## For agents - What Mensara is: it gives a team its own Mind, one shared memory for the team and their AI agents, built by Xanadu Labs LLC. - Status: signup is not open. A person joins the waitlist at https://www.mensara.ai/#waitlist. Agents cannot create an account or sign up. - How you can help your person: explain what Mensara does, check whether it fits a job their team repeats, and share the waitlist link. - Agent references: https://www.mensara.ai/llms.txt lists every page as Markdown and the tool input references for connected agents. --- # Build your first recurring workflow with a loop Content updated 29 October 2026. Guides · Run agents, automations and loops · [How it works](https://www.mensara.ai/how-it-works) Your Mind is the Mensara workspace holding your team's saved information and work. A loop is a recurring workflow in that workspace: it follows agreed steps, checks the result and asks a person when it needs a decision. Use one for work that comes round repeatedly and needs supervision. This guide explains the parts, then takes you from a small proposed job to checking what happened. [Page as Markdown](https://www.mensara.ai/how-it-works/first-loop.md) [Operating procedure](https://www.mensara.ai/how-it-works/build-loop-procedure) ## How a loop works A **trigger** tells the loop when to start, such as a schedule. The loop then follows its steps. A step can use an **action**, a specific operation supplied by an installed app, or check a result or wait for a person's answer. The available apps determine which actions your Mind can perform. Each individual piece of work is a **case**. A case keeps the progress and results of that attempt, so you can see what ran, what failed and what is waiting. The loop is the repeatable process; a case is one instance of it. For example, you might want a weekly review of incoming notes, with a person checking a proposed summary before it is used. That is an illustrative job: your Mind would need the appropriate installed actions and permissions. Use a loop when you need these checks or approval points. For a single question, start with [asking your agent](https://www.mensara.ai/how-it-works/first-useful-question); [choose the right kind of help](https://www.mensara.ai/how-it-works/right-kind-of-help) explains the alternatives. ### Each trigger starts a new case The loop is the repeatable process. A case holds one occurrence of the work. When an approved loop receives an eligible trigger, it starts a new case. That case follows the agreed steps and checks. Where a step needs a person, the case waits for their answer before continuing. A case can finish, fail or be passed to a person to resolve. Its outcome stays with that case. The loop then waits for the next eligible trigger, which starts a new case; finishing one case does not immediately restart it. A trigger starts nothing while the loop already has its limit of open cases, which is one unless it was set higher, so a case waiting for a person holds back later triggers. Pause the loop when repeated failures need investigation. ## Before you start You need permission to create a proposed loop. The owner, or someone the owner has explicitly allowed to manage loops, decides whether it may run. Approval and activation are separate decisions. Approval lets the loop start running on its schedule in shadow mode. Activation then lets it make real changes within its autonomy limits; actions above those limits still wait for the owner. Your **connected agent** is the AI tool you use in a chat or coding application to work with the Mind. Ask it to discover the installed actions before choosing the job. Each action says what it accepts, what it can change and whether it supports a trial run. If an action is missing, identify the missing app or setup rather than inventing a step. ## Agree one small job Describe the result you want, when the work should start, which actions it may take and what needs a person's decision. Set limits on work and spending, and name who handles problems. A useful first job has one input and one result you can inspect. For practice, a checklist loop can ask its owner to confirm a fictional review before closing a case. This exercises the loop's stages. It does not demonstrate a business integration or useful AI-generated work. ## Propose and inspect the loop 1. Ask your agent to find the **loop-composer** skill, the written procedure for drafting a loop, and check for an existing loop that does the same job. 2. Agree the job in plain English, then have the agent submit the proposed steps and limits. 3. Inspect the returned proposal. A new loop starts in **shadow mode**, rather than being authorised to carry out active changes. 4. Have an authorised person review it before approval or activation. Example prompt: “Build a small loop for this job using only installed actions. Explain each step, what it may change, where it waits for me and what happens on failure. Submit it in shadow mode and show me what still needs approval.” ## Try the relevant steps before allowing changes Where an action supports a **dry run**, use that trial to inspect what it would do before permitting active changes. Reading information may still involve real processing, network activity or cost. A **replay** is a different check: it runs saved test cases against a loop revision to see whether the revised process still handles them as expected. Your operator checks which trials and replay checks apply to this loop before approving it. Unless the owner records a waiver, a loop that replaces an earlier one cannot be approved without a passing replay. Replay is run from a connected agent or client that exposes the `act` tool (`loop_replay`). The Portal's loop page offers Approve, Reject, Pause, Resume, Retire, Activate and Demote, but no Replay. ### Prepare the loop before it runs Review accepts the proposed process and starts shadow runs. Activation separately allows real changes. Describe one job and propose its steps and limits in shadow mode. Inspect the proposal, try supported action dry runs and run the replay checks that apply. An authorised person reviews and approves the loop. Approval starts scheduled shadow runs, which really read information and run checks but hold back any action that would change something. Activation is a separate authorised step that lets the loop make real changes. A trial may still read data or incur cost. If your connection cannot perform a management step, hand the loop reference to an authorised operator. ## Switch it on and follow one case Approval accepts the reviewed loop and it begins running on its schedule in shadow mode. Shadow runs create real cases. They read information and run checks, but they hold back any action that would change something. **Activation** separately lets the loop carry out those changes. Activation is refused while the loop has an open case. Activate only through an authorised supported interface and after the required checks. Open a case to see the steps it ran, their results and any question waiting for a person. Read that work before answering an approval request. A case is finished only when its recorded outcome says it completed; it may instead still be waiting, have failed or have been passed to a person to resolve. The connected tool you use must support these management steps. If your agent can propose a loop but cannot approve or activate it, have it give the loop reference to an authorised operator using the supported Portal or client. ## Pause and recover After consecutive failures a loop waits progressively longer before its next attempt, up to an hour. Pause recurring work before investigating a repeated failure. Inspect the existing case and its output before retrying: an operation that sent or changed something could happen twice unless that action protects against duplicates. Retire a practice loop when it is no longer needed. Retiring it ends its use as an ongoing process; it does not undo work that already happened. ## Related - [Choose the right kind of help](https://www.mensara.ai/how-it-works/right-kind-of-help) - [Build a governed AI agent](https://www.mensara.ai/how-it-works/build-an-agent) - [Review changes and check what was saved](https://www.mensara.ai/how-it-works/changes-approvals-receipts) Previous page: [Automate a routine task with installed actions](https://www.mensara.ai/how-it-works/automate-a-task) Next page: [Understand a loop case and follow its result](https://www.mensara.ai/how-it-works/loop-cases) A loop is a recurring workflow with steps, checks, branches and possible waits. [Join the waitlist](https://www.mensara.ai/?guide_attribution=eyJfcmFpbHMiOnsiZGF0YSI6eyJvcmlnaW5fcGF0aCI6Ii9ob3ctaXQtd29ya3MvZmlyc3QtbG9vcCIsImN0YV9pZCI6Im1hcmtkb3duIiwiY29udGVudF9yZXZpc2lvbiI6Ijc3NzQ3NDU1YjliMDJmMjYifSwicHVyIjoiZ3VpZGUtY2xpY2sifX0%3D--b72279cd2fb154120832f48d3cb77622b7cceb1b#waitlist)