Skip to content
← All Glossary Terms

AI Recipes

Definition

An AI recipe is a reusable, single-operation AI task with a structured prompt, typed input schema, and typed output schema. Recipes are the fundamental building blocks of AI automation in JieGou — each one performs one well-defined operation, like scoring a lead, drafting an email, or extracting invoice data.

How AI Recipes Work

A recipe consists of three parts: an input schema defining what data the recipe needs, a prompt template that structures the LLM request, and an output schema defining what the recipe returns. When you run a recipe, JieGou fills the prompt template with your input data, sends it to your chosen LLM provider (Claude, GPT, or Gemini), and parses the response into structured output fields. This means the output is always machine-readable and consistent — not free-form text that downstream steps need to parse.

Recipes vs. Raw Prompts

Unlike a raw ChatGPT prompt, a recipe enforces structure. Inputs are validated against a schema before the prompt is sent. Outputs are parsed into typed fields after the response arrives. This means you can chain recipes into workflows and trust that the data flowing between steps is always in the expected format. Recipes also support knowledge base context via RAG, brand voice guidelines, and version history — features that disappear when you paste prompts into a chat window.

Creating Recipes

JieGou offers three ways to create recipes. First, install a pre-built recipe from a department pack — each pack includes 7-10 recipes designed for your team. Second, use the conversational AI agent to describe what you need in plain English and have it generate the recipe. Third, build a recipe manually in the recipe editor, defining the prompt, input fields, output fields, and model configuration. All recipes support AI-assisted prompt optimization that suggests improvements based on your test runs.

Using Recipes in Workflows

Recipes are designed to be composable. A workflow chains multiple recipes together with branching, loops, and approval gates between them. Each recipe step in a workflow automatically maps outputs from previous steps to its inputs. For example, a lead qualification workflow might chain Prospect Research → Lead Scoring → Email Drafting, with each recipe receiving context from the previous step. This composability is what makes recipes powerful — each one does one thing well, and workflows combine them into end-to-end processes.

Recipe Evaluation and Quality

JieGou includes built-in evaluation tools for recipes. Bakeoffs let you compare how different LLM models perform on the same recipe using multi-judge scoring. Token tracking shows cost per run. Version history lets you A/B test prompt changes. These tools ensure your recipes produce high-quality, cost-effective results over time — not just on the first test run.

See it in action

Start building AI automations with recipes and workflows today.