LIMEN13

For MSSPs

AI agent security for managed security providers

Your customers are deploying AI agents that act on their systems. LIMEN13 lets you offer control, containment and evidence for those agents as a service, with one isolated engine per customer.

In plain words. If your company looks after other companies' security, you can run LIMEN13 for each of them. Every customer gets a separate copy, so one customer's information and keys never mix with another's.

Why

A new service on a model you already run

A gap your customers already have

Agents receive tool access and API keys faster than anyone reviews them. Customers will ask who is watching.

Work your analysts know

Policies, alerts, containment, evidence and a monthly report: the discipline a SOC applies to endpoints, applied to agents.

Evidence you can hand over

Each customer's record is signed with that customer's own key and verifies on its own.

Isolation

One customer, one engine, one of everything

Customers share no database, no queue and no session.

What is separated for each customer, and how
For each customerHow it is separated
Policy enforcementOne engine process per customer, loaded with that customer's policies only. No shared memory, cache or queue.
Credentials and security boundariesEach customer's keys live under its own name or vault path. Stored service tokens are encrypted at rest and bound to their agent.
Evidence recordsOne evidence chain per customer, signed with that customer's own key. A second engine on the same folder refuses to start.
Agent inventoryAgents, people, sensors and collectors are defined per customer. One customer's keys mean nothing to another engine.
Incident recordsAlerts, containments and kill-switch actions are written to that customer's chain only.
Monthly reportComputed from that customer's verified evidence, under that customer's name.

How this is tested

An automated test starts two real engines on two workspaces. It checks that one customer's agent key, run token, console session, sign-in, sensor key and collector key are all refused by the other engine, and that each evidence chain verifies only with its own key.

What exists today

What you operate, and what does not exist yet

In release 0.10.1

  • One engine and one console per customer, each on its own hostname
  • Its own OS user or container, signing key and vault path
  • One engine image for all customers; the customer is chosen at start by its folder
  • Encrypted backups for each customer

Not built

  • There is no central console showing all customers on one screen.
  • Your analysts open each customer's console, or work from your SIEM, where every engine can send its alerts.
  • A central view is a direction we are considering. It is not a committed feature and has no date.

Operations

The workflow, customer by customer

  1. Onboard the customerCreate its workspace, signing key and hostname, and connect its sign-in.
  2. Register agentsRecord each agent with an owner, issue its key, and move its provider and service credentials into LIMEN13.
  3. Configure policiesStart in observation mode, review what would be refused, then enforce.
  4. Monitor executionsAlerts arrive in your SIEM, by webhook, in Slack or in Teams.
  5. Respond to incidentsContain a run or disable an agent, then read the run's storyline.
  6. Verify evidenceVerify the chain on demand, and compare it with off-site anchors when they are configured.
  7. Review the monthly reportEdit the draft's recommendations, then send it to the customer.

Evaluation guide

What to check in an MSSP pilot

Run these with two test customers, on non-production systems.

  • Start two engines and confirm that one customer's keys, sessions and sign-ins are refused by the other.
  • Verify each evidence chain with its own key, and confirm it fails with the other customer's key.
  • Send both customers' alerts to your SIEM and check that you can tell them apart.
  • Disable an agent mid-task for one customer and confirm the other is unaffected.
  • Generate the monthly report for each customer and compare it with what you observed.
  • Back up one customer's workspace and restore it to a new folder.
  • Measure the added delay per call with a workload like your customer's.
  • Record the operating effort per customer: it is your cost model.

MSSP licensing is quoted on request. Pilot and pricing. See the sample monthly report.

Evaluate it with two test customers.

We help you run two isolated engines and go through the evaluation guide together.