Ember with AI agents

Ember exposes its own operations as tools an AI assistant can call. The assistant is not clicking around a screen or exporting files for another program to parse: it works from the project and timeline its session is bound to, not whatever the window shows, and writes changes into the same undo history you use.

A first request to type. "Caption this timeline, use the Client Name style, add zooms and transitions, then export to this folder."

Type it in a new conversation and replace the style name and the folder with your own. The assistant reads the project and timeline its session is bound to, so you do not have to describe them.

Included with Pro and Max. AI agent access and Ember Engine are part of the Pro and Max plans. Lite includes the caption, styling and folder export workflow without the agent tools.

Orientation

Check what is possible right now, what is blocking it, which Ember build is running and which engine is installed.

Projects and media

List or create projects, import local footage, inspect the media pool, switch timelines and reopen a timeline from its file.

Review

Render stills at a time or a spoken word, make contact sheets over a span, inspect source footage, and render a short preview with audio.

Captions

Run a Generate pass, read captions by view or search, and edit text, timing, lanes, Styles, Sparks, Links, Formations and placement in one undoable batch.

Timeline

Read the timeline, and trim, split, extend, replace, enable, disable, ramp, retime, reframe or ripple-edit pieces. Grade a clip, match a camera, cut multicam, and place or edit graphics and their fields.

Audio

Report the mix's sounds and levels, preview and apply Trim Pauses, and design, render and place sound effects.

Colour and analysis

Read scopes, balance and match cameras, apply LUTs, and solve a camera for 3D pins.

History

Undo or redo, read the journal of agent calls, and checkpoint, diff or restore a timeline.

App and settings

Open a page or go back, read or change the Generate, Ignite and Export options like the user's own controls, and install Ember Engine when the plan includes it.

Delivery

Run Ignite, adjust or discard its live result, and render clips from a Resolve timeline or an Ember project.

The tool list follows what Ember is working on. With a Resolve session it offers the Resolve workflow; with one of Ember's own projects open it offers the local editing tools; with neither it offers everything so a project can be opened.

1. Open Settings > Account > AI assistants

   Ember lists the assistants it can configure. Connect one and Ember registers its MCP server, starts the bridge once, and verifies that the assistant can initialize and see Ember's tools before reporting success.

2. Or register from a terminal

   For an assistant Ember does not list, use Ember's mcp command with the assistant's name, for example Ember mcp --register <assistant>. The same command supports --status and --unregister.

3. Start a conversation

   Ask the assistant to use Ember. It reads Ember's own guide resource (ember://guide) for the workflow and each tool's schema for its fields; you never have to paste a prompt or a schema by hand.

A row that reads Not installed has nothing to connect: install that assistant first. Its Connect button does nothing until the assistant is on the computer.

Ember's check starts the bridge once and closes it. The assistant reads its own configuration when it starts, so restart it if it was already open when you connected. Codex and OpenCode are registered automatically once when they are installed. A connection you disconnect on purpose stays disconnected.

1. Status first

   The assistant checks Ember's state instead of trying an action to see whether it fails. Blockers name the fix: no Resolve connection, no project, no engine, a stale build.

2. Read before writing

   It reads only the facts it is missing: the caption blocks it will change, the timeline pieces it will move, the fields a graphic template accepts.

3. One batch per intent

   Edits arrive as one batch with every operation in it, so a group of changes is one revision and one undo step rather than a trail of partial states.

4. Generate once

   Generate and Ignite use your credits. A good assistant runs a pass once with the options you approved, then edits and restyles instead of regenerating to compare results.

5. Look at the frames

   Before delivery it renders stills or a short preview through the same path the export uses, so composition problems are visible before a full render.

6. Report the result

   The answer names what changed, any warnings, and where the exported files are. It should not claim a render or a check that did not happen.

An assistant works in whichever mode Ember is in. With DaVinci Resolve connected, it reads the active timeline and writes captions, polish and export there. With an Ember project open, it imports footage and edits the local timeline, captions, graphics and sounds, then reviews stills and previews.

A tool that does not appear in the current mode is not wanted there. That is why an assistant should read the tool list instead of calling something that does not apply.

• Paid work is explicit. Generate uses the same credits as the button, and paid work inside a run, such as Cast, does too. Rendering to a folder does not.
• Timeline and caption edits in an Ember project are revision-checked. A stale edit is refused instead of overwriting work that changed since it was read.
• Undo and history include agent edits. You can inspect and undo what an assistant did like any other change.
• Agent edits never bypass a refused operation. If Ember cannot verify a step, the assistant reports the refusal rather than substituting a different result.
• An assistant works from the project it opened and the files you ask it to import or inspect; it does not scan your disk.

• Captions: generate, verify and edit

  The words the assistant generates and rewrites, and the editor it edits them in.

• Ignite: captions and polish in one pass

  The one pass that places the approved captions and the polish around them.

• Export: renders and social delivery

  Rendering the clips to a folder, and the files a finished render leaves.

Every page here is written to be read by a machine as well as a person. /llms.txt lists the topics with one-line summaries so an assistant can find the right page before it starts a task.

• Start at /llms.txt for the index, then read the topic that matches the job.
• The agent page describes capabilities and setup, not implementation details.
• Inside Ember, the assistant reads ember://guide and each tool's schema, which are more current than any website copy.