Appearance
AI nodes
The AI section of the palette has two kinds of node. Agent is a full AI turn: a brain, a prompt, a budget, and the tools to read and write your workspace. The Quick AI actions each do one small job and hand their text to the node after them.
Both need AI to be available for the workspace. A node that cannot run is shown locked in the palette with the reason: AI is off, or no model can call tools. See AI settings.
Agent
| Node | What it does |
|---|---|
| Agent | Runs an AI agent with a prompt: it reads and writes tasks, notes and fields over as many steps as it needs, as Truetask AI or as one of the board's agents. |
The brain
Agent is the one control that says who the step runs as.
- Leave it on Truetask AI and the step runs as the house voice, on the workspace's models.
- Pick one of the board's agents and the step runs under that agent's name.
- A model picker sits beside the brain, so one automation can use a different model from another.
An agent that was set up on a computer through the desktop app is different: the step runs there, with that computer's coding tool, model and effort. The node says so in amber when you pick one. That computer has to be on, signed in and connected every time the automation runs, or the run fails. See Desktop agents.
The prompt
Write what you want done. The agent reads the task and the board, calls the tools it needs, and ends with a message. Placeholders such as {card.title} and {card.description} fill in from the task that fired the automation. See Placeholders.
The budget
| Field | Default | Range |
|---|---|---|
| Turns | 8 | 1 to 20 |
| Tokens | 8000 | 500 to 50000 |
| Seconds | 60 | 5 to 600 |
A run stops with whatever it has when a limit is hit. On top of the budget, one Agent node may make at most 20 write calls in a single run.
What the agent can do
The Agent node is granted every automation tool: reading and writing tasks, lists, tags, checklists, comments, custom fields, relations, notes, milestones, timers, dependencies, attachments and files, plus searching the workspace and fetching a URL.
It is never given the destructive ones. Deleting tasks, tags, custom fields, comments and checklist items is off the table, because an automation has nobody to confirm with. A run as Truetask AI also skips the tools that only a board agent may call.
The output
The agent's final message is available to the nodes after it as {agent.output}. Add it as a comment, write it to the description, or put it in a field. To write it to a different task, put a Target Card node in between.
Quick AI actions
| Node | What it does | Where the result goes |
|---|---|---|
| Smart Hydrate | Fills in a sparse task from its title: a description and tags, following your prompt. | Applied straight to the task |
| Generate | Writes text from a prompt and the task's context. | {ai.output} |
| Suggest Tags | Proposes tags for the task from its content, using the board's own vocabulary. | {ai.output} |
| Improve Title | Rewrites the task's title to be clear and specific. | {ai.output} |
| Summarize | Summarizes the task, its comments or a note into a short text for the steps after it. | {ai.output} |
Every one of them has a single optional Prompt field: extra instructions appended to the node's own default prompt. Leave it blank and the default does the work. Use it for board-specific tone, for example "use frontend tech vocabulary".
Except for Smart Hydrate, these nodes only produce text. They do not write anything themselves, on purpose: chain them into the action that should write.
| Node | Pair it with |
|---|---|
| Generate | Set Description, Add Comment or Write Note |
| Suggest Tags | Apply Tags |
| Improve Title | Set Title |
| Summarize | Add Comment, or anything that takes text |
Smart Hydrate
Smart Hydrate is the one that applies its own result. One AI call writes a description and picks tags in a single update to the task.
It writes a description only when the task does not already have one, and it prefers tags the board already uses, creating a new one only when nothing fits. Each half can be switched off workspace-wide under Settings > Integrations > AI, in the Features section: Suggest card descriptions and Suggest card tags.
Smart Hydrate is also what the Hydrate new cards starter automation runs. See Builder.
Where the models come from
Models are configured under Settings > Integrations > AI. The Models section is the list of models the workspace can use, and Routing picks which model serves each capability.
| Capability | Used by |
|---|---|
| Generate | Smart Hydrate |
| Tool use (agent) | The Agent node, and the Generate, Suggest Tags, Improve Title and Summarize nodes |
A model chosen on the Agent node itself wins over the routing.
Truetask Cloud only
Truetask AI is the managed option: models are included with your plan, with nothing to configure and no API key. Usage counts against a monthly allowance shown on the AI settings page. You can still add your own provider models alongside it. See Plans and billing and Models.
Self-hosted only
There is no managed option. Add your own provider models under Settings > Integrations > AI and route each capability to one of them. Tool use needs a model with real function calling: a weak caller invents tool names and the agent goes nowhere.
Editing the default prompts Self-hosted
Every AI node has a default prompt behind it, stored in the workspace rather than hard-coded. Workspace administrators can edit them in the Prompts dialog. The Smart Hydrate node's config panel has a Customize hydrate prompts link straight into it.
The link only appears when you have permission to manage prompts and your workspace has prompt customization. On Truetask Cloud the defaults are fixed, and the per-node Prompt field is how you steer them.