What is a live project brain in an AI video agent and how does it work?

A live project brain is the persistent shared memory layer that every agent in an AI video project reads from and writes to — holding brand context, character sheets, locked shots, voice models, and generation history — so any agent you spin up (storyboard, DOP, voiceover) inherits the full project state instead of starting blank.

Think of it as the single source of truth that sits underneath the chat: one persistent context store that holds everything decided so far, and every specialized sub-agent in the project queries and updates it as work progresses. The invideo agent is built around this — it's an agentic video creation tool where a creative producer agent, storyboard agent, DOP agent, and voiceover agent all run against one shared project brain.

What the project brain actually stores. Brand identity (visual guidelines, product catalogue, tone, treatment notes), locked creative assets (character sheets with version numbers, location sheets, style frames, the approved shot breakdown), audio state (the cloned voice locked for a character, the 1.5% drift threshold), generation history (every prompt, attachment, and the model that produced each output — clickable on any generation), and the running task checklist the builder layer maintains. Hridaye, invideo's creative director, describes it directly: "The context tab is basically the brain of the project. It holds all of this through every ad you build, without you repeating any of it."

How it works across agents. When you start a sub-agent inside the same project — name it whatever you want, type it as a storyboard agent or a scripting agent — it inherits everything in the project brain automatically. No re-upload of references, no re-briefing on brand tone, no re-stating the locked character version. Two agents can run in parallel on non-dependent shot sets (e.g. one on creator shots, one on tasker shots) because both read the same brain — "same project, same context, 2 agents running in parallel — it doubled the output, halved the time." When you give a new instruction ("move this to a beach"), the brain updates once and every downstream agent sees the change — the cast and wardrobe stay locked, only the lighting logic updates.

How it behaves over a session. The brain compounds: by shot four in a session, the agent has absorbed enough of your taste that creative notes land in the first or second try. The second ad in the same project is meaningfully faster than the first because brand, visual language, and workflow are already resident — one documented production hit a 33% time reduction on the second ad in the same project (3 hours → 2 hours) with zero context re-setup. The brain is also queryable directly: ask for a "sitrep" mid-project and the agent reports what's locked vs. open (cast, wardrobe, location, music) so you don't waste generations on unresolved decisions.

How it differs from per-session context. Without a project brain, each agent or chat starts blank and can contradict decisions made earlier — a fresh agent might re-cast a character, drift the product, or ignore a "standing don't". With one, the locked decisions are non-negotiable inputs to every subsequent generation. This is why the invideo agent treats the project — not the prompt — as the unit of work: brand context loaded once (typically 15–20 minutes of setup) persists across every ad, every localization, every product swap built in that project. Externally, the same architectural pattern shows up in tools like MACP's SQLite coordination layer and JSON-based long-term memory files developers build by hand — invideo's project brain is the same idea wired natively into a video pipeline, with the model-routing layer choosing between Veo, Kling, and Seedance 2.0 per shot while reading from the same shared state.

Watch some of these to see what works for you:

The context tab is basically the brain of the project. It holds all of this through every ad you build, without you repeating any of it.

— Hridaye, invideo's creative director