Routines and webhooks
Run agents on a schedule, collect their reports in one thread, or trigger work from an external event.
Routines
A routine runs once, at a repeating interval, or on selected weekdays. Each run starts a fresh execution using the selected bot's model, permissions, tools, and computer configuration.
Run receipts distinguish queued, active, waiting, completed, missed, failed, and cancelled work. GlinkBot must be running when a local routine becomes due.
Results arrive in the bot's chat
Each run works in a conversation of its own, out of the way, and only its result comes to the bot's chat: one message from the bot, with a small line above it naming the automation and when it ran. Reply to it like any other message; the bot knows what its automation found. A run with nothing to tell you stays quiet: no message, no unread dot, no notification, and the automation's card says Nothing new.
When a run needs you, it asks in that chat. A question in words shows needs you on the message; answer it there, or react 👍 or 👎. A choice or an approval the run waits on appears in the chat too, and answering it there, or reacting 👍 or 👎, carries the run on. A sign-in, a connection or a key it needs is named in the chat, with See run to it. When a run can't finish, the chat says so once, with what to fix.
Nothing lands in the middle of your own conversation: if you are chatting with the bot when a run finishes, its message waits and appears right after the bot's reply to you.
Files a run links in its message open from the chat, just as they would in the run's own conversation. Use See run on a message for the run's full conversation, then Back to results to return. Run logs keeps the execution history, including failures and waiting runs. Background run conversations stay out of the normal sidebar and mobile thread pickers.
A bot schedules, changes, pauses, resumes, and deletes its own routines from chat without asking, and each change shows as one line with Undo (see Changes a bot makes). A routine for another bot asks first unless the conversation has Full access.
Without a chosen thread, each run reports into the bot's main thread as it is when the run starts, wherever the routine was made, so results follow you to the conversation you are in. Changing the destination affects future runs, not the messages older runs left. If you delete a results thread, its hidden execution threads become visible so you can still inspect their work and respond to approvals.
Remember the last run
Each run starts fresh by default. For work that builds on last time, such as a QA pass, monitoring or follow-ups, turn on Remember the last run under the routine's instructions. Each run then starts with the previous run's report (up to 2,000 characters), so the bot knows what it found before. It is available for repeating routines that one bot runs; team goals don't carry it yet. You can also ask the bot in chat to turn it on.
Run when the app starts
If your computer is often off when a routine is due, open the routine's Advanced options and turn on Also run when GlinkBot starts. The routine then runs once each time GlinkBot opens, as well as on its schedule. Its run logs show these runs as On startup. A routine that GlinkBot is already catching up on at startup runs once, not twice, and an occurrence missed by more than 12 hours is covered by the startup run instead of being reported as missed. Switching to another workspace and back, or GlinkBot recovering from a crash, doesn't count as opening the app again. Pausing a routine stops its startup runs too. It's available for repeating routines, and you can also ask the bot in chat to turn it on.
A routine's own model
A routine runs on its bot's model unless it has its own. Give it one through the API with modelSelection (the same shape as a bot's: instanceId, model, and an optional effort or variant), for example a light hourly check on a smaller model while the bot chats on a stronger one: PATCH /api/routines/<id> with {"modelSelection": {"instanceId": "claude", "model": "claude-haiku-4-5"}}. Each run's thread then uses that model; if its AI isn't available, the thread falls back to the bot's model. The model has to keep the bot's permission level, as a bot's own model change does: with Custom approval it must run on the bot's own engine, and with Auto-accept edits on an engine that has that level. Other models are refused, and if the bot's permissions change later so the routine's model no longer fits, runs use the bot's model. Room goals always use their bots' own models, and a limit-reset routine stays on the model of the conversation it continues. Send "modelSelection": null to go back to the bot's model.
When the AI's usage limit stops a run
If a bot's AI reaches its usage limit during a run and the limit resets within 12 hours, GlinkBot starts that run once more a minute after the reset, and tells you so. A repeating routine's scheduled run is the exception when the routine's next scheduled run comes no later than one of its intervals after the reset (an hour for an hourly routine, a day for a daily one): that next run does the same work, so the stopped run fails with the usual usage-limit notice instead. Runs you start with Run now, one-time routines and routines that run when a limit resets still start again at the reset, and a bot with a backup AI still switches to it at once.
Clear a folder's unread dots
Right-click a folder or open its … menu, then choose Mark folder as read. This clears unread dots without opening each thread. It does not answer approvals, dismiss questions, or stop any running work.
Webhook triggers
Webhooks start the same queued task executor from an external HTTP request. They are independent from schedules and use a dedicated receiver on 127.0.0.1:8800 by default.
The receiver exposes only health and secret hook routes. It does not expose the broader GlinkBot API.
Bearer authentication is preferred because it keeps the secret out of URLs and most access logs. A capability URL is available for senders that cannot set headers.
A bot that sets up its own webhook
Reaching a local webhook
On the desktop app the receiver listens only on this computer. To accept events from the public internet, proxy only the webhook receiver through a narrowly configured relay or tunnel. Do not expose the main harness port.
GlinkBot documentation, reviewed for the 0.0.4 codebase. Public download availability can differ. Licenses and acknowledgments.