If data cannot leave the institution, security must run inside it.
Public institutions run agents on citizen data. AvertGate installs inside the institution's network, makes decisions there and keeps records there.
Decision flow
The request never left the network
fetch_urlhttp://169.254.169.254/- Cloud metadata address is closed
- Private networks and localhost are closed
- Decision made inside the institution
Everything that leaves the network is a decision.
For institutions the problem is not only an agent doing the wrong thing; it is being unable to show by which path and with whose approval data left.
Install inside
The Data Plane runs in the institution's network; calls and results do not leave.
Deny by default
A tool that is not listed does not run. Authority is only what the task needs.
Document approval
Risky actions go to an authorized approver; the approval and reason are written to the record.
See it. Limit it. Prove it.
Example rules for an applications agent.
Lookups are free, deletion is closed, outside network access is limited.
- Network shield. Requests to the cloud metadata address, private network ranges and localhost are blocked.
- Local model. Semantic inspection can run inside the institution with Ollama or vLLM.
- One record. The decision, reason and rule id are written to a SHA-256 chained record.
Agent management
Every call in the live stream.
Take the rules out of code.
Allow, limit and approval rules per tool and argument live in a single YAML file. A change needs no redeploy.
- Deny by default. A tool that is not listed does not run.
- Argument limits. Fields such as amount, SQL and URL are bound to a rule.
- Human approval. Risky calls wait for approval in Slack.
Policy editor
The rule is written on one screen.
Every decision is recorded.
The decision, reason and rule id are linked with SHA-256; a record changed afterwards is detected.
- Reasoned decision. Every event carries the reason and rule id.
- Redacted arguments. Raw data never reaches the dashboard.
- Session trail. Calls in the same session are read together.
Session trails
Every session is traced end to end.
Agent management
Every agent in the institution, with its status and call count.
Frequently asked
Does it work without internet access?
Yes. Deterministic checks are local already; for semantic inspection a local model is installed.
In what formats is it delivered?
As a standalone binary or an enterprise Docker image.
How does procurement work?
A demo and POC on your own agents come first; then an annual contract is signed.
Guarantees
Four fixed principles behind every decision.
- 3 decisionsEvery call ends in allow, block or human approval.
- Fail-closedIf a check errors, the call is not run.
- SHA-256 chainEvents are chained, so any change after the fact is noticed.
- Your data stays with youTool calls and results never leave your network.
Put a gate in front of your agents.
Every deployment starts with a demo and a POC on your own agents.