AI agents that work through the night and wait for your sign-off.

Midnira builds and runs AI agents for law firms and other paper-heavy businesses, and uses agents to make penetration tests go further. The agents do the reading and the legwork. Nothing is filed, sent or reported until a person has checked it.

One example night on a 24-hour dial Midnight is at the top. An outer lane shows AI operations, from a fax at 11:42 PM to staff approval at 8:05 AM. An inner lane shows a penetration test, from the test window opening at 2:30 AM to the operator reproducing a finding at 8:40 AM. A marker shows your local time. Two small dials show how many proposals are waiting for sign-off and the current phase of the moon.
AI operationsPenetration testingA person's decisionYour local time3 waiting for sign-offMoon phase

One night, as an example

AI operations

  1. Operations agent

    A 14-page fax arrives. The agent splits it into separate documents and proposes a folder for each.

  2. Operations agent

    A court notice lands in the inbox. The agent reads the hearing date and drafts a calendar entry.

  3. Staff

    The filings and the calendar entry are approved. One folder is corrected first.

Penetration testing

  1. Testing agent

    The test window opens. Every target is checked against the signed authorization.

  2. Testing agent

    An invoice endpoint returns another customer's record. The agent flags it for the operator.

  3. Operator

    The finding is reproduced by hand, rated high and written up.

We build the agents, then we stay to run them.

For a law firm, the agents read incoming mail, faxes and court notices, propose where each document belongs and who needs to see it, and then wait. Staff approve each filing with a click.

See AI operations
Received by email 1:07 AM

Notice of hearing

File to
Matter 2291 › Court notices
Calendar
Hearing, Thursday at 9:30 AM
Notify
Assigned attorney
Approved by staff 8:05 AM

Penetration tests that cover more ground, led by an operator who has done it by hand.

In an agentic test, AI agents take on the wide, repetitive work of mapping the attack surface, trying the known paths and re-checking fixes. That leaves the operator's time for the chained, context-dependent attacks that tools miss. Every finding in your report has been reproduced by a person.

See penetration testing

A sample finding, as it appears in a report

Invoice API returns other customers' records

Severity
High
Where
GET /api/invoices/{id}
Flagged by agent 3:18 AM Reproduced by operator 8:40 AM

One practice, pointed in two directions.

Running agents in production shows you how they fail. Breaking into systems shows you what failure costs. Midnira's operations work is built by a penetration tester, and its testing uses agents built by someone who runs them in production. The same three rules apply to both.

People approve what matters

Agents propose and people decide. The approval step is part of the design and cannot be switched off for convenience.

Try the morning queue
Code decides what must be right

Where there is one correct answer, such as a deadline, a filing rule or a scope boundary, it is enforced by code. Models are used where judgment helps.

Try the scope gate
Everything leaves a record

Each action is logged with its inputs and the person who approved it, so any result can be traced back to its source.

See the record

How your data is handled

The agents run on Claude
Claude is the AI model made by Anthropic. It is the model behind the agents we build and run.
Your data does not train AI models
The documents, messages and test results that pass through the agents are not used for model training.
Ask us the rest
Where the agents run, who can read the logs and how long data is kept depend on the engagement. We answer in writing before any work starts. See the questions firms ask

A small practice, on purpose.

Derek Fish, founder of Midnira

I'm Derek Fish. I started Midnira in 2026, and the person who scopes your work is the person who does it.

I have spent eight years in security, across penetration testing, red team work and managed detection and response. I have also built and operated fleets of AI agents in production, where they handle a company's daily operations under human supervision.

I have earned eight GIAC certifications, including GXPN (exploit development), GPEN (penetration testing), GCIH (incident handling) and GCED (enterprise defense).

Derek Fish, founder

Derek on LinkedIn

Tell us what you want handled, or what you want tested.

Send a few lines about the work, or book a 30-minute call. A person reads every message and replies.