Lockwork 1 READ Agent level, go to the levels

Operations agents

Agents that do the work. You keep the key.

Lockwork agents clear your operations queue: billing fixes, access requests, ticket triage, on-call chores. Each one works at the level you grant it. Anything it can't undo waits for a named person to approve.

Request access Give it a task

30-minute working session. Bring one real task.

Replay of a scripted run at a fictional company.
TaskJOB-0412 · Billing

Refund last night's duplicate charges.

Planrevised

Refund the 3 confirmed duplicates. Retry the other 11.

    Effect—

      Needs a key

      Issue 3 refunds · $1,240.00

      Key holder: Dana Ruiz, Finance

      Why agent pilots stall

      The demo goes well. Then security asks two questions: what is it allowed to touch, and what did it do at 2 a.m. last Tuesday? If nobody can answer, the pilot ends there.

      Give it a level.

      Every tool an agent can use sits at one of four levels. You grant them per agent and per system, and you can lower one in the middle of a run.

      Level 1 Read

      Looks, never changes. Logs, tickets, dashboards, invoices.

      JOB-0412 Read the billing run log, 02:00–02:14

      No one approves. Every read is logged.

      Level 2 Draft

      Prepares the change and stops: replies, refunds, pull requests, config diffs.

      JOB-0412 Draft refunds for the 3 duplicates

      A person sends it, or throws it away.

      Level 3 Act

      Makes changes it can undo, inside limits you set: retries, reassignments, restarts, suspended accounts.

      JOB-0412 Retry the 11 card errors

      No approval inside the limits. One click reverses the run.

      Level 4 Key

      Anything it can't undo, or anything over a limit: refunds, deletions, production rollbacks.

      JOB-0412 Issue 3 refunds · $1,240.00

      A named key holder approves, every time.

      Sandbox

      Give it a task.

      Type something your team does by hand. The sandbox is scripted, so nothing real runs, but the levels are the ones the product uses.

      Verdict—
      L– ———

      Highest level this task needs

      Plan

      Waiting for a task.

        Effect—

          Needs a key

          This step needs a key. In the sandbox, the key is yours.

          Scripted simulation. No API calls; nothing leaves your browser.

          Every step is written down.

          Each run leaves a log your auditors can read: the tool it called, the records it touched, the level it used, and who held the key. Export it, or stream it to your SIEM.

            October at Harrow & Pike: 14,212 steps, 312 keys turned.

            From a 60-day pilot

            Illustrative figures for a fictional customer, Harrow & Pike Logistics.

            2,140
            tickets closed with no one touching them
            41min
            median time to first action on overnight alerts, down from 6 h 10 min
            0
            irreversible actions taken without a key
            3min
            3 minutes, median wait for a key holder

            We stopped arguing about whether to trust it. Now we argue about which level to give it, which is a much better argument.

            Priya Natarajan, Head of Platform Operations, Harrow & Pike (fictional)

            Request access

            Lockwork is in early access for operations teams of 20 to 500 people. The first month looks like this:

            1. Day 1A 30-minute working session. Bring one task your team does by hand; we run it with you in a sandbox against your tools' test accounts.
            2. Days 2–30A 30-day pilot at Levels 1 and 2, free. The agents read and draft. Nothing changes unless your team sends it.
            3. ThenRaise levels when the log has earned it.

            After the pilot, plans start at $1,500 a month for 2,000 runs (illustrative pricing).

            We reply within one business day with two times for the session. No sales sequence.

            Runs in your cloud or ours. Customer data is never used to train models. SSO, SCIM, and separate credentials for every agent.

            Questions

            What happens when an agent gets it wrong?

            At Levels 1 and 2 nothing has changed yet, so you discard the draft. At Level 3 every action has a recorded undo, and one click reverses the run. Level 4 actions don't happen without a person.

            Which tools does it work with?

            Anything with an API or a command line: ticketing, billing, identity, cloud consoles, chat. Each connection gets its own credentials and its own level.

            Does it replace my team?

            It takes the queue nobody gets to. The decisions stay with the people holding keys.

            Can we run it in our own cloud?

            Yes. The agent runtime deploys into your VPC, and the log stays there unless you export it.

            Start with one task.

            JOB-0412 · Refund last night's duplicate charges. · Level 4

            Request access

            Bring it to the session. We'll run it with you, at the level you choose.

            Read the docs
            Request access