Manage conversations

Suppose you explore two directions for an ad, then move on to delivery checks. Give each piece of work its own conversation:

• Perfume ad | Cool direction | Storyboard
• Perfume ad | Warm direction | Storyboard
• Final delivery | Captions and checks

The conversation list lets you reopen any conversation and review its instructions, Agent activity, and results. The nodes and files remain on the same canvas, so a new conversation can still reference approved material.

When should you start a new conversation?

• The new task is unrelated to the current direction.
• The old conversation contains conflicting requirements.
• You want to keep the approved result without carrying every historical instruction forward.
• Different deliverables need independent records.

Add the nodes and attachments this task needs, such as the current brief, approved script, and product reference.

Agent does not decide which instructions from an older conversation still apply. Start with one sentence that states the new goal, then reference only the current material.

When should you branch?

Use a branch when the earlier discussion and source material still apply, but you want to try a different next step:

• Try live action and animation from the same script.
• Keep the same character but explore another setting.
• Create different durations from an approved storyboard.

A branch keeps every message up to the point where you branch. The original conversation remains available, while the new direction gets its own history.

If the goal, source material, or deliverable has also changed, start a new conversation instead.

What should you save to the canvas?

Save approved directions, scripts, character standards, product rules, and final settings as canvas nodes or files:

1. Separate final decisions from discussion.
2. Label nodes as approved, under review, or old.
3. Reference only current material in a new conversation.
4. Keep rejected options for history, but do not use them as generation input.

When you open a new conversation, reference these nodes directly. Agent can use the current version without making you search through the full conversation history.

What can you do while Agent is running?

• Stop when the goal or reference is wrong.
• Add to Queue by sending the next instruction while Agent is busy. It will wait in Queue and run after the current task finishes.
• Keep organizing the canvas by moving nodes, checking references, or preparing material for the next step.

For example, while Agent is generating four storyboard frames, you can send:

After all four frames are ready, name them Shot 01 through Shot 04 and arrange them in one row on the canvas.

The instruction appears in Queue without interrupting the active generation. If you add several instructions, Agent processes them in the order you sent them.

If the current task uses the wrong goal or reference, stop it first, then correct the material and instruction. Avoid queuing contradictory requests, such as keeping a background in one message and removing it in the next.

What should you check after a task fails?

1. Read the task status and the error shown on the page.
2. Check the network connection and available Tapies.
3. Confirm that referenced nodes and attachments are still accessible.
4. Reduce the task to one generation step.
5. Change the model or settings based on the error, or send the task again later.
6. Keep the error details and canvas link when contacting support.

Next: Apps