Image: Martini
Why it mattersA video team using several AI tools can now point them at one shared project. Billing and version history travel with it.
A video studio uses Claude and a custom script tool to make a short. It ends with three folders of files and no idea what each one cost. Martini said on Sunday that Bar is its answer. The product gives AI agents and vibe-coded video tools one shared workspace. YC lists Martini in its Winter 2026 batch, and the launch page had 3 votes as of 27 October.
Bar is Martini's API, MCP server, and CLI. Agents and homemade tools connect to one video project. Humans and agents work on the same canvas and timeline. Each shot keeps its own history: the model, the prompt, the references, every take, and the person or agent behind each one. Spend is broken down by person, agent, project and model. Martini says you can set caps per team member and per project. These are Martini's own claims from the launch page.
Which agents it works with, and what it costs
The MCP server targets Claude Code, Codex, Cursor, Claude and ChatGPT. A team's own code can call the HTTP API, with TypeScript and Python clients, or the CLI. That is per the launch page's own FAQ.
Billing has two shapes. A team can bring its own fal.ai key, and generations bill to that account at fal's own prices. For models not on fal, Martini charges in its own credits, called olives. The page gives no price for olives. There is also no customer count or revenue figure, which the launch page does not state.
What Martini says customers have done with it
Martini named four unverified use cases on the launch page. An animated series with "1M+ views". TV commercials "airing internationally". Shows for TV networks. Short-form content for "viral consumer apps". None of these is named, and the view and reach figures are the company's own.
One customer testimonial sits on the page: "What takes us three days in other tools took five hours in Martini." Who said it is not named. It is the company's own quote until an outside source confirms it.
Who is building it
The two founders named on the launch page are Koh and Long. Koh is a cinematographer whose work screened at Cannes' Directors' Fortnight, per the page. Long was the first growth engineer at Supabase. Those are the company's own stated biographies.
What this changes for a team using AI tools to make video
Three problems move. Several people and several agents can work on one project without passing a folder around. Each shot carries its history, so a project can build on the last one. Spend by agent and by project becomes a report. The fal bill is one less thing a producer loses track of. Bar exports to Premiere Pro and DaVinci Resolve as XML. An editor can finish in the tool they know. That is per the launch page.
Three things stay unverified. The customer testimonial is unnamed. The named use cases are unnamed. There is no outside benchmark against Supabase or any other backend. A buyer should ask for a reference customer before trusting the "three days to five hours" number.
Source
Martini Bar: AI video production infrastructure, Y Combinator Launches. Martini, the company.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.
