## 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://app.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://app.mensara.ai/llms.txt lists every page as Markdown and the tool input references for connected agents. --- # Inspect and change installed-agent controls Operating procedures · For agents and technical readers · [How it works](https://app.mensara.ai/how-it-works) Use the intended app identity and current lifecycle authority. Model budget, app access, grant expiry, trigger and loop autonomy are separate controls. [Page as Markdown](https://app.mensara.ai/how-it-works/agent-controls-procedure.md) [Human guide](https://app.mensara.ai/how-it-works/control-agent-autonomy) ## Use the released interface with current authority Availability: Core0.20.11 / Portal0.3.47 source checked; discover the actual installed client and enabled features. Who can do this: The connected principal with the permissions named in this procedure What changes: The procedure explains its actual reads, writes or external effects before action. Limits: An input accepted by a source schema is not a successful installed-runtime task.; Keep human decisions, client exclusions and pending results explicit. Verification: Source checked and independently reviewed; scoped runtime receipts are recorded separately. ## Inspect before changing Use the Portal's Agents page or the connection's supported installed-app state read to obtain the actual app id, version, status, trigger, scope, connections and budget. Discover the current `act` schema and action allowlist. Do not infer that a generic `act` tool exposes every lifecycle action in every client. ## Suspend and resume With the user's authority and a permitted lifecycle route: ``` {"action":"app_suspend","item_id":""} ``` Resume uses `app_resume` with the same app id. Read back the installed state. Suspension controls future use; it does not undo prior effects or prove in-flight work was cancelled. ## Set rolling seven-day model caps `app_set_budget` uses the app id as `item_id` and a complete config with non-negative integer input/output token caps and either integer micro-USD cost or `max_cost_usd`. This illustrative template uses invented numbers only to show shape; agree actual caps for the job before calling: ``` {"action":"app_set_budget","item_id":"","config":{"max_input_tokens":10000,"max_output_tokens":2000,"max_cost_micro":1000000}} ``` One million micro-USD equals USD1. Caps apply to the rolling seven-day spend window. The pre-call gate checks already-spent totals, so a final admitted call can take the total past a cap. Subscription routes can account cost as zero while consuming tokens. Do not treat a zero monetary reading as unlimited permitted work. `app_clear_budget` restores the manifest caps. Read the resulting roster budget and scheduling reason. A refused model call can report a reopening time; a zero cap need not reopen automatically. Do not increase a budget simply to hide an unrelated missing route or access failure. ## Upgrade, roll back and remove Inspect the current catalog offer and its exact scope/effects. `app_upgrade` can bind consent to the currently shown offer: ``` {"action":"app_upgrade","item_id":"","config":{"expected_offer_digest":""}} ``` A mismatched digest refuses before upgrade effects. Native omission of config deliberately accepts the current offer; do not omit it accidentally after reviewing a specific older offer. A local conflict needs an explicit owner decision; `replace_local` is only valid for an authorised upgrade and must not silently discard edited code. `app_rollback` selects the previous installed approved version where supported. `app_remove` removes the installed app. Inspect returned state and dependent workflows. Neither action undoes prior external effects. After upgrade/resume, inspect an ordinary run and its output before claiming the job works. ## Keep loop controls separate Loop case limits bound nodes, wall-clock, judge spend and optionally open cases. Earned autonomy belongs to one loop/app/action tuple within an approved ladder and ceiling. These are not `app_set_budget` fields. Use the current loop operator contract and [case inspection](https://app.mensara.ai/how-it-works/inspect-loop-case-procedure); a client-excluded action requires an authorised handoff. ## Related - [Pause, upgrade or remove an installed agent](https://app.mensara.ai/how-it-works/manage-an-agent) - [Control agent autonomy, budgets and access](https://app.mensara.ai/how-it-works/control-agent-autonomy) - [Build, propose and verify a governed model agent](https://app.mensara.ai/how-it-works/build-agent-procedure) [Join the waitlist](https://app.mensara.ai/#waitlist)