Agent Mode

Agent Mode is an AI-powered editing assistant built into modelBridge. Instead of manually picking models, setting parameters, and importing results, you describe what you want in plain language — and the agent handles the rest.

It reads your timeline, understands your project, and takes action directly in Premiere Pro. Think of it as a knowledgeable assistant sitting next to you who can operate the plugin, run quality checks, and help you make creative decisions — all through a chat interface.

What it can do

Timeline editing

The agent can read and modify your Premiere Pro timeline:

Move, trim, split, and delete clips — describe the edit, and the agent executes it
Adjust properties — scale, position, rotation, opacity, audio levels, speed
Apply effects — add or remove video/audio effects and transitions
Color correction — adjust Lumetri Color parameters (exposure, contrast, highlights, shadows, temperature, tint, saturation, vibrance, sharpening)
Manage markers — add, remove, and read sequence markers
Control tracks — mute/unmute audio and video tracks
Set playhead, in/out points, and clip labels

Quality checks

Ask “run a QC check” and the agent inspects your sequence for common issues:

FPS mismatches between clips and sequence
Resolution mismatches
Gaps on V1 (video track 1)
Audio issues (silent tracks, level inconsistencies)
Codec inconsistencies
Flash frames (very short clips)
Disabled clips still on the timeline
Pacing analysis

The agent prioritizes findings by severity and explains each issue with specific clip names and timecodes.

AI model operations

The agent has full access to modelBridge’s model catalog:

Search and browse 1,100+ fal.ai models with filtering
Install models directly from the catalog
Get model details — parameters, constraints, pricing, generation count
Recommend models based on what you’re trying to achieve

Multi-step workflows

Because the agent can chain actions, it handles complex tasks that would normally take many manual steps:

“Reframe this sequence for Instagram (9:16)” — adjusts scale and position on every clip
“Normalize all audio to the same level” — reads levels, calculates adjustments, applies them
“Add a cross dissolve to every cut” — finds all edit points and adds transitions
“Close all the gaps on V1” — detects gaps and ripple-deletes them
“Copy the color grade from clip A to clips B, C, and D” — reads effects and replicates

Setting it up

1. Add your Anthropic API key

Agent Mode uses Claude (by Anthropic) as its AI engine. You provide your own API key — modelBridge never charges for agent usage beyond what Anthropic bills directly.

Create an account at console.anthropic.com
Generate an API key
In modelBridge, go to Settings and paste your key in the Anthropic API key field

2. Open the Agent tab

Once a valid API key is saved, the Agent tab appears in the plugin. Click it to open the chat interface.

3. Start a conversation

Type a message describing what you want. The agent reads your current sequence, selected clips, and project state before responding. It has full context of your timeline from the first message.

Custom instructions

Click the gear icon in the Agent tab to open the customization overlay. Here you can set personal instructions that shape how the agent behaves:

Preferred models: “Always suggest Kling v3 for video generation”
Working style: “Keep responses short and direct”
Safety preferences: “Always ask before deleting clips”
Creative direction: “I work on corporate videos — suggest professional, clean styles”

Your instructions persist across sessions and apply to every conversation.

What it costs

Agent Mode uses your own Anthropic API key. You pay Anthropic directly at their published rates — modelBridge adds no markup.

Typical usage for a 2-hour editing session costs roughly 50–60 SEK per month (approximately $5–6 USD). Costs depend on conversation length and complexity — quick questions are cheap, multi-step editing sessions with many tool calls cost more.

You can monitor your spending on the Anthropic dashboard.

Model selection

The Agent tab includes a model dropdown where you can choose which Claude model powers the assistant:

Claude Sonnet (default) — fast and capable, best all-rounder for most editing tasks
Claude Haiku — fastest responses, good for quick questions and simple operations

Switch freely between models depending on the task complexity.

How it understands your project

When you send a message, the agent can call tools to read your project state:

Sequence info — FPS, resolution, duration, track count, work area, in/out points
All clips — name, dimensions, source FPS, media type, position, duration (paginated for large sequences)
Selected clips — only what’s currently selected
Audio tracks — track count, clip names and positions
Effects — the full effect stack on any clip, including Motion properties
Project bin — all items in the project panel
Track state — which tracks are muted or locked

The agent reads what it needs for each request — it doesn’t dump your entire project into every message.

What it cannot do

Being transparent about limitations:

Cannot add or delete tracks — track structure must be managed manually
Cannot insert clips from the project bin — it can read the bin but not place clips on the timeline
Cannot set keyframe animation — properties are set to static values only
Cannot do time remapping — speed changes are uniform, not ramped
Cannot access external URLs — it works within your local project and the fal.ai catalog
Cannot modify Premiere Pro preferences — project settings outside the panel are read-only
Cannot see audio waveforms or loudness meters — it reads clip metadata, not visual scopes
Undo may be unreliable — Premiere Pro’s Cmd+Z doesn’t always reverse scripted edits. The agent will warn you before risky operations and suggest saving first

Privacy

Conversations are sent to Anthropic’s API using your personal key. modelBridge does not store, log, or transmit your conversations to any other service. Your prompts, creative direction, and project details stay between you and Anthropic.

See Anthropic’s usage policy for how they handle API data.