Agency automation setup guide

How it works

Claude Desktop starts a separate local companion included with VideoBatcher. ChatGPT, Gemini and Perplexity send requests through your private VideoBatcher link; the open desktop app answers them using the same tools and permissions. No customer-hosted server, terminal or extra AI API key is needed for these four setup options. You bring your own AI account, which makes its own model calls. If you separately allow AI media sharing, it can also request sampled images from a video.

• •After you confirm the guided setup, your AI can use files anywhere you point it to — on your computer, an external drive or a synced folder. Hidden, private and system folders stay off-limits.
• •VideoBatcher checks the job against your preset rules, limits and file-safety rules before anything renders.
• •In Review first mode, you approve or reject it in the desktop app after checking captions, save folders, and local previews. In Automatic mode it starts rendering directly.
• •Preview tools return local file paths only — never image attachments or automatic uploads to your AI client.

Download videos without connecting an AI

Downloads requires verified Agency access, but no AI connection or AI permissions. Only download content you own or have permission to use.

1. Open the dedicated Downloads tab in the desktop sidebar. Paste public Instagram post/Reel links or TikTok video links, one per line, then click Find videos.
2. Choose from the results using thumbnails or checkboxes. You can multi-select, use Select all, or Clear selection. A single result is selected for you. If a preview is unavailable, the app shows that rather than inventing an image.
3. Click Choose download folder… to save in your chosen folder, or keep the app's default download folder. Click Download selected; only checked videos are downloaded. Changing the folder affects new batches only.
4. Select completed files and click Use in editor to repurpose them with your usual effects, captions and presets. This adds them to your media list and opens Effects; nothing renders until you click Generate. You can also follow progress, show a completed file in its folder, or cancel remaining downloads.

To use downloaded files in AI batches later, just tell your AI where they are. Downloads your AI starts go to Downloads › VideoBatcher Imports unless you ask for another folder.

Desktop setup

The steps below are identical on macOS and Windows — automation lives inside the same VideoBatcher desktop app you already use.

1. 1.

   Activate your Agency access. Automation needs verified Agency access, including an eligible Agency trial or invited beta access. Standard alone and the free desktop trial do not unlock it. Activate your license in VideoBatcher, not in your AI chat.
2. 2.

   Open AI Automation → AI Connection. Use the desktop app to approve access and review jobs. Keep VideoBatcher open while using a private-link connection. Claude Desktop starts a separate local companion, which must stay running for the jobs it owns. There is no always-on background rendering service.
3. 3.

   Set up AI access once. Click “Set up AI access…” and confirm. Your AI can use files anywhere you point it to and save finished files where you ask—including supported external and synced folders. There are no folder lists to maintain. When you don't say where, files go to VideoBatcher Exports; you can change this default folder in Settings. Originals and hidden, private and system folders stay protected.
4. 4.

   Choose how batches run. The setup confirmation enables Automatic mode and creates Agency Starter. Valid jobs can then start without per-job approval. To check every job yourself, choose “Review before running” in Settings and save before asking your AI to submit work. You can also change the default folder, presets and limits there. Agency Starter allows small brightness, contrast and saturation adjustments; your AI can add captions per job.
5. 5.

   Choose your AI and follow its steps. Under “02 / Connect your AI,” choose Claude, ChatGPT, Gemini or Perplexity. Install the Claude Desktop extension or add your private VideoBatcher link in the other app's connector settings. Then ask your AI to call videobatcher_resources. Creating an extension or link does not mean the AI is connected yet.
6. 6.

   Ask your AI to make a small batch. Use the example request below to find a clip, add captions and make two versions. In Review first, check the job before approving; in Automatic, watch progress directly. Your AI cannot approve a job, change modes or grant itself new permissions.
7. 7.

   Track, inspect, or cancel. Closing the desktop app cancels the jobs it owns rather than leaving anything rendering unattended.

Note: saving new permissions or configuration cancels any jobs already running and invalidates pending ones — they'll need to be resubmitted.

Connect your AI

In VideoBatcher, open AI Automation → AI Connection → 02 / Connect your AI. Complete Quick setup or review Settings first, then choose your AI. All four options use your saved settings, default folder and job limits. You need your own eligible AI account; provider availability and account rules can change.

Available tools

There is no tool to delete originals, run shell commands, approve a job, change modes, or grant permissions. Browse and export access stays inside the scope you confirm on the desktop.

Automatic vs. Review first

Previews are capped at the first 24 outputs per job. In Review first, larger jobs will flag that the remaining outputs weren't visually previewed — keep jobs smaller if you want every output checked before approval.

Advanced: job format (multi-folder example)

You do not need to type JSON or folder paths for the guided setup: ask your AI to call resources and browse_media, then use the real paths it returns. For advanced manual setups, a job describes sources, where finished files go, and caption settings. The example below is illustrative; your AI must replace every path and preset with values returned by your installed app.

Say “Save 10 versions in D:/Clients/Summer.” The AI can create the destination without a settings step. Each job writes a new dedicated subfolder, leaving originals and earlier outputs untouched.

For “Upload the finished versions to my connected Drive folder,” your AI needs its own authenticated connector that can upload local files. VideoBatcher supplies a delivery manifest marked local_ready_not_uploaded; it does not include built-in Drive/Dropbox OAuth or upload support. The AI must obtain the provider's confirmation before saying “uploaded.” Saving in a synced folder also does not prove cloud synchronization.

Presets & supported features

Guided setup provides Agency Starter so you can try a simple batch without creating a preset. It enables bounded brightness, contrast and saturation adjustments (−3 to +3); ask your AI to add captions in the job. For advanced results, save and approve your own preset in the desktop app. Jobs carry its supported video/image settings:

• •Color & look: brightness, contrast, saturation, hue, color temperature, gamma, sharpen, noise.
• •Frame & motion: zoom, pixel shift, rotation, mirror, trim, speed, and pitch (video).
• •Audio EQ and a maximum output resolution cap (Original, 4K, 1440p, or 1080p) to keep renders fast on modest hardware.

VideoBatcher 3.10.0 also supports these features from an approved saved preset. Configure them manually in VideoBatcher first:

• •Audio pools: saved tracks and audio settings for video jobs.
• •Watermarks: saved logo and text watermark settings.
• •Device Metadata Profiles: the supported device profile selected in your saved preset. No GPS or capture-location fields are accepted.

1. Configure the features in the desktop app and save a preset.
2. In Automation, approve that preset. Its audio, logo and font files are used wherever you saved them.
3. Ask your AI client to use that preset by name. Start with a small job in Review first mode.

Turning a feature on in the manual workspace alone does not apply it to MCP jobs: automation starts from the saved preset, not unsaved controls. Your AI can adjust supported audio, watermark, and device-profile settings for a job using the preset's assets. Enabled features stay on; the AI cannot switch them off or enable disabled features. Each job records the resolved settings for review and checks its referenced files again before execution.

Per-job adjustments (feature_overrides)

A job can nudge the approved preset's audio, watermark, and device-profile settings for that job only — without changing what's saved on the preset itself:

Overrides can only stay inside what the preset already enabled — they can't turn a disabled feature on, turn an enabled one off, or swap in a new logo/audio file.

Let your AI use a caption pool

Ask your AI to write several captions and rotate them sequentially or randomly across outputs, or explicitly use the selected captions saved in your approved preset. It can also assign captions to particular outputs. This does not edit your desktop caption library. Omitting captions means no caption is added.

Caption timing

By default a caption stays visible for the whole final video. For video jobs, captions.timing can instead show it for just a range or the final few seconds of the finished output (after any trim/speed changes) — useful for a call-to-action that should only appear near the end:

{"mode": "range", "start": 2, "duration": 3} shows the caption for a specific window instead. Timing is video-only — a job cannot request non-whole timing for a still image, and images ignore this field.

Update older builds: VideoBatcher 3.10.0 supports approved saved presets with audio pools, watermarks and Device Metadata Profiles through automation. Older builds may reject these features. Install the latest version before following this guide.

Audio pools cannot be applied to still images; use a separate preset without audio for image jobs. Missing assets and unsupported settings produce an error, rather than silently dropping a feature. Your normal manual desktop workflow is unchanged.

Still images with Device Metadata Profiles enabled export as JPEG to retain device EXIF metadata; images with profiles off remain PNG. Previews are still images, so they do not play audio. Video previews apply the saved effects before extracting a frame and may take longer to generate.

A few other things automation doesn't do yet: video/audio-only effects (zoom, trim, speed, pitch) on image inputs, or characters outside the current default/emoji fonts — both return an explicit error instead of a silent workaround.

Safety & things to know

• •Caption text is always treated as data, never as commands. Raw FFmpeg arguments, unknown fields, and capture-location/GPS metadata are rejected.
• •Auto-fit checks that text is readable — it doesn't understand faces, products, motion, or platform overlays. Review placement yourself.
• •Video previews show representative frames, not every frame.
• •Permissions and job contents are re-checked at approval and at render time — a changed input or setting needs a fresh job.
• •If you retry a submission, reuse the same idempotency key; a changed request with that key is rejected rather than double-processed.
• •Cancelling preserves your originals and anything already completed — check results before resubmitting failed work.
• •This connector restricts its own operations — it isn't an OS-level sandbox for an AI client that has been separately granted shell or unrestricted filesystem access.
• •Agency checks your license online before starting new jobs. There's no offline entitlement cache — a render already running can finish through a brief outage, but new jobs need a successful check first.
• •Export destinations need to support hard links — NTFS, APFS, and ext4 work; FAT/exFAT and some network drives don't. Render to a supported local drive first.

Getting help

Update to VideoBatcher 3.10.0 first. If tools do not appear, follow the connection checks above. If something still fails, contact us with your app version, operating system, AI app and the exact error. Never send your license key, private connection link or account credentials.