What Agent can do in Nodespell

Nodespell Agent works through a fixed set of actions. It can read your graph, search the catalogs, edit nodes and connections, work with Creative Tools and Assets, and run parts of the graph with your approval. This page lists every action group, with an example request for each, and the things Agent does not do.

Agent capabilities at a glance

In the table, “Changes the graph” means the action works only in Edit mode. “Needs approval” means Agent always asks first, even in Edit. See Modes and approvals.

What you can ask for	Example request	Changes the graph	Needs approval
Explain a graph, a node, or a chain	“Explain what this graph does.”	No	No
Find nodes on the current Level or anywhere in the Project	“Find every node that uses the portrait prompt.”	No	No
Trace connections upstream or downstream	“What feeds into @Final grade?”	No	No
Count and summarize the graph	“How many image models are in this Project?”	No	No
Find a node type or Creative Tool to use	“Find a faster video model for this branch.”	No	No
Read a node’s or Creative Tool’s settings and models	“What parameters does this upscaler have?”	No	No
Create nodes	“Create an image node from this prompt.”	Yes	No
Duplicate a selected chain, up to 20 copies	“Duplicate these nodes three times with a different direction.”	Yes	No
Rename nodes	“Rename the shots Shot 01 to Shot 04.”	Yes	No
Change parameter values	“Set the aspect ratio to 9:16 on both image nodes.”	Yes	No
Connect or disconnect nodes	“Connect this audio into the video model.”	Yes	No
Delete nodes	“Delete the unused upscaler.”	Yes	Yes
Create a Creative Tool	“Add an image-to-video Creative Tool after @Hero still.”	Yes	No
Check what a model switch would drop	“What would I lose by switching this tool to another model?”	No	No
Switch a Creative Tool’s model	“Switch this tool to a cheaper model.”	Yes	Yes
Find Project Assets and place them on the graph, up to 20 at a time	“Add the three logo files from Assets to the graph.”	Yes	No
Check a node’s results and review result media	“Did the last run of @Hero shot produce anything usable?”	No	No
List recent runs and their status	“What’s still running?”	No	No
Estimate the credit cost of running nodes	“Find the most expensive step in this graph.”	No	No
Prepare a run with one to four outputs	“Get two takes of @Hero shot ready to run.”	No	Yes
Start a prepared run	“Start it.”	No	Yes
Follow a run’s progress	“Tell me when the video finishes.”	No	No
Cancel a run	“Cancel the run you just started.”	No	Yes
Ask you a question with choices	Agent asks when your request is ambiguous.	No	No

Starting a run does not edit the graph, but it spends credits. For the full generation flow, see Generate with Agent.

What does Agent not do?

These limits come from the actions Agent has. Agent can’t work around them, even if you ask.

Agent does not	What to do instead
Spend generation credits without your approval	Approve the run when you’re ready.
Change the graph in Ask or Plan	Switch to Edit.
Arrange or tidy the layout for you	See Organize the graph. Agent places new nodes automatically.
Edit clips in the Timeline or layers inside a compositor	See Timeline and Compositing.
Use Snippets or Custom Nodes	See Snippets and presets.
Import images or video from the web	Upload the file and attach it. See Context and references.
Retry a failed run by itself	Ask Agent to prepare the run again, and approve it.
Listen to audio	Agent can read a transcript when one is available.
Keep working if you switch Level or Project mid-request	Start a new request in the Level you want.
Run two requests at once in one Project	Queue the follow-up, or wait.

Agent works only inside the open Project. It can’t read or change other Projects.

Tips for requests that work

• Name the target. Include selected nodes, or mention a node with @, so Agent doesn’t have to guess.
• Pick the mode first. Use Ask for questions, Plan to review a change before it happens, and Edit to make it.
• Say the outcome and the constraint. “Make a 9:16 version of @Hero shot and keep the prompt” works better than “make it vertical”.
• Ask for an estimate before a big run. “What will running this cost?” works in every mode and spends no generation credits.
• Give one task per message. Queue the next step rather than putting several unrelated tasks in one request.
• Use attachment roles. Mark a face reference as Character and a look reference as Style.
• Check the steps. Each action Agent takes appears in the conversation. Select a node chip to find that node on the canvas.

Frequently asked questions

Can Agent build a whole workflow from a description?

Yes, within the actions on this page. In Edit mode, Agent can create nodes and Creative Tools, set their parameters, and connect them. Running the result still needs your approval.

Can Agent use any model in Nodespell?

Agent searches the same node and Creative Tool catalogs you see in Nodespell. It uses models that are available to you. Running them still depends on your plan and credits.

Can Agent see the images and videos it generated?

Agent can check a node’s results and inspect up to four results at a time, sampling a few frames from each video. It does not listen to audio. Treat Agent’s comments on how a take looks as a starting point, and review the take yourself in the Results tab.

Can Agent delete my nodes?

Only with your approval. Agent asks before every deletion, and you can undo a deletion with Cmd/Ctrl+Z.

Related

• Agent overview — open Agent and send a first request.
• Modes and approvals — when Agent asks first.
• Context and references — mentions and attachments.
• Generate with Agent — estimates, runs, and takes.
• Creative Tools — the tools Agent can create and switch.