IJFW: A Localized Workflow Framework for Unifying Multiple AI Coding Agents

IJFW is an open-source framework that allows unified configuration of 13+ AI coding agents (including Claude Code, Codex, Gemini, Cursor, etc.) with a single installation. It implements shared local memory, intelligent task routing, standardized workflows, and multi-AI auditing functions, running entirely locally without additional costs. It aims to solve the fragmented challenges faced by developers in the multi-AI era, such as high tool-switching costs and memory gaps (where agents cannot share context), providing a connection layer for multi-AI coding collaboration.

With the explosion of the AI programming assistant market, developers face the problem of rich tool choices but high switching costs. A deeper issue is memory gaps: architecture decisions discussed in Claude Code need to be re-explained when switching to Cursor, and business logic sorted out by Gemini has to be restarted in front of Copilot, leading to repeated communication, token waste, and broken experiences. IJFW was created to address these pain points, acting as a connection layer that allows multiple AI agents to share local memory, follow unified workflows, and audit each other.

Installation and Supported Agents

Cross-platform installation via npm command: npm install -g @ijfw/install && ijfw-install, which automatically configures agents already installed in the system. Currently supported: 6 active (Claude Code, Codex, Gemini, Windsurf, Hermes, Wayland), 2 standby (Cursor, Copilot), and 5 planned (OpenCode, etc.).

Core Functions

• Shared Local Memory: Persistent storage across sessions/agents (project knowledge base, conversation history, review records, etc.), stored locally to protect privacy and available offline.
• Standardized Workflow: Covers 5 phases (understanding, design, implementation, review, delivery), gently guiding agents to follow best practices via prompt templates.
• Multi-AI Auditing: After the main agent completes a task, it automatically triggers a second-line AI audit (e.g., code generated by Claude Code is checked for security vulnerabilities by Gemini), reducing the risk of single-point failures.
• Intelligent Task Routing: Matches the optimal agent based on task type (e.g., Claude Code for complex architecture, Codex for quick generation).
• Prompt Cache Optimization: Templates common instructions and reuses project context to reduce redundant token consumption.
• Design Contract: Defines specifications like visual design, components, and code style via DESIGN.md to ensure consistency in output across agents.

• Individual Developers: Eliminates the hassle of repeatedly explaining project backgrounds, allows flexible agent switching without context loss, and multi-AI auditing provides quality assurance for solo projects.
• Small Teams: Unifies collaboration standards, new members get up to speed quickly via shared memory, and multi-AI auditing provides an objective review perspective.
• Technical Leaders: Observes the development process through standardized workflows, audits AI decision trails, and optimizes team workstyles.

• Learning Curve: Requires understanding the shared memory structure and workflow phases to fully unleash its potential; deep usage requires some investment.
• Agent Compatibility: Best support for Claude Code; other agents may have varying feature coverage.
• Local Storage Limitations: No built-in multi-device sync/backup mechanism; users need to solve this via Git/Syncthing themselves.
• Audit Cost: Multi-AI auditing increases API calls; users need to balance quality and cost.

Feature	IJFW	Aider	Continue	Cody
Multi-agent support	✅13+	✅Multi-model	❌Single agent	❌Single agent
Shared memory	✅Local files	✅Git integration	❌Session-level	❌Cloud
Workflow engine	✅Built-in	❌None	❌None	❌None
Multi-AI auditing	✅Supported	❌None	❌None	❌None
Design contract	✅DESIGN.md	❌None	❌None	❌None
Pure local operation	✅Yes	✅Yes	⚠️Partial	❌Cloud
Open-source and free	✅Yes	✅Yes	✅Yes	❌Commercial

IJFW is positioned as a multi-agent coordination layer, forming differentiated competition with Aider (focused on Git integration), Continue (IDE plugin), and Cody (commercial tool).

System Requirements

Node.js 18+, Git, Bash shell (Windows users can get it via Git for Windows).

Installation Commands

• Cross-platform: npm install -g @ijfw/install && ijfw-install
• Windows PowerShell: iwr https://raw.githubusercontent.com/TheRealSeanDonahoe/ijfw/main/installer/src/install.ps1 -OutFile install.ps1 && .\install.ps1 (if blocked by execution policy, run as administrator: powershell -ExecutionPolicy Bypass -File .\install.ps1).

Verification and Plugin Installation

• Verification: ijfw help (terminal) or ijfw help --browser (browser).
• Claude Code plugin: In Claude Code, execute /plugin marketplace add ~/.ijfw/claude and /plugin install ijfw.

IJFW represents a new paradigm in AI tool usage: from single-point tools to a collaborative ecosystem. It recognizes that different AI agents have their own strengths and enables collaborative work through technical architecture, aligning with the trend of integrating AI into human workflows. For multi-AI tool users, IJFW can reduce repetitive communication and improve efficiency. Its open-source nature and active community lay the foundation for continuous improvement, and it is expected to become an important infrastructure for AI-assisted development in the future.