Playbooks
Save a graph of agent nodes with gates between them, then run it as one unit from any thread.
A playbook is a saved graph of agent nodes joined by edges, with optional gates between them. You run it as one unit.
Parts of a playbook
| Part | What it is |
|---|---|
| Node | One agent run. It picks an installed agent and model, and gives it a role and an instruction. |
| Edge | Connects two nodes. The upstream node's output becomes the downstream node's context. |
| Gate | A pause on an edge: a verify command, or your approval. |
| Loop | An edge that sends work back to an earlier node, bounded by an iteration cap. |
What a run handles for you
- Automatic handoffs between nodes.
- Verify gates and human approval gates.
- Iteration caps and timeouts.
- A fresh session if a node's agent crashes.
- A transcript for each node that you can inspect.
- Re-running from a chosen node.
A playbook run appears on the Fleet board like any other thread.
Palisade never swaps in a different agent when a node's agent is unavailable.
Running a playbook
Type |= in any thread to run a saved playbook.
Use Test run on the canvas to try it out.
Editing on the canvas
- On an empty canvas, select Add your first node.
- Connect nodes by clicking one port, then another.
- Select a node or edge to open its inspector.
- Edits are staged. A bar reads Unsaved changes with Discard and Save.
- Discard reverts to the saved playbook.
- Validation problems appear once you have unsaved changes.
From the Playbooks panel you can Duplicate a playbook or Delete it after a confirmation.
Where playbooks live
Playbooks are project-scoped.
They are saved as JSON under .palisade/chains/ in your repository, so they ship with the repo.
Your team gets the same playbooks when they clone it.
Not yet built
- Scheduling and event triggers.
- Drag to connect from a port.
- Edge labels and multi-select.