Phewsh Plugins + ion

Phewsh ION - Harness routes - pre check node
Phewsh ION - Harness routes - post check node
Phewsh ION - overview / action input blank
Phewsh ION - connections ( initiate collaborations with other users, local nodes, installed harnesses, etc)
Phewsh ION - Intent TLDR
Phewsh ION - overview / action input partially filled ( optional constraints revealed)
Phewsh ION - Record ( source of truth and interaction from 3rd party, cli, multiplayer clients, etc)
Phewsh - Plugin ( development)
Phewsh - Claude connector
phewsh via chatgpt plugin
phewsh via chatgpt plugin
Phewsh - Plugin information
phewsh via claude connector
Phewsh - Plugin information ctd

Inspiration

Phewsh started as a way to capture human intent and turn it into executable AI-assisted workflows.

As I built with AI tools, a larger problem became obvious: I was constantly switching between Codex, Claude Code, ChatGPT, and other AI systems. Each tool was powerful, but every transition meant losing context, repeating decisions, and rebuilding understanding.

The question became:

Will the next AI know what the last one learned?

During OpenAI Build Week, that question pushed Phewsh beyond a CLI workflow into Phewshion and Phewsh Plugins: a way for multiple AI tools to collaborate around one user-owned project record.

What it does

Phewsh Plugins connect AI tools like ChatGPT, Codex, Claude Code, and other MCP-compatible environments to one shared project record.

Instead of replacing existing AI tools, Phewsh gives them a common foundation:

Project context
Decisions
Handoffs
Revision history
Attribution
Verified project truth

MCP provides the connection between tools.

Phewsh provides the durable project layer around those connections.

In the demo, ChatGPT records a decision through the Phewsh connector, Phewsh ION updates the project revision live, and Claude Code continues from the same project context without requiring the project to be explained again.

One project. Many AI tools. One verified record.

How we built it

The original Phewsh system was built around intent-driven AI development:

Capturing goals before execution
Connecting local repositories to AI harnesses
Maintaining project context across workflows

Build Week expanded that foundation into an interoperability platform.

We built:

Remote MCP connectivity
ChatGPT developer connector support
Claude integrations
Authentication flows
Cross-tool handoffs
Revision-guarded project updates
Phewsh ION, a live project record interface

Codex accelerated development of the interoperability layer, including MCP adapters, authentication, testing, and workflow verification.

GPT-5.6 helped guide architecture decisions, challenge assumptions, improve verification strategies, and refine the final product direction.

Challenges we ran into

The hardest challenge was not connecting AI tools.

MCP already provides a communication layer.

The harder problem was creating trust between AI systems.

When multiple AI tools collaborate:

Which decisions are authoritative?
What changed?
Which tool made the change?
What context should move forward?

Phewsh solves this by making project truth explicit, attributed, and user-owned.

Accomplishments that we're proud of

We built a working interoperability workflow across multiple AI environments.

The demo shows:

ChatGPT writing into a real Phewsh project record
Live revision updates in Phewsh ION
Claude Code retrieving the same project context
Multiple AI systems contributing to one continuous history

We are especially proud that the project itself demonstrates the problem it solves: Phewsh was built by switching between AI tools, and it now helps those tools work together.

What we learned

The future of AI work is unlikely to be one perfect AI assistant.

Different tools have different strengths.

The important layer is not replacing them. It is helping them collaborate around what matters most: the user's intent, decisions, and project history.

We learned that interoperability is not just about moving information between systems.

It is about preserving understanding.