Kolbo.AI

Queue a read-only inspection of the active Premiere Pro sequence or After Effects composition. No approval is needed. Premiere returns size, duration, playhead_seconds, tracks and a clip list (track, name, start_seconds, end_seconds) sorted earliest first; After Effects returns the comp and its layers. The list is truncated to max_clips (default 200) with the full count in clips_total / layers_total.

Import an SRT caption file hosted on Kolbo storage onto the active Premiere Pro sequence. Premiere Pro only. The editor must approve it in the Kolbo panel. Create the SRT with transcribe_audio first when needed.

Import one Kolbo media item into the open Premiere Pro or After Effects project. It lands in the Kolbo.AI / My Media bin for its type (Videos, Images, Audio). The editor must approve it in the Kolbo panel. Returns a command record; poll adobe_get_command_status.

List the caller's Premiere Pro and After Effects windows that have the Kolbo panel connected to AI agents. Call this before the first Adobe command. If none are listed, ask the user to open the Kolbo panel and click the AI agents button in its header.

Import one Kolbo media item and place it on the Premiere Pro work sequence (free track at the playhead) or the active After Effects composition. There is no time or track parameter. The editor must approve it in the Kolbo panel.

Run ExtendScript inside the connected After Effects (or Premiere Pro) for real motion graphics: shape layers, trim paths, text animators, effects, masks, expressions, cameras, precomps. `code` is a FUNCTION BODY: use log(...) for progress and `return` a JSON-serialisable result. In After Effects the whole script is one undo step. The editor sees the exact code in the Kolbo panel and must approve it. Scripts have full access to the project and the computer, so never read or write files, call system.callSystem, or touch the network unless the user explicitly asked. Read workflows/after-effects-motion.md before writing motion graphics, then verify with adobe_capture_frame.

Extract a color palette from 1-5 reference image URLs using fast local pixel analysis (no LLM, free). Does NOT save anything — pass the returned colors to create_color_palette to save + optionally activate it.

Turn a script or scene description into a SYNCI music search.

AI helper that turns a video/voiceover script into stock b-roll search terms. Returns { queries[], mediaType, keywords } you can feed into search_stock_media to auto-source matching footage/photos. Use this first when the user gives you a script rather than explicit search keywords.

Understand a video with Kolbo's official video-understanding model (agentic Gemini: it navigates the timeline itself — frames, audio and transcript on demand — instead of sampling fixed frames, so long videos are cheap and timestamp / "when does X happen" / counting questions are answered directly). Pass a public https video URL or a YouTube URL plus an optional question in `prompt`; with no prompt you get an exhaustive description + verbatim transcript. Billed by the analyzer's real token usage. Local file → upload it first (create_upload_ticket / upload_media / media_upload_widget) and pass the returned URL. For subtitle files use transcribe_audio instead. LOCAL FILE (on the user's machine, or attached to this chat)? NEVER reply that you cannot upload files — Kolbo uploads it for you. Call `media_upload_widget` so the user picks the file (claude.ai web/mobile), or `create_upload_ticket` and POST the file yourself if you can run shell/HTTP commands; either way you get an https:// URL to pass here. Ignore any "absolute local path" wording in the args below — this server cannot read the caller's disk, so a path will fail.

Run paid video, music, speech or sound understanding on a previously extracted interval only. Video focus requires clip evidence; audio focus accepts audio or clip. Useful for motion/continuity, instruments/rhythm/mood, speech context and sound design. For word timestamps/SRT use transcribe_video_evidence. Returned analysis timestamps are clip-relative: add source_offset_seconds. Real model usage is billed, and identical successful requests reuse saved analysis without new charges. An uncertain previous paid request is never automatically resubmitted.

Archive a project — hides it from the default project list without deleting anything. Fully reversible with `unarchive_project`. (Permanent project DELETION is intentionally not available via the API — it cascades to all content and stays an in-app action.)

Apply a bounded list of structured Blender operations. This changes the scene and may delete data; preview the exact operations and obtain user approval in Blender when requested.

Capture the active Blender viewport to the extension-managed cache and optionally add it to the caller's Kolbo media library. This creates a file/media record but does not modify the scene.

Execute Python inside Blender with full machine-level authority. The code can read files, access the network, run processes, and alter the scene. Show the exact code to the user first; it always requires in-Blender approval unless trusted mode is visibly active.

Perform a sensitive Blender file operation. Opening, replacing, or writing an arbitrary path can overwrite data and always requires explicit in-Blender approval unless the user enabled trusted mode for this process.

Read the current state, absolute expires_at, and bounded result/error for one Blender command owned by the caller. Commands and idempotency claims expire after 24 hours. awaiting_approval is not a polling state: stop and wait for the user to approve or deny it in Blender.

Queue a read-only scene inspection in Blender. Returns a command record; if status is queued or running, use blender_get_command_status. If approval is required, stop and let the user approve in Blender.

Import one Kolbo media item or Kolbo-owned HTTPS media/CDN asset into Blender. URLs are restricted to the extension's exact v1 host allowlist and revalidated after redirects. GLB imports as a named collection; images support plane/material/world; video imports to the sequencer.

List only the caller's currently registered Blender processes. Use this before any Blender command when more than one Blender window may be connected.

Start a Blender still or animation render. V1 animation renders are host-enforced at 250 scene frames and 100,000,000 pixel-frames. Rendering can consume substantial machine resources and writes managed output files; approval may be required in Blender.

Inspect the connected Blender runtime and return relevant official Blender API/manual URLs. The extension does not fetch those documentation pages or execute returned code.

Undo the most recent Blender change in the selected session. Undo is itself state-changing and can discard later work, so do not call it speculatively.

Browse the licensed SYNCI catalog. Playback remains watermarked; final use requires acquire_clean_music_track.

Soft-delete up to 1000 media items in one call. Items go to trash (30-day recovery). Owner only — items not owned by the user are silently skipped (count returned in response). Use this for "clean up all my old [type]" or "delete the failed generations from yesterday".

Move up to 1000 media items to a different project in a single call. Caller must own ALL items AND have access to the target project — if any item isn't owned by the caller, the entire operation is rejected (atomic).

Move MANY sessions into one project in a single call — the tool to use when reorganizing a user's library ("file all my Acme work into the Acme project", "clean up the API Generations bucket"). Each session carries ALL of its generations and media with it. Prefer this over looping `move_session`: one call handles up to 100 sessions, while `move_session` is rate limited per call. Sessions of mixed types (chat + image + video) can go in the SAME call. Each session moves independently, so one that cannot move — a generation still running, a Creative Director session, or one you lack edit access to — does NOT block the rest; check `failed[]` in the result and report those to the user. Resolve session ids with `list_sessions` and the target project id with `list_projects` first.

PERMANENTLY delete up to 1000 media items. NOT REVERSIBLE — removes from MongoDB, S3, folders, and source generation records. Owner only. ALWAYS confirm with the user before calling; this is the bulk equivalent of `permanently_delete_media`.

Restore up to 1000 trashed media items at once. Owner only. Returns the count restored and how many ids weren't in trash (already active or not owned).

Cancel this account's pending or processing URL-download job. Does not delete a previously completed file.

Request cancellation of a Flow run. Stops future work; already submitted provider jobs may still finish and cost credits. Poll until confirmed; cancel_requested is not cancellation confirmation. Keep late results attached to the original run.

Cancel a running generation and release/refund its credits. Works for every generation type the SDK tracks (image, image edit, video, image-to-video, video-to-video, elements, first/last frame, lipsync, music, speech, sound, 3D, motion graphic, global image/video edit, creative director). Use when the user says "stop", "cancel that", "never mind" while something is still running, or to clear a stuck generation. Returns how many credits came back. A generation that already finished or failed cannot be cancelled — you get a 409 with its current status, which is not an error worth retrying. For Shorts Creator jobs use shorts_cancel instead.

Cancel a running trend run and release its unspent credits. Free-trial runs cannot be cancelled.

Stop preparation/extraction and expire your private snapshot. A paid provider request already accepted may still complete and be charged. Source media in the library is unchanged.

Fetch messages in a chat conversation. Returns role, content, model, and any media URLs attached to each message.

List the user's chat conversations across ALL projects, most-recent first. Returns session_id, name, project_id, and activity timestamps. Pass `project_id` to narrow to one project (resolve names via `list_projects`).

Send a chat message to Kolbo AI. Starts a new conversation (omit session_id) or continues an existing one. Returns the assistant response when complete. Supports image/video/audio analysis via media_urls — pass public URLs and the model auto-routes to a vision-capable model (e.g. Gemini) when media is detected. Supports web search and deep think modes. LOCAL FILE (on the user's machine, or attached to this chat)? NEVER reply that you cannot upload files — Kolbo uploads it for you. Call `media_upload_widget` so the user picks the file (claude.ai web/mobile), or `create_upload_ticket` and POST the file yourself if you can run shell/HTTP commands; either way you get an https:// URL to pass here. Ignore any "absolute local path" wording in the args below — this server cannot read the caller's disk, so a path will fail.

Check your remaining Kolbo credit balance.

Strip voices that are still faintly audible in a "without dialogue" (M&E) track produced by separate_audio_stems. Use it only when the user actually hears speech bleeding through — it is not part of the normal flow, and it trades fidelity to do its job (each cleanup pass re-separates the bed, so the M&E comes back less faithful than the original split; it runs up to 3 passes until the voices are gone). Typical on dense crowd scenes; unnecessary on clean dialogue. Costs 17 credits, charged even when the track turns out to be already clean (the analysis pass still runs) — the response says which happened. Runs inline. LOCAL FILE (on the user's machine, or attached to this chat)? NEVER reply that you cannot upload files — Kolbo uploads it for you. Call `media_upload_widget` so the user picks the file (claude.ai web/mobile), or `create_upload_ticket` and POST the file yourself if you can run shell/HTTP commands; either way you get an https:// URL to pass here. Ignore any "absolute local path" wording in the args below — this server cannot read the caller's disk, so a path will fail.

Clone a custom TTS voice from an audio sample (the user's voice or any voice they own the rights to). Costs credits (provider-dependent) — confirm with the user before firing. Providers: "elevenlabs" (recommended, 2s–180s sample) or "deepdub" (2s–300s). After cloning, the voice appears in `list_voices` and can be used with generate_speech. Free plan caps at 3 custom voices. LOCAL FILE (on the user's machine, or attached to this chat)? NEVER reply that you cannot upload files — Kolbo uploads it for you. Call `media_upload_widget` so the user picks the file (claude.ai web/mobile), or `create_upload_ticket` and POST the file yourself if you can run shell/HTTP commands; either way you get an https:// URL to pass here. Ignore any "absolute local path" wording in the args below — this server cannot read the caller's disk, so a path will fail.