Skip to content
Prism

What Prism is

Prism connects the AI you already pay for to the app you already work in. It is an MCP server: your AI client talks to it, it talks to a panel inside the creative app, and what comes back is native, editable work.

Your AI client  ──MCP──▸  Prism  ──▸  a panel inside your app  ──▸  real layers
You bringYour own AI. ChatGPT, Claude, Gemini through Antigravity, Kimi, Qwen, or anything that speaks MCP
Prism bringsThe tools, the app knowledge, and the connection
You getNative layers and keyframes, not a script and not a rendered clip
You keepYour subscription. Prism takes nothing for the model that drives it

Already pay for ChatGPT, Claude or Gemini?

That is the only AI subscription you need. Prism plugs into it. There is nothing else to buy and no API key to generate.

You already haveWhat you need from Prism
ChatGPT, Claude, Gemini through Antigravity, Kimi, Qwen, or any MCP clientA free account and the panel
A paid plan with any of themNothing extra. Prism does not resell model access
No AI subscription yetAny of the above. Free tiers work

You also keep everything that assistant already knows about you — your past projects, your naming, the way you brief. Prism reaches After Effects from inside that conversation, so you are not starting again somewhere new.

Test it with what you haveSign in, paste one address, ask it about your comp
Keep your interfaceNo second window, no second bill
Change your mindSwitch client whenever. The address is the same for everyone

You choose the model, and you keep choosing

Any clientChatGPT, Claude, Gemini (through Antigravity), Kimi, Qwen, Codex, Cursor, OpenCode, Cline
Any modelIncluding anything you reach through OpenRouter, or a model you host yourself
Any timeSwitch whenever you like. The address is the same for everyone and carries no secret, so nothing in your setup moves

Prism has no way to know which model is driving: the connection carries the client, not the model. It authenticates a person, offers a set of tools, and executes what the client asks for. There is nothing per-client to build, so a new client generally connects as soon as it ships spec-compliant MCP support.

Which client connected is recorded, by name and version, as product analytics.

That also means the conversation stays where you already work. You are not learning a second interface, and the context you have built up in your assistant comes with you.

Built to spend fewer tokens

A connected agent costs money per call, so the architecture is shaped to send less. Five choices do the work:

A gateway, not a catalogueYour client loads three tool schemas rather than forty-five. The rest are fetched by name on demand through discover
Reads are scopedThe default project read returns compositions without their layers. Full detail is opt-in, because on a real project it is large
A scene is one callA whole composition is described in one document and built in one round trip, rather than a call per layer
Dry run before touching anythingA scene is validated and compiled without reaching After Effects, so structural and timing mistakes are caught before a round trip pays for them. Text and layout boxes still need real measurement in After Effects
You write TypeScriptThe server compiles it to the ES3 engine After Effects runs. The model works in its native idiom instead of burning turns on a 1999 dialect

The last one is the largest. Hand-written ES3 costs an agent reasoning budget and repeated corrections on syntax the model rarely gets right first time, and none of that produces work in your timeline.

What you are left with

The output is your project, not a picture of it.

LayersNative. Select, move, restyle
KeyframesReal. Drag them, retime them, re-ease them
TextStill text. Retype it, change the font
ExpressionsReadable, and yours to edit
UndoOne step in After Effects history

Nothing depends on Prism afterwards. Cancel your plan and the work stays exactly as it is, because it was never anything other than an After Effects project.

The loop

1
Read
Your AI asks Prism what is actually in your project — comps, layers, timing, real names. Not a guess from a template.
2
Reason
It plans against what is there.
3
Write
Prism executes inside the app. One undo step. Ask for a checkpoint first and it saves a timestamped copy of your project.
4
See
Frames come back inline, so it checks its own work instead of assuming.

What it runs on today

AppState
After EffectsShipping · get started
More appsThe MCP surface is app-agnostic by design; the panel is the only per-app part