YOUR WORK, WITH THE WHOLE BACKSTORY

Your code has a backstory.
Bring it into ChatGPT.

The decisions in your Codex task. The changes still on your machine.
Give ChatGPT the context behind your next move.

Self-hosted · Read-only local access · MIT open source

YOUR LOCAL WORKSPACE

Codex

Task history the why
Local repository the what
Unpublished changes the now
One task. Your chosen scope.
Context RelayAuthenticated MCPContext retrieved on demand →
YOUR CONNECTED CHAT

ChatGPT

More than the code.
The reasoning behind it.

ReviewDebugPlanPrompt
Evidence for a better next step.

You return the report. Codex checks, implements and validates.

CONTEXT IN. CLARITY OUT.

Pick up the work.
Keep the reasoning.

01

Link what matters.

Choose one Codex task and its repository. Limit paths and approve individual untracked files before sharing.

02

Ask with context.

ChatGPT retrieves relevant discussion, source, diffs and project instructions through your authenticated bridge.

03

Bring the result back.

Return a Markdown report to Codex. Refresh the evidence, reconcile changes and validate the next step.

Guided collaboration, not automatic synchronization. Retrieved context is sent to ChatGPT.

FIVE WORKFLOWS. MANY NEXT MOVES.

A better starting point
for your next question.

Stop rebuilding the backstory.
Start with the evidence that matters.

01 / CODE REVIEW

Review the change
before it leaves
your machine.

Connect an unpublished diff to the decision that shaped it. Get findings with source references, not guesses.

quota.mjs− return count <= limit;+ return count < limit;
02 / BUILD & FIX GUIDANCE

Find the missing piece.

Investigate a bug with task history and current files. Implement with separately available, authorized editing tools.

03 / CONSULTATION

Make the tradeoff clear.

Discuss architecture using the stack, constraints and decisions your project already has.

04 / PROMPT CREATION

Turn an idea into a brief.

Carry the goal, constraints and acceptance criteria into a focused implementation prompt.

05 / PROJECT SETUP

Give your project a home.

Generate Markdown guidance and source documents. Add them to ChatGPT without replacing your existing instructions.

The bridge supplies reads and guidance. Editing, running tests and GitHub actions require separate tools and permissions.

AN ILLUSTRATIVE WALKTHROUGH

The bug is in the code.
The clue is in the conversation.

Imagine a reservation limit. The task says a full quota must reject the next request. A local edit accidentally lets one more through.

Run the offline synthetic demo ↗
“At the limit, reject another reservation.”
return count <= limit;

Equality should reject. Change the comparison to <, then verify the boundary test.

Synthetic example, not a live connection or a recording. The offline demo uses fictional history and a temporary repository.

YOUR INSTANCE. YOUR BOUNDARIES.

Share the context.
Keep control of access.

You run the bridge and supply your own authentication. Each link is tied to one task and repository.

Understand the security model ↗

Self-hosted does not mean all data stays local. Retrieved content is sent to ChatGPT. Redaction is best effort, not a guarantee.

Permitted tracked filesReadable
Nonignored untracked filesIndividual approval
Ignored & secret files, including .envExcluded
Local shell & file writesNot exposed
Task linksExpire & revoke

OAuth authentication · exact owner checks · path scope
32 MiB history default · optional 64 MiB retry

BUILT TO BE YOURS

Your own bridge.
A guided way to get there.

Before you start

Codex Desktop + local repository

Node.js + pinned pnpm

Auth0 + Cloudflare + domain

ChatGPT custom MCP access

The documented live path is Codex Desktop on macOS, with ChatGPT Pro. Check account availability first. Linux, WSL2, standalone CLI onboarding and Plus-specific acceptance remain unverified. Platform details

A FEW GOOD QUESTIONS

Know what
you’re connecting.

Is this a shared service or a one-click plugin?

It is runnable, MIT-licensed software you host yourself. You configure your own authentication and HTTPS tunnel. Installing someone else’s connection does not link your local files.

What does it cost?

Context Relay is free and open source and adds no model API billing. Your ChatGPT subscription, domain renewal and provider usage are separate. The documented personal setup targets Auth0 and Cloudflare Free plans; check current eligibility and quotas before setup. It does not provide unlimited model usage.

Do I need ChatGPT Pro?

Our walkthrough used Pro. OpenAI’s developer guide lists Plus eligibility, while its Help Center describes narrower availability. Verify developer mode and custom MCP access in your own account before upgrading. Check availability

Can ChatGPT build or fix my project?

Its available GitHub tools may edit, create branches, commit or manage pull requests when authenticated and authorized. Context Relay supplies local evidence; it does not supply those write tools. Local changes and remote commits still need reconciliation.

Do I reconnect for every message?

In the same conversation, reuse the valid supplied link and ask for fresh evidence. New local changes are fetched on demand. Links expire after 24 hours; renew the link when needed, not your Auth0 account or tunnel.

What happens when my computer is offline?

Fresh local reads stop. Previously retrieved information may remain in the ChatGPT conversation, but is not current evidence. Restart the bridge and tunnel, check the link and retrieve again.

LESS RE-EXPLAINING. MORE MOVING FORWARD.

Bring the backstory
to your next breakthrough.

Set up Context Relay

Read the code. Own the connection.