How giselles-ai/giselle works
Project type and purpose
• This is a full-stack, self-hosted AI agent builder called “Giselle.” It consists of two Next.js applications (“playground” for experimenting and “studio.giselles.ai” for the production UI), a collection of shared TypeScript packages implementing core engine logic, data models, workflow orchestration, integrations, and utility libraries, plus database migrations and infrastructure-as-code for deployment.
• The README highlights that Giselle enables “agentic workflows” (chaining LLM calls, file operations, and integrations) in a cloud or self-hosted environment.
File structure
• Top-level “apps” directory:
• apps/playground – a Next.js 13 app for interactive sandboxing of agents.
• apps/studio.giselles.ai – the main Next.js 13 application with pages, API routes (auth, files upload, webhooks, workspace-run), and middleware.
• “packages” directory: dozens of internal packages providing:
• giselle-engine (core HTTP handlers, Next.js adapters, React contexts)
• data modeling (data-type, data-mod, flow)
• UI components (text-editor, workflow-designer-ui)
• SDK (giselle-sdk)
• Integrations (github-tool, supabase-driver, stripe via services/external, language-model providers)
• Telemetry, usage-limits, utilities, test-utils.
• Configuration and infra: Drizzle ORM config, SQL migrations, Supabase storage driver, Vercel/turbo/pnpm config, GitHub Actions workflows, Opentelemetry setup, CLI tools under “tools.”
README for additional insights
• Confirms a cloud-hosted service with auth, usage plans, free tier, and a self-hosting guide.
• Emphasizes AI-driven workflows that connect to Langchain-style LLM providers.