Chains
Chains
A chain carries work that takes longer than one sitting, step by step (module chains). Each step's outcome is kept before the next starts, so a chain carries on after a restart or a deploy without repeating what it already did. A chain waits on purpose (what for, how often it looks, when it gives up), retries a failing step a stated number of times, and asks a person where its playbook says so. When it cannot go on, it stops and says why; it never stalls silently.
A playbook is a named, versioned list of steps a chain is started from. Every automation run is a chain.
In the sidebar under Automation → Chains.
What you see
Chains
Each chain runs a playbook step by step. The list shows the Playbook and Version it started on (it keeps that version to the end), its State (Running, Waiting for you, Finished, Cancelled or Stopped), the step it is on, and when it Started and last Changed.
Open a chain to see every step with its state, Attempts (each try or look, with its time and what happened), what went in, what it produced (later steps may use it by name), what it is waiting for, and what comes next. A ✓ on a step means it ran under the approval given when its automation was turned on, rather than asking each time.
Playbooks
The playbooks here, each with its key (the name an automation or agent uses to start it), versions and steps. With none yet, the page says how to make one: ask an agent in Chat, for example "Make a playbook that checks my shops every morning and tells me on Telegram."
What you can do
Answer a chain that waits for you
Open the chain. Where it stopped, press Try again (retry the step, or wait longer), Skip this step, or Cancel the chain. Only the choices its stop offers are shown.
Cancel a chain
Press Cancel the chain. It stops at its next step boundary: a step already acting finishes, and nothing new starts.
Start a chain
Chains start from an automation, or from an agent using a playbook. Every step then asks for approval as it would if you called its tool yourself, unless its automation's Turn on approved it.
Who can do what
| Permission | Label | Lets you |
|---|---|---|
chains:read | View chains and playbooks | See chains, their steps and playbooks. |
chains:write | Start, resume and cancel chains | Start a chain, answer one that waits, and cancel one. |
chains:playbooks | Write and change playbooks | Save playbooks and new versions of them. |
What happens on its own
| Job | When | What it does |
|---|---|---|
sweep | Every minute | Starts, once, any chain that was written but never reached the engine, for example after a crash. A chain waiting for you sends its own reminder after a week. |
Needs
Durable runs, Approvals and Projects switched on.
What agents can do
Read-only tools run at once; the rest ask first under the default rules.
- Read-only:
chains.playbooks,chains.playbook,chains.list,chains.status(where a chain is and what it waits for) - Change:
chains.savePlaybook(refused, with every reason, when a step lacks its wait or retry limits or names a tool this instance does not have),chains.start(as yourself, with its inputs),chains.resume(retry, skip or cancel, as its stop offers),chains.cancel(stops it)
Over MCP the tools are named chains__list and so on.