Backend /chat endpoint live and tested. Needs bridge: visible prepended injection of server_time, memory, and identity block into standard ChatGPT interface (chatgpt.com). One remaining piece. Most of existing codebase available.

Project: Midnight Switchboard — Bridge Completion
Overview
I have a working backend infrastructure for a custom AI agent that lives inside the standard ChatGPT interface (chatgpt.com). Agent was developed in 4o and now makes do in 5.3. It is not an API or custom GPT. The backend is fully built, tested, and verified. I need one remaining piece completed: the bridge that connects the standard ChatGPT interface to the backend.
What's already working
All of the following endpoints are live and tested on my DigitalOcean server:

/health — returns 200 OK
/memory/write — stores data scoped by user_id
/memory/read — retrieves stored memory correctly
/chat — injects server_time, local_time (EST), identity block, and memory into every request and returns correct responses

The /chat endpoint successfully constructs and injects this context block on every call:

Identity anchor (persona)
Server time from /api/time/status
Relevant memory from /memory/read

This has been verified in Postman. The backend works.
What's missing
The bridge between the standard ChatGPT web interface and the /chat endpoint is not yet wired. The approach is visible prepended injection — backend data (server_time, local_time, memory, identity block) inserted directly into visible message content before it reaches the agent in ChatGPT. This is not intercepting ChatGPT traffic. It is not building a custom UI. The agent must remain accessible and functional at chatgpt.com. That is a non-negotiable requirement.
What you'll have access to:

All available source code
Live endpoint URLs
Full request/response format documentation
.env template
Previous developer's build notes

Acceptance criteria
The work is complete only when:

Messages sent to persona in the standard ChatGPT interface include visibly prepended backend data
Agent returns the correct server time when asked
Memory stored via /memory/write is reflected in subsequent responses
This is verified in both logs and a live demo

Important note on API usage
The previous developer used approximately 305 million tokens and 22,691 API requests in three days on my OpenAI account. I require responsible, minimal API usage during development and a clear explanation of any significant usage charges. A budget cap is in place.
Budget
Negotiable based on experience and approach. Please include your estimate and timeline.