Agents overview

Agents are scheduled AI runs that work autonomously — Strata executes them on the schedule you set, with the tools you allow, and (optionally) emails you the result so you can act on it without ever opening a chat.

What an agent run is — vs. a chat conversation

A chat is interactive: you ask, the AI answers, you steer with a follow-up. An agent is one-shot: you write the task once, and Strata runs that same prompt on a recurring schedule (or on demand). There is no follow-up because there's no human in the loop during the run. Each execution:

• Runs in a fresh session with no chat history.
• Uses only the tools you explicitly enabled.
• Has a hard runtime cap (30–600 seconds, default 300).
• Has a hard token cap on its written response (500–100,000 tokens, default 20,000).
• Ends with a status of success, failed, or timed out, and the full output is stored in the agent's run history.

Each run is its own chat

Every real run opens a chat of its own, the way a scheduled task's runs do in Claude. The agent's instructions are the first message, labelled Ran manually, Ran on schedule or Ran from its webhook, and the run's answer appears under them with the steps it took. While the run is still working the chat says so, and the answer appears when it finishes. Reply in that chat to ask about the result or take the work further: the reply is an ordinary chat turn, with the model you pick. The chat's title bar reads Agents / and the agent's name; Agents opens the agent, and so does View agent in the title's menu.

Run chats stay with their agent: they are not listed under your recent chats. Open one from the agent's run history, from the run's notice in your Inbox (Open the chat), or from View on the notice that appears when you press Run now. A test run opens no chat, since it acts on nothing, and neither does a run another agent starts. Deleting an agent deletes its runs' chats too, as deleting a chat does. If your settings ask Strata not to keep agent output, a run's chat keeps neither the instructions nor the answer, only a note saying so.

Use agents for recurring reports, monitoring, inbox triage, OneDrive intake — anything you wish would just happen on its own.

The Agents page

Click Agents in the sidebar. The page shows your agent table with these columns: Agent, Status, Schedule, Model, Next run, Last run, plus a row-action cell on the right. Schedule reads in words on a 12-hour clock (Weekdays at 9:00 AM, Every 15 minutes), with the agent's time zone under it when the time of day matters. Next run and Last run read from now (in 2h, 3h ago · failed); hover one for the full date and time in the agent's time zone (Mon, Sep 21 at 9:00 AM EDT). An agent that needs attention says why in a line under its name. Above the table, All, Needs attention and Paused filter the list, each with its count, and the filter is kept in the page's address (?show=attention or ?show=paused), so a reload or a shared link keeps it. An N running note appears while runs are going, a Search agents field appears once you have more than eight agents, and the Create agent split button sits on the right. On a phone each agent is a card whose lines are labelled (Schedule, Next run, Last run) and the model is left out.

The Model column shows which AI model the agent runs on — you pick it in the footer of the agent form's Instructions box. The available models come from your organization's Model Catalog (the set your admin has enabled), so the list reflects the same models you see in chat. See Models for how the catalog works.

Status words

The same words appear in the list and on an agent's own page; hover one for what it means.

• Scheduled — runs on its schedule.
• On demand — has no schedule; it runs when you press Run now or call its webhook.
• Running — a run is going on now, with how long it has been running. A test run shows the same way.
• Paused — the agent is switched off and won't run on schedule: you paused it, or Strata paused it because its model is no longer allowed (see the note below). Resume starts it again, unless its model still isn't allowed for you. Run now still runs it.
• Expired — past its expiration date, so it no longer runs. Edit it to extend the date.
• Stopped — Strata stopped it, for example because its database connection was removed or its owner's account was deactivated, and it won't run again until it is edited. The reason shows under its name.
• Failing — its last five runs failed in a row, so Strata backs it off (a run refused because its model isn't allowed doesn't count as a failure here). It is not stopped: about once an hour the scheduler lets a single probe run through, and the first probe that succeeds clears the streak and returns the agent to Scheduled on its own. Fixing the underlying issue (or editing the agent) just lets that probe succeed sooner — you never have to re-enable it by hand.
• Failed — its last run ended in an error or ran out of time. It is still scheduled to try again.
• Skipped — its last run did not start, for example because a limit was reached or the AI service was busy. It runs again on its schedule.
• Interrupted — its last run was cut off before it finished, for example by a restart.

Needs attention, above the list, gathers the agents that are Failing, Failed, Stopped, Expired, Skipped or Interrupted.

Note

Agents also stop running org-wide while a subscription is inactive. Runs are skipped rather than failed — they show as Skipped in the run history (Skipped — not started when you open one), and the run records "Organization's subscription is inactive — agents are paused until billing is resolved". Nothing is lost: once an admin restores billing at Organization → Billing, agents resume on their normal schedule without being re-armed.

Note

An agent is paused if its pinned model is no longer allowed for you and a run you didn't start by hand finds it: its schedule, its webhook, or another agent invoking it. That happens when an admin turns the model off for your organization (Strata first shows them which agents will pause, and whose, and asks), or when your role, a role one of your groups gives you, or your own model access now leaves the model out (the save warns the admin which agents it will pause; a change your organization's identity provider makes through SCIM doesn't). The run fails, the agent shows Paused, and your Inbox gets a notice titled <name> stopped (Stopped · its model is not allowed) that names the setting leaving the model out and where an admin can allow it (your role or group roles in Organization › Roles, your own model list in Organization › Users, or your organization's list in Organization › Models), lists the models you can switch the agent to (or, when none of the models left to you is a chat model, says an admin can allow it, or another chat model), and ends This agent is paused until then; press Resume on it to start it again. The run's error in the run history ends the same way. Run now is refused for the same reason but pauses nothing: the Agents page shows Could not start run with the same sentence, the run is recorded as failed, the agent is left as it was, and your Inbox shows <name> failed. Either notice arrives beside a report you haven't read yet rather than replacing it. The same notice isn't posted again within 24 hours, so pressing Run now again doesn't stack them. Once an admin allows the model for you where the notice says, or you edit the agent to use a model that's still allowed, press Resume. Until then Resume is refused with Could not update agent: This agent can't resume. <Model> isn't available to you: your role doesn't allow it. Pick one of: …. Edit it to use another model, then resume it. (the reason can also be your organization doesn't allow it or it isn't in your own model list). If Strata couldn't read the roles your groups give you and, without them, the model isn't allowed, it pauses nothing: the run is recorded as not started, a scheduled run tries again in 15 minutes, and Run now shows Could not start run with Could not check which models this agent's owner may use. Try again in a moment.

Starred agents show a small star icon next to their name. Use Star / Unstar in a row's ... menu to flag the agents you check most often.

Row actions

Each row's action cell has three controls:

• Pause / Resume — toggles the enabled flag. A paused agent stops appearing in the scheduler poll; its next_run_at is cleared. Its webhook answers 409 Agent is paused, and another agent that tries to invoke it is refused with The agent "<name>" is paused. Its owner can resume it on the Agents page. Run now still runs it.
• Run now — kicks off a manual run immediately, independent of the schedule. The run is tagged trigger_type=manual in run history. If Strata can't start it, the page says Could not start run with the reason instead of Run started: the agent's model isn't allowed for you, it is already running, you've reached your organization's daily limit (You've reached your organization's limit of N agent runs per member per day. Runs start again at midnight UTC.), the subscription is inactive, or the organization is suspended. A run that hasn't started a minute after Run started is reported as The run did not start.
• The ... menu — Edit, Duplicate, Test run, Star / Unstar and Delete. Edit and Duplicate are shown but unavailable, with the reason, for an agent created from a locked organization template and when agent changes are turned off for you. Test run runs the agent once without the tools that send, save, write or run other agents (see Dry runs).

The agent's name is a link to its own page, /ai-tools/agents/<agent-id>, so it can be opened in a new tab or bookmarked; clicking anywhere else on the row opens it too. The agent's page shows its status and next run, what is wrong when something is (with View run, Run again and the fix), its runs (Runs, Success rate, Avg duration and Total spend for the runs shown, then Recent runs with All and Failed and Show more), its instructions and its settings, and, closed at the bottom, Webhook trigger. Its bar has Pause / Resume, Run now, Edit (which opens the agent form) and a ... menu with Duplicate, Test run, Star and Delete. A run you open has an address of its own, /ai-tools/agents/<agent-id>/runs/<run-id>, which is where the Inbox's View this run leads.

Deploy your first agent

Click Create agent at the top right of the Agents page. The split button gives you two choices:

• Create a custom agent — opens the agent form with empty defaults. See /docs/agents/create.
• Create from a template — opens the Templates view, listing every template you can start from. Templates published by your admin appear under From your organization; the Kronisys-curated set appears under Built-in templates. See /docs/agents/templates.

You can also ask for an agent in a chat in the web app ("every weekday at 8am, summarize my unread client emails"). Strata drafts it as a card, and nothing is created until you choose Create agent. See Create an agent from a chat.

The current built-in templates are:

Org limits on agents

Your admin can constrain how many agents you can deploy and how often they can run, under Organization → Agents:

• Agents per member — a per-org cap. Hitting it returns Agent quota reached: your organization allows up to N agent(s) per user. Delete an existing agent to make room. (HTTP 429). A limit of 0 blocks new agents altogether: Agent quota reached: your organization currently blocks new agents. Contact your administrator.
• Minimum time between runs — the shortest cadence the org allows. A schedule below the threshold returns Schedule is too frequent: your organization requires at least Ns between runs (this schedule fires every ~Ms). Pick a less frequent cadence. Independently of any org setting, Strata enforces a platform floor of one run every 5 minutes — a faster cadence is refused even when your admin has set nothing, with the message Schedule is too frequent: scheduled agents run at most once every 5 minutes.... An admin can only make the interval longer, never shorter than this floor.
• Runs per member, per day — a cap on how many runs a single user's agents may start in a day (counted since midnight UTC). Once it's reached, further runs are skipped for the rest of the day. This counts every trigger — scheduled runs, webhooks, manual Run now, test runs, and linked-agent (invoke_agent) runs alike — so hitting the cap by hand starves your own scheduled runs too. The count resets at midnight UTC. Run now past the cap says so: You've reached your organization's limit of N agent runs per member per day. Runs start again at midnight UTC.
• Cost ceilings — admins can also cap what a run may consume: Tokens per run and Spend per run stop a run mid-execution once reached, and Monthly spend per agent skips that agent's runs until the calendar month resets. These are what make short (every-few-minutes) schedules safe to allow.
• Freeze agent changes — an org-wide switch that blocks creating and editing agents for everyone, admins included. Existing agents continue to run on schedule, and you can still pause them.

Note

Self-serve trials carry two limits your admin cannot raise. For the length of the trial, each member may deploy at most 2 agents, and every schedule is floored at one hour between runs. Both are clamps rather than replacements: an admin who set something tighter still wins. Both lift when the trial converts to a paid subscription, and neither applies to legacy indefinite-trial organizations.

See /docs/admin/rate-limits for how admins configure these and /docs/agents/per-user-governance for per-user restrictions that apply on top.

Where to go next

• /docs/agents/create — the agent form, end to end.
• /docs/agents/schedules — friendly options, custom cron, timezones, expiration.
• /docs/agents/notifications — getting emailed when an agent runs.
• /docs/agents/webhooks — triggering an agent from an external system.
• /docs/agents/permissions — what tools an agent can use and how to keep it safe.
• /docs/agents/templates — org-wide templates published by your admin.

Related

• Create an agent
• Templates
• Schedules
• Notifications