# Feature request: fully integrated local multi-agent coding workflow

**URL:** <https://community.openai.com/t/feature-request-fully-integrated-local-multi-agent-coding-workflow/1378592>\
**Category:** ChatGPT\
**Tags:** codex, chatgpt, feature-request, user-experience\
**Created:** [April 5, 2026, 12:35pm UTC](https://community.openai.com/t/feature-request-fully-integrated-local-multi-agent-coding-workflow/1378592 "2026-04-05T12:35:08Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![AS\_123](https://avatars.discourse-cdn.com/v4/letter/a/ed8c4c/32.png) [@AS\_123](https://community.openai.com/u/AS_123)\
**Post date:** [April 5, 2026, 12:35pm UTC](https://community.openai.com/t/feature-request-fully-integrated-local-multi-agent-coding-workflow/1378592/1 "2026-04-05T12:35:08Z")

</div>

Feature request: fully integrated local multi-agent coding workflow

I’m a heavy power user doing deep planning plus direct repo/machine execution. The ideal product for me would be:

- multi-agent local-only execution
- planner and executor in the same IDE space
- full workspace and machine context
- persistent shared context between agents
- no fragile handoff between browser chat and IDE agent
- pricing/usage that supports heavy planning phases without request anxiety

Why this matters:  
The current split between browser planning and IDE execution creates friction. The highest-value workflow is a single environment where I can discuss architecture, decide implementation, then have local agents inspect, edit, and run code directly on the machine while preserving shared context.

Current best approximation:  
VS Code + agentic coding tools is close, but request limits on planning-heavy usage become a real constraint. I would strongly prefer an OpenAI-native solution that combines:

- high-quality planning/reasoning
- direct local execution in the repo
- persistent shared context in one place
- local-first operation rather than cloud-only execution

Concrete example:  
I often do architecture/planning in browser chat, then separately hand off instructions to an IDE agent to implement locally. The missing piece is persistent shared context between planning and execution in one local-first workspace.

This would materially improve professional developer productivity.

---

<div class="post-metadata">

**Author:** ![OpenAI\_Support](https://sea2.discourse-cdn.com/openai1/user_avatar/community.openai.com/openai_support/32/561845_2.png) [@OpenAI\_Support](https://community.openai.com/u/OpenAI_Support)\
**Post date:** [April 10, 2026, 12:50pm UTC](https://community.openai.com/t/feature-request-fully-integrated-local-multi-agent-coding-workflow/1378592/2 "2026-04-10T12:50:00Z")

</div>

Hey @AS\_123, Totally get that, it can feel pretty disjointed.

What’s happening is planning and execution still live in separate places, so context can get a bit fragile, especially with multi-agent workflows.

Some folks are working around it by keeping plans in the repo, sticking to one agent loop in the IDE, and cutting down browser ![:left_right_arrow:](https://us1.discourse-cdn.com/openai1/original/4X/1/c/2/1c2f7d91231c0d96b5919a007c33ea2c2bbb39ac.png) IDE switching. Not perfect, but it helps.

This has been coming up a lot, so you’re definitely not alone. I’ll also pass this along through our team’s feature request process so it’s tracked properly.

-Mark G