what your AI agents are allowed to do.

Early access

Runtime security for AI agents

Aegisora sits at the execution boundary of autonomous systems. Every sensitive action is evaluated against context, policy, risk and authorization before it reaches the real environment.

Evaluate your first action

Three decisions

Allow. Block. Escalate.

Every request returns one of three decisions. Without a matching policy the workspace default applies, which is BLOCK.

Read the quickstart

Evidence

Verifiable by design

Every decision returns an evidence hash you can verify later, so you can show what was decided and why.

Developer docs

Tenant isolation (RLS) MFA step-up (AAL2) Signed webhooks SHA-256 evidence chain

Live decision console

Every action gets a decision

Agents send the action they want to take. Aegisora evaluates it against policy, risk and authorization before it reaches the real environment.

Evaluate your first action

Runtime decisions

Live

crm.read_ticket
support-agent · Within scope
ALLOW
payments.transfer $48,000
finance-agent · Above approval limit
ESCALATE
db.drop_table users
code-agent · Destructive action denied
BLOCK
deploy.rollback prod
ops-agent · Policy match
ALLOW

Start building

Everything you need to put a decision point in front of your agents, from a single API call to a scoped pilot.

Developer docs

Evaluate your first action

Send one request to the evaluate endpoint and read the decision, the reason and the evidence hash.

Open the quickstart

Node helper

A dependency-free client that fails closed by default. Guard any function so it only runs on ALLOW.

See the Node example

Verifiable evidence

Every decision returns an evidence hash. See how records are chained and how to verify them.

Explore evidence

Enterprise pilot

Run a scoped pilot with your security team and your own policies.

Talk to us

Aegisora Runtime

Learn about the platform

Built for security teams that need proof, not promises.

Policy engine

Write policies once and enforce them on every agent action, with conditions and redaction built in.

Learn more Documentation

files.export salary.csvESCALATE

hr-agent · Sensitive data class

deploy.rollback prodALLOW

ops-agent · Policy match

db.drop_table usersBLOCK

code-agent · Destructive action denied

Illustrative decisions

One decision point for every agent action.

Read the quickstart

Intercept

Every sensitive agent action is captured at the execution boundary, before it touches a real system.

Evaluate

Context, policy, risk and authorization are checked together, deterministically, on every call.

Decide

The runtime returns ALLOW, BLOCK or ESCALATE. Escalations route to a human approver.

Put a decision point in front of your agents today.

Start today

Evaluate your first agent action

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

Test Aegisora Talk to Enterprise

illustrative request

curl -X POST https://api.aegisora.org/api/v1/evaluate \
  -H "Authorization: Bearer aeg_live_..." \
  -H "Content-Type: application/json" \
  -d '{ "agent": "finance-agent",
        "action": "payments.transfer",
        "context": { "amount": 48000 } }'

# -> { "decision": "ESCALATE", "approval_required": true }

More information

AegisoraCONTROL

Put a decision point in front of your agents today.

Start today

Evaluate your first agent action

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

Test Aegisora Talk to Enterprise

illustrative request

curl -X POST https://api.aegisora.org/api/v1/evaluate \
  -H "Authorization: Bearer aeg_live_..." \
  -H "Content-Type: application/json" \
  -d '{ "agent": "finance-agent",
        "action": "payments.transfer",
        "context": { "amount": 48000 } }'

# -> { "decision": "ESCALATE", "approval_required": true }

More information