1flowbase
中文GitHub
Use cases

Connect your models. Build on what they create.

Start with a unified AI endpoint, compose models around your tasks, and turn conversation data into the foundation for your next application.

01

Give your team a shared AI entry point

The limitation

Different applications may use different model providers and API protocols. Your team needs a common way to access and distribute those capabilities.

The approach

Use the AI Gateway to publish models and workflows through OpenAI Chat Completions, OpenAI Responses, and Anthropic Messages, while retaining complete conversations in PostgreSQL.

01Applications and Agents
↓
02AI Gateway
↓
03Models and published workflows
↓
04Conversation data in PostgreSQL
02

Let models work together on a task

The limitation

Planning, execution, and review can have different capability and cost requirements. One model does not have to perform every step.

The approach

Compose a workflow in which one model plans, a worker model executes, and another step reviews the result. Publish it as one virtual model, or make a worker model a callable tool.

Explore a workflow example →
01Primary model: plan
↓
02Worker model: execute
↓
03Review the result
↓
04Virtual model response
03

Build an AI improvement loop from conversation data

The limitation

To improve an AI application, you need to understand what people ask, how Agents respond, where tasks fail, and which calls consume resources.

The approach

Use complete sessions stored in PostgreSQL as a foundation for your own usage, quality, and cost analysis. Apply the findings to prompts, harnesses, routing, and model composition.

01AI usage
↓
02Complete conversation data
↓
03Analysis
↓
04Prompt, routing, and workflow improvements
04

Build an Agent-ready business application

The limitation

A working AI application also needs business data, APIs, and a usable interface. Those pieces need to connect with its models and workflows.

The approach

Combine dynamic PostgreSQL tables, an API-first backend, workflows, and React Blocks. Use MCP list / get / call to discover and invoke exposed capabilities. Templates and improved building without source-code context are in progress; complex apps currently benefit from an Agent working in the project context.

01Business data model
↓
02APIs and AI workflows
↓
03React application interface
↓
04Agent access through MCP