An Agent session in Nodespell is a conversation. It keeps your requests, Agent’s replies, approvals, and results together for one Project. You choose which AI model Agent thinks with, and that choice affects speed, quality, and credit use. While Agent works, you can stop it, redirect it, or queue the next request.
Manage conversations
Section titled “Manage conversations”The header of the Assistant panel shows the current conversation’s title. A new, untitled conversation is called New chat. Agent titles a conversation from your first message.
| Task | How |
|---|---|
| Start a conversation | Select the + button (New chat) in the header. |
| Switch conversations | Select the title to open the list, then pick a conversation. The list is grouped into Today, Last 7 days, and Earlier. |
| Search | Type in Search conversations at the top of the list. The search matches titles, including archived conversations. |
| Rename | Open Conversation options (…) and choose Rename. Type a name of up to 48 characters, then press Enter to save or Escape to cancel. |
| Archive | Open Conversation options and choose Archive. The conversation moves to the Archived submenu at the bottom of the list. |
| Restore | Open the Archived submenu and select the conversation. |
| Delete | Open Conversation options, choose Delete conversation, and confirm. |
Archiving only hides a conversation. It doesn’t change the graph or its takes. You can’t switch, rename, archive, or delete conversations while Agent is working.
Choose the Agent model and reasoning effort
Section titled “Choose the Agent model and reasoning effort”The model menu sits at the right of the message box. It shows the current model’s name. The menu groups models by cost, cheapest first. Hover over a model to see what it is for and its ratings for Speed, Accuracy, and Cost.
| Group | Typical use |
|---|---|
| Cost effective | Quick, low-cost replies for simple questions and small edits |
| Regular | Everyday graph building and planning. One model here is marked Recommended. |
| High quality | Long, multi-step work and hard problems. Slower, and costs more. |
Agent starts with a model from the Cost effective group. Nodespell remembers your choice in this browser. The list of models can change over time, so rely on the menu rather than a fixed list.
When the selected model supports it, the menu also shows Reasoning effort, with Low, Medium, High, and Extra high. Medium is the default. Higher effort can take longer and use more credits. The model menu is hidden while Agent is working.
What does Agent use credits for?
Section titled “What does Agent use credits for?”Agent uses credits in two separate ways.
| Use | When it is charged | Where you see it |
|---|---|---|
| Agent reasoning | For each request, based on the Agent model and how much work the request takes | Usage tab of the Tasks panel, listed as Editor Agent |
| Generations | Only after you approve a run, like any run you start yourself | The Tasks panel, like other runs |
Agent needs a small available credit balance before it can start a request. Each request also has a spending cap. If a request reaches the cap, Agent stops, and the usage entry shows Budget reached. Per-model prices change, so this page doesn’t list them. See Plans and credits for balances and plans.
Stop, steer, queue, or resume a request
Section titled “Stop, steer, queue, or resume a request”While Agent is working, the message box changes. Its placeholder reads Add a follow-up, or steer the current work.
| Control | What it does |
|---|---|
| Stop | Stops the current request. Completed changes are kept, and work already submitted may still finish. |
| Steer | Appears when you have typed something. It stops the current work and continues with your new message instead. |
| Queue | Sends your message after the current work finishes. Pressing Enter while Agent is working also queues the message. |
| Queued next | Lists queued follow-ups in order. Remove one with its × button. You can queue up to eight. |
| Resume | Appears after a stopped request. Agent rechecks the current graph and finishes only the unfinished steps. |
| Reconnect / Retry | Appears if the connection was interrupted. It recovers Agent’s saved answer. |
-
To redirect Agent mid-request, type the correction, for example “Use 9:16 instead”.
-
Select Steer. Agent stops the current work and continues with your correction.
-
If you stopped too early, select Resume to finish the remaining steps.
If Agent fails three times in a row, it pauses with The assistant paused after repeated failures. Edit the prompt before trying again. Rewrite the request before you send it again.
Only one Agent request can run in a Project at a time. If you see Another assistant request is already running in this project., stop that request or wait for it to finish.
Frequently asked questions
Section titled “Frequently asked questions”Does deleting a conversation undo Agent’s changes?
Section titled “Does deleting a conversation undo Agent’s changes?”No. Deleting a conversation removes only its history. Graph edits and takes stay in the Project, and you undo edits in the graph.
Which Agent model should I use?
Section titled “Which Agent model should I use?”Start with the Recommended model in Regular for most graph work. Use Cost effective models for quick questions, and High quality models for long, complex tasks.
Does Stop cancel a generation that already started?
Section titled “Does Stop cancel a generation that already started?”Not always. Work already submitted may still finish. Ask Agent to cancel the run, or cancel it in the Tasks panel.
How many follow-ups can I queue?
Section titled “How many follow-ups can I queue?”You can queue up to eight follow-ups while Agent is working. Agent sends them in order after the current request finishes.
Related
Section titled “Related”- Agent overview — open Agent and send a first request.
- Generate with Agent — the credits a generation uses.
- Tasks panel — the Usage tab and running work.
- Plans and credits — balances, plans, and estimates.
- What Agent can do — what a request can include.