Onboarding Agent and Story Agent
Starting a project from source material with the Onboarding Agent, asking and delegating inside a project with the Story Agent, and how proposal review keeps you in control.
Tutorial: Start a Project with the Onboarding Agent
The former Onboarding Agent panel is retired. Use Thruline Agent in the project workspace for guided setup; there is no separate Onboarding Agent switch or launcher.
You can start setup from Project Overview without first opening Documents, Storyboard, Canvas, or Production. The agent loads the project data it needs before continuing. If loading fails, restore the connection or reload the project and retry.
Ask Thruline to set up a project using your brief and chat attachments. It reviews the sources, asks which Script or asset foundation you want, then asks whether to generate images and confirms visual style. Use the normal import and creative reviews to apply changes and edit image prompts and model settings. Workflow checklists are saved in this browser and account; accepted work keeps its existing recovery records. After importing, Thruline asks whether you want reference images before preparing them. Existing styles in populated asset categories are reused; only missing styles need an answer. Your supplied wording is prepared directly for review and saved when you apply it. You can ask for unrelated work during setup. Standalone actions do not advance its checklist; only actions explicitly linked to setup return its next-step guidance. Pause holds the next action and preserves your review edits. Resume continues unfinished work. Cancel stops further actions and requests cancellation of the current workflow and known generation jobs; saved changes remain, and already submitted jobs may still finish. When applying a prepared Script, the Episode dropdown shows the new episode title or lets you choose an existing episode whose Script will be replaced.
After images and supplied references are handled or skipped, Thruline offers to turn your Script into planned Shots and Sequences. You can leave this for later. If you continue, the Generate storyboard card asks for maximum seconds per Shot and confirms replacement before automatically applying the storyboard. It does not generate Shot images or videos. An assets-only project needs a Script first.
Give it what you have: paste or upload a script, drop in notes, attach reference images, or describe the idea in a sentence. It reports what it recognized under What I found and asks a clarifying question if something is ambiguous.
Choose your project starting point. Depending on your material it offers to use your script as is, to write a script, to produce an ordered scene outline, or to set up assets only.
Confirm the Script stage. The first episode's Script Doc is written, Scenes, Characters, and Props are identified, and Documents opens so you can read the result while the agent waits.
Choose the visual direction. It proposes a Character, Scene, and Prop style from your material; edit the wording or pick another suggestion.
Confirm the Visuals stage. Uploaded images are placed on the first episode's Canvas and, when they clearly match an asset, tagged to it. Selected assets get generated reference images, funded by the platform rather than your credits, which arrive on the board as they finish.
Complete onboarding closes the panel. The launcher in the bottom-right corner keeps the transcript so you can reread what was done.
The Story Agent
Thruline Agent replaces the retired Workspace Story Agent. Open it from the workspace header to work with your current project and selection; there is no separate Story Agent switch or launcher.
What It Can Do
| Term | Meaning |
|---|---|
| Answer and brainstorm | Discuss story, structure, character, and visual ideas in plain conversation with your project as context. |
| Read the project | Look up exact Scenes, assets, Shots, Sequences, and generation history to answer questions such as which Scenes still have no approved video. |
| Prepare script edits | Draft rewrites, new Scenes, or a new episode from source material as a preview you can read before anything is written. |
| Prepare generations and exports | Set up an image or video generation, or a Project files export, with the model, references, and estimate frozen for your review. |
| Take you somewhere | Offer to open a Scene, Sequence, or asset. Navigation is also a proposal, so it never moves you unexpectedly. |
In Documents, selecting text and choosing the Story Agent affordance opens the same chat with that passage attached, which is the quickest way to ask for a rewrite of one speech or one Scene.
Create a storyboard with Thruline
When the Storyboard workflow tools are enabled, ask Thruline to storyboard an episode. The original Storyboard wand remains available.
Choose an existing Script, import one through the document review, or ask Thruline to develop your story or plot into a Script. Empty and starter Scripts need story content before planning.
Generated Scripts receive the same asset analysis as imports before Storyboard. Accepted Scenes, Characters and Props are registered for the episode; uncertain candidates remain for review. Wait for registration to finish, or use Retry or Cancel. Existing asset images and descriptions are preserved.
When the Script is ready, the configuration card opens directly. Enter the maximum seconds per Shot there; the field has no model-based upper limit. Actual Shot lengths vary with the story. The planner also links recognized episode assets in Shot descriptions; this does not attach or generate asset images.
Confirm the episode replacement before planning starts. The Storyboard Agent automatically applies its validated Sequences and Shots, then refreshes the board. There is no second Apply step.
Reviewing Proposals
Agents never write, spend, delete, or navigate silently. Anything with an effect arrives as a proposal card in the chat with the exact change, cost, and warnings, and it happens only when you approve it.
| Control | What it does |
|---|---|
| Proposal card | Describes the action, the target such as a Script Doc or Sequence, and for generations the model, references, and credit estimate. |
| Approve | Applies the change through the same paths you would use by hand. The card then shows the result or an error. |
| Dismiss | Discards the proposal. A dismissed card cannot be approved later from another tab. |
| Dependent proposals | Some plans arrive as a stack, for example create an episode then generate its Scene images. They apply in order and stop if an earlier step fails. |
Thruline Agent
The rail header has Chat and History tabs, with a save-status dot beside Chat. The director mascot shows whether Thruline is idle, working, waiting for your review, or paused. Your messages appear as bubbles; agent responses remain open text. Attachments wrap as compact pills above the message field; click a filename to preview it. During work, Send remains available with Stop beside it. Script previews scroll within chat and offer Read expanded. Generation reviews put Cancel on the left and Generate on the right.
Thruline can carry out related actions one at a time, using earlier results and your original request to choose the next action. Continue in the same open conversation to refer to earlier results, uploaded files, or project references. Apply and Generate reviews still control changes. A prepared draft or accepted generation is not proof that the requested work is finished. Keep the chat open to continue the same execution; after refreshing or starting a new chat, use Chat history to read saved conversations. Attachments remain subject to their availability and expiry; existing saved documents and generation jobs are independent of chat history.
Use the + attachment menu or drag files into chat to attach images, videos, audio, or UTF-8 text files. Canvas media and asset images can also be selected. Compact attachment tiles have removal controls and @ tags that insert at the chat cursor. Click a thumbnail to open its preview; video and audio previews include playback controls. The agent receives metadata, not their sound or video content. Text files provide their first 20,000 characters, with longer files marked as truncated. Image and video payload reviews, including image batches, start with the project aspect ratio. If the model does not support it, they use 16:9, or another supported choice when 16:9 is unavailable. You can change the ratio before generating.
Ask for a character sheet, a location reference, or a prop reference image. The agent uses suitable starting guidance while honoring your instructions for views, layout, style, background, and other details. Review and edit the prepared prompt before Generate. General image requests need no category. Editing an existing image follows your supplied images and requested changes. Creating a reference image does not automatically attach it to an asset or replace its primary image.
When available, ask Thruline to generate images for the current episode’s storyboard. Choose all Shots, selected Shots, or Shots missing images; the active variation is used. References start with attached Shot Media plus the primary/first image of each Asset tagged in the Shot description. Asset images receive bound prompt tags and remain editable in the payload review. Edit each prompt and add, remove, or reorder images from Canvas, uploads, or Assets before approval. Reference images use bound @ tags just like the image generation pill. The batch captures the Shot details when requested. Review edits affect this batch; later edits to Shot descriptions or media do not alter its payloads. Empty Shots offer Replace all, Fill missing details, or Cancel; missing Script context leads to Script development. Review the selected Shots, model, parameters, project aspect ratio, and recalculated total for the selected funding mode once. The total is visible below the model settings. If a credit estimate fails or credits are insufficient, retry the estimate without losing prompt edits. The approved allowance or credit choice stays with the batch. Results form a new Canvas grid and become their tagged Shots’ keyframes while keeping older images. Progress appears in a collapsed Shot list, with Queued for images awaiting generation and an animated Processing indicator. Open storyboard stays available through completion, then disappears on your next request. Completion automatically reports how many images were generated and attached. Keep the workspace open; accepted generations continue if you leave, and reopening an interrupted batch resumes remaining work and tagging. Pause stops new operations while already-started work settles; Resume continues from saved progress. Cancel also attempts to stop accepted generation jobs and reports any cancellation it cannot confirm. Completed images remain, and cancelled batches do not restart when reopened.
Uploads and reference selections appear immediately as reference cards with a loading indicator while processing continues. You can keep editing your prompt. Failed items stay visible: use Retry when available, or Remove to leave them out. Generation becomes available when required references are ready or removed. Canvas imports and copies show movable placeholders; generation previews appear while submission is prepared. If a result cannot be confirmed, check the destination before adding it again. Retry loading refreshes an existing result without generating or charging again.
When enabled, creative preparation tools can analyze your sources, draft a new Script, extract reusable assets or an ordered Scene outline, match attached images, propose visual guidance, and prepare asset images. Review and apply content before it is saved. Scene, Character, and Prop guidance can be edited separately. Generated and revised Scripts show an unsaved preview. Continue in the chat box to ask questions, request changes, or switch tasks; retained drafts are available without a resume button. Insertion review offers Apply script or Cancel, with the formatted Script draft preview below the replacement or new-episode selector. Request text changes in chat. Image batches use the Canvas model browser and shared model and advanced settings beneath it. Select the assets you want to generate and edit their individual prompts before Generate. The funding estimate updates with your settings. Existing images are preserved. When Studio Package is selected, the batch uses its included allowance; otherwise it uses ordinary credits. The approved funding choice stays with the batch. Reopen an interrupted preparation to check its retained jobs; closing a review does not undo saved content or accepted generation. New tools start in Draft; administrators can enable them for Testing before release.
Progress identifies when your Script is being written or revised. Each Scene uses one formatted heading; the screenplay starts directly with the first Scene, without a title page. Writing or revising returns an unsaved draft to chat. The Script draft preview remains visible. Ask questions, request changes, or switch tasks in the chat box, including during insertion review; there is no separate resume or discussion button. Ask to insert the draft when ready. The insertion review shows the same Script draft preview, with a new-episode or replacement selector above it; nothing is added until you click Apply. After saving and asset analysis finish, the review closes automatically. Image generation uses only your selected assets and the reviewed batch settings. The inline and expanded readers show the same complete retained Script. Drafting does not generate a second screenplay in chat prose. After insertion completes, the approved Script remains available in a collapsed, read-only receipt. Both Script readers use Fountain formatting. Earlier drafts collapse when you continue chatting; a revision preview appears only when ready. Choose View expanded for the larger reader.
Shots are listed as 01.1.2, without repeating Scene or Sequence text. Sequences use Scene 01, Seq 1. Numbering matches Storyboard; episode titles distinguish repeated labels. Agent replies display headings, bold and italic text, bullet and numbered lists, quotes, code, links, and tables. Wide code blocks and tables scroll within the chat. Your messages remain as typed.
Attach a Canvas image and ask which Shots and Sequences are connected to it. The agent looks up the Shot tags shown on Canvas and the Sequences containing those Shots, including tagged variations that are not currently selected. A Sequence containing a tagged Shot does not mean the image is directly tagged to that Sequence. If the lookup is unavailable, retry after the workspace finishes loading.
While processing, Thinking… appears between calls and Calling 1 tool… appears during tool execution. Expand tools beneath an assistant response to see the number of tool calls, including repeated calls. Admins can also inspect function names in order. Resize the chat by dragging its left edge.
When enabled, the agent can create Canvas text and frames. New objects appear in the area you are viewing unless you request a specific position. Use Show on Canvas beneath each created object to find it later. To reply to a Canvas comment, open the Canvas containing that thread first; comment searches can include other Canvases.
Identify planning targets by episode and Scene number or heading, Sequence name or label such as Scene 01 Seq 1, or a Shot label such as 01.1.1. Scene 1 and Scene 01 are equivalent. A Shot number without a variation suffix selects its approved variation, otherwise A; an explicit suffix selects that variation. The agent can create or update a Sequence in an explicitly named episode without switching your view. If several targets match, it asks you to choose. Numbering follows the current order.
Created Notes documents include Open Notes document. Created Shots include Open Shot: this opens Shot details in Storyboard when available, or selects the Shot in the Canvas Timeline and opens its detail rail. You can also ask the agent to add a named Description section to a Character, Scene, or Prop, or edit one existing section while preserving the others.
Ask the agent to comment on a Shot variation, Production take, Script document, Notes document, or Canvas. For documents, identify the exact passage to comment on. To reply, identify the comment and its area; ambiguous threads need clarification. Canvas comments still require the owning Canvas to be open.
Ask for assets in an episode to see its library assets, including manually added assets. Ask for assets in the script for a script-derived breakdown. Document lists group each episode’s Script and Notes, with project Notes listed separately.
Ask for a particular episode, asset type, comment surface, or status to narrow an answer. Episode lists show titles in saved order. Counts cover the matching collection without listing each record. For analysis, the agent reads relevant details in pages and explains any remaining gaps. You can ask a follow-up to explore a specific part of the answer.
Ask Thruline to preview a voice to get a playable audio sample in chat. Use play, pause, and seek to listen; a failed sample can be retried. You can also ask for models within a family by its name or ID. Large model lists and capability details are read in pages, so ask a follow-up for more detail.
Import script and the agent’s importDocument workflow share the same review. Attach a document or paste text in chat. Fountain scripts retain their formatting; other recognizable screenplays are formatted before review. For ordinary text, confirm Notes; for uncertain content, choose Script or Notes. Select Replace or New (there is no Append option), choose the destination, then click Apply import. A new Script creates an episode; new Notes can belong to the project or an episode. Script imports add high-confidence assets and keep uncertain suggestions for review. Notes do not create assets. PDFs contribute readable text only; images are not transcribed. Open document takes you to the imported text. Resume import reopens unfinished work from Documents. Retry resumes failed preparation, and cancelling asset analysis retains text already imported.
| Control | What it does |
|---|---|
| Pick on Canvas | Click an image, video, or audio node to attach it. Picking does not show a loading spinner. Click the chat text box, press Escape, or choose Return to chat to cancel and keep your draft. Loading starts only after selection. Invalid selections explain what to choose; use X to dismiss the notification. |
| Choose from assets | Choose an asset image using the same browser as the generation composer. Typing @ also opens this browser and inserts a bound image tag. The browser expands above the attachment tiles, with your prompt below them. |
| Attach documents | Drop files into chat or use Upload files / Attach documents. Text files stay as attachments rather than being pasted into the prompt. PDF and DOCX documents support up to 8 MB each. Images added or imported to Canvas include Show on Canvas. Upload JPEG, PNG or WebP images up to 20 MB each; MP4/MOV video or MP3/WAV audio up to 50 MB each; and UTF-8 TXT, Markdown, Fountain, CSV or JSON text up to 1 MB each. Each batch is limited to 100 MB. Chat and source analysis accept up to 32 files at a time; each generation follows its selected model’s reference limits. Source analysis accepts up to 100 MiB of image/document files, uses smaller image copies and reports shortened text. Audio/video analysis uses metadata only, without transcription or video-content analysis. A conversation supports 64 attachments. Private uploads expire after 24 hours; retained project outputs remain available. |
Video and dialogue requests open an editable preview before generation starts. The preview uses the model you name, or selects a compatible default when you have not named one. Your requested video mode and settings are kept when choosing a model; unsupported choices stay in review for correction. Choose the video model and mode below the prompt, edit the prompt and settings, and use Add reference beside the media previews to choose Canvas media, uploads, or asset images using the same source menu as Canvas generation. Use @ tags to refer to attached media; tags keep their media identity when references are reordered. First/last-frame labels show the order. For dialogue, edit turns and check each Character’s assigned voice. Click Generate to submit, or Cancel to stop. You can request generation from any workspace surface. After video submission, Show on Canvas opens its generating location or finished output. The agent can read your current project, episode, surface and detail view from workspace state. For a video of a planned Shot or Sequence, ask Thruline to prepare its video. It gathers the selected Shot descriptions, camera direction, tagged asset context, each tagged asset’s first image, and attached Shot Media, then enriches the prompt inside this same review. A Sequence keeps its Shot order and shares repeated references. Wait for preparation to finish, edit the draft, then click Generate. Missing media or incompatible timing must be resolved first. If you only say “prepare this Shot,” Thruline asks whether you want an image or a video when the conversation has not established that choice.
Ask who created attached media to check recorded human attribution, or ask which model or tool generated it to see the recorded public model name and generation operation. You can also ask for the submitted prompt, recorded generation type and settings, image output size, reference media, dialogue turns, or generation status. Details are returned when recorded; missing values are not guessed. Ask for both to receive both. Missing creator information is never replaced with a model name. Internal infrastructure stays private. Audio/video inspection reports metadata only; it does not watch, listen, or transcribe.
Chat history
Chat history keeps your saved Thruline conversations private to you within each project. The first message becomes the title; chat names cannot be renamed. Saved chats remain available after a refresh and on another device when you sign in to the same account and project.
Open Thruline Agent and choose the History tab in its header. The Chat tab keeps your current conversation and draft open.
Search by chat title or browse Today, Yesterday, Previous 7 days and Earlier. Select a row to read it; use Load more for older chats and All chats to return to the list.
Choose Chat to return to your draft and ongoing work. New chat saves the current conversation before opening a fresh one. To remove an older conversation, open its Chat options menu, choose Delete chat and confirm. Generated media stays in the project; the current live chat has no delete option.
Agent Availability
Thruline Agent is always enabled when available to your account. Non-admin users use GPT-6 Luna automatically. Only platform admins see the model selector; there is no agent on/off switch.
Open Thruline Agent from the project workspace header. Close the chat rail to return to the workspace; closing it preserves the conversation. The retired Workspace Story Agent and Onboarding Agent no longer have workspace launchers.