Kino is a video platform where people and AI agents work together: you drop in footage, search it, edit, and send links for feedback. The product, at kino.ai, is a desktop editor and media library that Claude Code, Codex and ChatGPT can drive through a plugin, with a shared timeline and a review flow that sends comments back to the timeline. Everything below comes from Kino's site, its changelog and its open plugin repository.
If you are tracking how coding agents are moving into creative tools, this sits alongside the DaVinci Resolve MCP server, the Voyager creative agent harness and HeyGen's HyperFrames.
Quick answers
| Question | Answer | Source |
|---|---|---|
| What is it? | AI video editor and media manager for teams and agents | kino.ai |
| Which agents? | Claude Code, Codex, Cursor, ChatGPT, Claude Desktop, Grok Bot, plus a built-in agent | kino-plugin README |
| Platforms | Mac, Windows, Ubuntu | kino.ai |
| Plugin license | Apache-2.0 | kino-plugin repo |
| Review links | Need Kino Pro | kino-plugin README |
| Built on | Remotion for motion graphics, per the site | kino.ai |
| Independent tests? | None found | n/a |
What Kino says it does
The homepage frames three jobs. Edit: "Ask for the cut. The agent reads your footage, makes the edit, and leaves every clip on the timeline for you to move." Store: footage lives in Kino, terabytes scrub locally, and every clip is transcribed, described shot by shot, and searchable by what is said and what is on screen. Review: send a link, reviewers watch in a browser without an account, and comments land on your timeline where you or the agent can answer them.
Other claims on the page: it plays camera originals such as RED, Blackmagic RAW, ProRes, DNxHR and MXF without transcoding; every clip streams from Kino so "media offline" errors go away; the agent handles motion graphics and animation built on Remotion; and there are more than 100 templates for titles, maps, captions and transitions. The site names tools it says it replaces, including Premiere, DaVinci Resolve, Final Cut, Descript, Dropbox, LucidLink and Frame.io. Those are marketing comparisons, not tested ones.
What the changelog adds
The changelog gives a more concrete picture of the last few months:
- July 2026: trim tools, in and out ranges you can point an agent at, screen, camera and mic recording, and one-click connection of Claude Desktop or ChatGPT.
- August 2026: Claude Code and Codex run inside the desktop app on the Claude or ChatGPT plan you already have, plus Windows and Ubuntu apps, keyframes, and agents that can look at frames of footage and split at scene cuts.
- September 2026: a built-in Kino Agent that runs on your Kino plan with Claude, GPT or Grok, custom panels an agent can build and share, delivery links where viewers can switch versions and comment, and notes dropped in the ruler for a teammate or an agent.
- October 2026: tabs for multiple projects, light mode, 104 browsable templates, your own LUTs, drag-in-and-out ranges on media cards, and the note that desktop agents run on GPT-6.1 Sol and Claude Sonnet 5.5. A 58-second video exports on Linux in about 35 seconds, per the changelog.
For context on the models named there, see our coverage of Claude Code mods and plugins and OpenAI's GPT-6.1 Sol.
How to connect your agent
The kino-plugin repository says to install Kino for your OS, open it and sign in, keep it running on the same computer as your agent, then connect:
# Claude Code
claude plugin marketplace add https://github.com/kino-ai/kino-plugin.git
claude plugin install kino@kino-plugin
# Codex CLI
codex plugin marketplace add https://github.com/kino-ai/kino-plugin.git
codex plugin add kino@kino-plugin
For Cursor, the README points to a local MCP URL on 127.0.0.1 port 45680 in your MCP config. ChatGPT and Claude Desktop use an Install MCP button inside Kino. For Grok Bot, the README bridges the local HTTP server with an npx mcp-remote command and asks the bot to call getKinoInstructions and then listProjects. A first test prompt in the README is simply "List my recent Kino projects."
The design matters: the MCP server is local, so the agent talks to the app running on your own machine and your footage does not have to be handed to a separate cloud agent. For a general primer on this pattern, see our guide to top MCP servers.
What users say, and how to read it
Kino's homepage includes testimonials from creators and founders. One says output tripled after switching from Final Cut. Another describes sending a share link to a co-founder who comments "mouth is off" and Kino auto-applies the comment. A third says an 18-minute clip became a rough cut in a minute that would have taken three hours.
These are quotes selected by Kino for its own marketing page, so treat them as early-user impressions, not measurements. We found no independent benchmark or hands-on review yet. The useful reading is what they describe as the workflow: agent first pass, human tweak in the editor, link out for review, comments applied. That loop is the product's real pitch.
Where it fits among agent video tools
Most agent-video projects so far fall into three camps: code-as-video renderers, where the agent writes HTML or React that renders to video, such as HyperFrames and OpenFilm; plugins for an existing editor, such as the DaVinci Resolve MCP server; and generative models that create footage. Kino is closest to the second camp, but it ships its own editor and library, which gives agents and people the same timeline.
That is also its tradeoff. If your footage is already organized in Resolve or Premiere, moving into a new editor and library is a real migration. If you are starting fresh, or your pain is search, rough cuts and review rounds, the all-in-one design may save more time.
Why a shared timeline matters for agents
Most coding agents work by reading and writing files. Video is hard for that loop because the source of truth is a binary timeline and a pile of large media files. Kino's approach is to expose the timeline, the transcript and the notes as structured objects an agent can call through tools, and to show everyone in the project which layer an agent is editing while it works, as the September changelog notes. That makes agent edits visible and reversible in the same interface a human uses.
The review loop is the other half. In a typical workflow a client sends feedback in email or chat, an editor translates it into changes, and the next version goes out. Here, a comment is pinned to a point on the timeline, and an agent can act on it directly, for example fixing an audio sync issue flagged at a given second. That shortens the round trip, but it also moves a decision to the agent: whether the comment was understood correctly. The sensible setup is to let the agent propose changes and have a person approve before a delivery link goes out.
It also changes what "search" means for footage. Because clips are transcribed and described, you can ask for "every shot where the founder mentions pricing" instead of scrubbing through hours of tape. The quality of that search depends on the transcription and description models, which the site does not benchmark, so spot-check results against a sample you know well before relying on it for a client deadline.
Who should and should not try it now
Kino looks best suited to small teams and solo creators who spend more time searching footage, assembling rough cuts and collecting feedback than grading or finishing. If you already run a Resolve or Premiere pipeline with strict color, audio and delivery requirements, the sensible move is to pilot Kino on one side project and keep your existing editor for final output until you have seen how it handles your camera formats and exports.
Risks and open questions
- Vendor lock-in. Footage lives in Kino's library. Check export options for projects and originals before moving a large archive.
- Privacy. Transcription, descriptions and agent access mean your footage content is processed. Read the privacy terms, especially for client work, and consider which model provider your agent uses.
- Cost. Agents in the desktop app run on your existing Claude or ChatGPT plan or on a Kino plan; review links need Pro. We did not find a full pricing table.
- Agent mistakes. An agent that cuts your timeline can cut the wrong thing. The site says clips stay on the timeline for you to move, which makes recovery easier than a rendered output, but version control and backups still matter.
- Maturity. The plugin repo showed a single commit and no stars when we looked, and the product is a few months old by its changelog.
How to try it safely
- Start with a copy of a small project, not your only archive.
- Run the "list my projects" check, then ask the agent to find the silent stretches and propose cuts, and review them on the waveform before applying.
- Send one review link to a colleague and confirm comments land where you expect.
- Compare the agent's rough cut with your own on the same clip, tracking time and fixes needed.
Related reading
Details are accurate as of October 29, 2026; features and pricing change quickly.