1flowbase README

1flowbase

Community: WeChat | WeChat Official Account | Twitter

From AI Gateway to a Complete Application System.

1flowbase is a low-code native backend foundation built on an AI gateway. Beyond core AI gateway features, it lets you build your own application system on top of the AI chat data stored in the gateway.

Why 1flowbase

An Application Foundation Designed for Agents

1flowbase is API-first by design. If a traditional GUI can take over the entire system through APIs, then turning those APIs into MCP lets an Agent take over everything as well.

1flowbase Business Architecture

Technical Architecture

1flowbase Technical Architecture

1flowbase is a self-hosted AI Gateway and AI application runtime.

Start by unifying model access, retain conversation data in the same runtime, and keep building Workflows, APIs, business data, and React applications on top of it. Every capability is API-based and can be exposed to Agents through MCP to understand, build, and manage.

Dispatch intelligence, retain data, build applications.

Philosophy

The gateway is not the end: Full AI conversations passing through the gateway are stored in PostgreSQL, becoming a reusable data asset for cost analysis, model quality evaluation, prompt / routing optimization, and business system development.
A "model" can be a Workflow: High-capability models plan, judge, and audit; low-cost models execute. The combination is published as a single virtual model, and a model can itself be used as a Tool by another model.
Three protocols, one entry point: Conversion between Anthropic Messages, OpenAI Chat Completions, and OpenAI Responses, so your application layer doesn't change when providers do.
Designed for Agents: Every runtime operation available in the GUI is built on APIs, which are exposed to Agents via MCP. The MCP Gateway uses progressive list / get / call tool discovery instead of stuffing every Tool into the context at once.
Low-code efficiency + native React freedom: Business tables can be created dynamically; UIs are written as React code blocks with access to the entire React ecosystem, not a closed drag-and-drop editor.

text

Human → GUI ─┐
             ├→ API → AI Gateway / Workflow / Data / React UI → AI Application
Agent → MCP ─┘

Quick Start

Docker

Deployed with Docker — no manual environment setup required.

Linux / macOS

Terminal

curl -fsSL https://raw.githubusercontent.com/taichuy/1flowbase/main/scripts/shell/docker-deploy.sh | sh

Windows PowerShell

powershell

irm https://raw.githubusercontent.com/taichuy/1flowbase/main/scripts/powershell/docker-deploy.ps1 | iex

Windows CMD

cmd

powershell -NoProfile -ExecutionPolicy Bypass -Command "irm https://raw.githubusercontent.com/taichuy/1flowbase/main/scripts/powershell/docker-deploy.ps1 | iex"

Follow the prompts to finish configuration and start 1flowbase. Storing full conversations can generate large data volumes; for high-traffic production, plan storage capacity, retention, backup, and compliance in advance.

Git

Requirements

Node.js >=24.0.0
pnpm 12.11.2 (run corepack enable and corepack install --global pnpm@12.11.2)
Rust stable toolchain (Cargo), version 1.99 or newer
Docker with Docker Compose (for middleware such as PostgreSQL)

Start

Run from the repository root:

bash

git clone https://github.com/taichuy/1flowbase.git
cd 1flowbase

Before the first source build on Linux with a user systemd manager, apply the development resource profile and put its user-level wrappers on PATH:

server.ts

bash scripts/shell/apply-resource-limits.sh
export PATH="$HOME/.local/bin:$PATH"

Rust normally uses half the available CPU and memory, targeting at least 2 logical CPUs and 8 GiB when capacity permits. Smaller machines may exceed the half-machine share; memory still leaves OS/service headroom. These are build budget targets, not verified minimum hardware requirements. Explicit local settings remain supported. Windows/macOS use the same CPU policy; Linux systemd memory limits do not apply there. See resource configuration for limits and overrides.

For first-time development, precompile the Rust test targets, then start the services:

bash

node scripts/node/reset-rust-cache/cli.js --warmup-only
node scripts/node/dev-up.js

--warmup-only fetches locked dependencies and compiles build/test targets in sequence without deleting existing caches or stopping services; it does not execute tests. For subsequent daily starts, run dev-up directly to reuse the caches.

dev-up starts the Docker middleware, installs frontend dependencies, builds and runs the backend, and starts the frontend. Other common commands:

bash

node scripts/node/dev-up.js status
node scripts/node/dev-up.js stop
node scripts/node/dev-up.js restart --frontend-only
node scripts/node/dev-up.js restart --backend-only

Logs are written to tmp/logs/. For more scripts, see scripts/README.md.

Comparison

Project	AI Gateway	Composite Models / Workflow	Full Conversation Data	Dynamic Business Data	MCP	React Apps
1flowbase	✅	✅	✅	✅	✅	✅
LiteLLM	✅	—	Partial	—	—	—
OpenRouter	✅	—	Platform-hosted	—	—	—
Dify	Partial	✅	✅	—	✅	Fixed app types
Supabase	—	—	—	✅	Ecosystem	Build your own
n8n	—	✅	—	—	Ecosystem	—

Roadmap

Out-of-the-box application templates (AI Gateway, multi-model router, enterprise AI Gateway, AI education platform, Agent application backend)
Better official MCP Tool descriptions so Coding Agents like Codex can build applications without reading the source code
Lifecycle management for AI conversation data

Tutorials

Repository Layout

text

web/          Frontend root, powered by pnpm + Turbo
api/          Rust backend workspace
api/apps/     Backend service entry points
api/crates/   Shared backend crates
api/plugins/  Plugin source workspace, HostExtension manifests and templates
docker/       Local middleware orchestration and self-hosted service stack
scripts/      Repository-level development, testing, verification, and debugging scripts

Contributing

Community contributions are very welcome. Before submitting a Pull Request, please run the following verification script:

bash

node scripts/node/verify.js repo

Development guidelines:

Friends

Linux.do - Learn AI, on L Station.
openai - Thanks to the early OpenAI Developer Program for supporting this project.
Aionui - Remotely control AI from your phone.
OfficeCLI - Office suite designed for AI Agents.
deepseek-pp - DeepSeek web chat browser extension.
MuseAI - Local AI companion, text adventure, and interactive fiction app.
FrontAgent - AI Agent system designed for frontend engineering.
RedBox - Localized AI creation workspace for Xiaohongshu creators.

License

This project is licensed under Apache-2.0.

Contributors

Star History

If you want Agents to build and operate self-hosted applications across AI, MCP, application backends, and React interfaces — give 1flowbase a Star.

Report a Bug · Request a Feature