Gamma builds design-first agents with Vercel
By Steven Van ·
Gamma details how its agent, image and deployment pipelines run on Vercel's AI SDK, Preview Deployments and Functions.
Gamma, the AI presentation tool, published details on how it built its agent and image pipelines on Vercel. The writeup covers the infrastructure behind Gamma Agent, the conversational editing feature the company launched in October 2025.
- Gamma built its agent orchestration on the Vercel AI SDK rather than custom code, so it can pass context between agents and manage conversation state without re-architecting its backend.
- With the AI SDK's ImageModelV3 interface, adding a new image model now takes about 30 lines of code (a model ID, cost formula, sizes and capability flags), with tracing and cost tracking handled by shared middleware. Gamma's image pipeline has generated more than 1.5 billion images across 60 models and 20 providers.
- The company relies on Vercel's Preview Deployments, production deployments and Instant Rollbacks instead of building its own release system. With about 20 engineers, Gamma averages more than 250 deployments a day, with a median deploy time of just over 7 minutes and a 99 percent success rate.
- Gamma runs its HTML-to-editor conversion layer, which builds a full Tiptap editor schema inside JSDOM for every AI-generated card, as Vercel Functions.
Vercel
The platform for frontend developers — deploy, preview, and scale web apps and AI agents with zero config.
View Vercel →