Concepts

This page explains the words ReelBolt uses and how the pieces fit together. Each section stands on its own, so you can jump straight to the one you need.

Workspaces in ReelBolt

A workspace is the container that owns everything you do in ReelBolt: your projects, their files, workflows, renders, custom agents and AI connections. It is also the unit that shares and bills: people inside one workspace work on the same projects, and people outside it cannot see them at all.

Every account gets a personal workspace of its own, always shown as Personal, holding just its owner. When you need to work with other people you create a team workspace and invite them into it, and you can belong to as many as you like and switch between them from the menu at the top left. Inside a workspace each member holds one of four roles — Owner, Admin, Member or Billing — and the role decides what they can do there.

Two things follow from this and are easy to get wrong:

• A project you can see is a project your workspace can see, not one only you can see. The exception is the personal workspace, which has just you in it.
• Platform administration is not workspace membership. The menu group called Platform belongs to whoever runs the whole installation, and it deliberately gives no view of any workspace's projects.

See Workspaces and teams for the roles in full, invitations, switching and leaving.

Projects in ReelBolt

A project is a container for one video effort: a product launch video, a tutorial edit, a feature teaser. It holds three things:

• the files you upload and the files produced while working,
• the workflows that describe how to turn those files into a video,
• the renders, which are the finished videos.

A project belongs to the workspace it was created in. Everyone in that workspace whose role allows project work — Owner, Admin and Member — can see it and run its workflows; someone outside the workspace cannot see it at all, and neither can a platform administrator, who manages accounts and AI services rather than other people's projects. Deleting a project permanently deletes everything in it. See Workspaces and teams.

Project files and their three categories

Every file in a ReelBolt project has a category that says where it came from. You will see these names when talking to the assistant or reading a file's details:

Category	Shown in the Files tab as	What it holds
userFiles	Sources ("Uploaded by you")	Everything you uploaded: code, footage, music, images, documents.
agentFiles	Generated ("Created by an agent")	Files agents wrote during runs, such as analysis results, generated clips and synthesized voiceover audio.
outputFiles	Renders ("Render")	Finished videos produced by runs.

Files can sit in folders. Folders are just a way of organising the list: they appear when a file is uploaded or moved into them, and an empty folder exists only until you put something in it. You can move files between folders, and even between categories, with Move to....

Text files get an automatic summary (a few lines about what the file contains) and are added to the project's search index so they can be searched by meaning. Video and audio files are skipped for both, because they are not text. The summary helps agents and the assistant understand your material without reading every file in full.

Workflows in ReelBolt

A workflow is a recipe: an ordered list of steps that turns your project's files into a video. Workflows belong to one project.

You can get a workflow in three ways:

• Automatically: every new project gets Promo video from your code v1 (called Quick Win Promo v1 in projects created before October 2026), a ready-made "promo from your code" workflow.
• From a template: on the project's Workflows tab, click New workflow, pick one of the cards under Start from a template and click Use this template. See Workflow templates.
• By building or describing one: add steps yourself in the visual builder, or ask the assistant to draft one and approve its proposal.

You can rename a workflow, change its steps, switch Require a brief before each run on or off, and click Save Changes. Running a workflow is done with Run.

Steps and step types in a ReelBolt workflow

A step is one job in a workflow. Steps run in order, and each step can use what earlier steps produced. Every step has a step type, which decides what kind of job it is:

• Agent: an AI agent does the work (writes a script, decides which shots to keep, builds an animation).
• ReviewLoop: an agent scores the result and, if it falls below a minimum score, sends the workflow back to an earlier step to try again, up to a set number of times.
• Conditional, ForEach and Parallel: control the flow, by choosing a branch, repeating over a list, or running several agents at once.
• Extract: trims a large earlier result down to just the parts later steps need, which makes runs cheaper and faster.
• VideoAnalyze: examines footage (pauses, scene changes, speech, what is on screen) without any AI decision-making of its own.
• VideoCompile: cuts and assembles the final video from the decisions made by earlier steps, adding graphics, music, sound effects, colour and inserts if asked.
• EditRoom, GraphicsRoom and ColorGradeRoom: several agents discuss a decision together, like a small editing team, and a director agent settles it.
• VideoGenerate: creates new video clips with a paid video-generation service.
• Voiceover: turns script text into narration audio.

Each step type is described in detail in Step types.

Agents: built-in and custom

An agent is an AI worker with a specific role and set of instructions. A workflow's Agent steps each name the agent that should do the work.

• Built-in agents ship with ReelBolt: code analysers, a scene translator that turns your interface into animated scenes, an animation strategist, a scriptwriter, a director, an author that builds and renders the video, reviewers, and video-editing specialists such as the story editor, motion-graphics planner, music supervisor, sound designer, colorist and shot director. Their instructions cannot be edited, because the rest of the system relies on what they produce.
• Custom agents are ones you create on the Agents page with New Agent: you give it a Name, a Description, a System Prompt (its instructions in plain language) and a Color. Custom agents can be edited at any time.

Each agent uses an AI service that the workspace has available: the connections the ReelBolt installation was set up with, plus — in a team workspace whose plan allows it — the workspace's own connections, which take precedence there. A platform administrator can point every agent everywhere at a different AI service; a workspace's Owner or Admin can do the same for that workspace alone — for a built-in agent type, and for the workspace's custom agents (for example a stronger model for the director). See Inference providers. The full list is in Agents.

Executions, runs and step results

An execution (or run) is one go of a workflow from start to finish. Each execution has a status:

Status	Meaning
Queued	Waiting for a free slot; only a limited number of runs work at the same time.
Running	Working through its steps.
Passed	Finished successfully.
Failed	A step could not complete; the error is shown on the execution page.
Cancelled	Stopped, either by you pressing Stop or because the system was shut down mid-run.

Every step that runs leaves a step result: what went in, what came out, its own status (Pending, Running, Completed, Failed or Skipped), any error, how many tokens it used, and which attempt it was (a review loop can run the same step several times). You can read all of this by clicking a step on the execution page. A step that fails is usually retried automatically before the whole run is marked Failed.

Outputs and renders

An output is anything a run produces. The most important outputs are renders: playable video files. Renders appear on the project's Renders tab, under Renders on the Files tab, and on the Overview tab under Latest renders. You can play, Download and Share them.

Some steps also produce non-video results that you can inspect on the execution page, such as a full footage analysis, the list of cuts, or a manifest of generated clips. Video-editing runs also produce a timeline you can browse in the Editor section.

The step cache: when a rerun reuses earlier results

When you run a workflow again, ReelBolt can reuse a step's earlier result instead of doing the work again, if nothing that step depends on has changed. That includes the step's own settings, the agent and its instructions, the AI service used, everything earlier steps produced for it, your user request, and the project's files. Change any of those and the step runs fresh. Reused results are kept for up to 7 days by default.

A reused step is marked reused from cache on the execution page, with how many tokens it saved. This makes reruns faster and cheaper, and for paid video generation it means an identical clip is not paid for twice.

Some steps are never reused by default, because they change things outside their own result or because a fresh judgement is the whole point: steps whose agent writes project files or builds and renders video, review loops, the graphics room, and the creative steps in the built-in templates. Analysis and cutting steps are reused by default.

Each step has a Cache Mode setting (the JSON key is cacheMode):

Cache Mode	Meaning
Default	Use the built-in rules above.
Always	Always allow reuse for this step, even where the built-in rules would not. Use with care.
Never	Always run this step fresh. Use it when you want a new creative take every time.

Your brief for a run (Require a brief before each run)

Every run can carry a short brief: when you click Run, the confirmation box always has a Your brief field for free text such as "keep it under 45 seconds and end on the pricing page". Leaving it empty is fine unless the workflow has Require a brief before each run switched on (on the workflow page; the setting is called requiresUserInput), in which case the run cannot start without one.

The request is passed to every agent in the run as extra context, and it is remembered with the execution, so Retry uses the same request. Because the request is part of what the step cache compares, changing it makes affected steps run fresh.

Costs and spending limits in ReelBolt

Most of ReelBolt's work uses AI services connected to your workspace (its own, or the installation's). Those services usually charge by usage; ReelBolt shows how many tokens each step used on the execution page, but the bill itself comes from the service provider, not from ReelBolt.

Paid video generation is the one place ReelBolt actively guards your spending. A VideoGenerate step must have a Max Spend cap (maxSpendUsd, in US dollars). Before buying anything, ReelBolt checks four limits and refuses the whole step, buying nothing, if any would be exceeded:

• the step's own Max Spend cap, compared with the estimated cost of every clip it would create,
• a per-project daily budget (20 US dollars per day by default),
• the daily video budget of your workspace's plan (a plan without video generation has none at all, so the step is refused),
• a daily budget across the whole installation (50 US dollars per day by default).

Days are counted in UTC. The project and installation daily budgets are set by whoever runs the installation; your workspace's daily video budget comes from its plan.

When you run or retry a workflow that generates video, ReelBolt asks whether to reuse previous clips (free: an identical earlier clip is used again) or Regenerate (costs money). The upper bound shown is the sum of the video steps' Max Spend caps; the real cost is usually lower. See Troubleshooting for what a BUDGET_EXCEEDED error means.