Feature Request: Let Codex/ChatGPT Use Secure User-Provided APIs as Agent Tools

Hello OpenAI team,

I would like to suggest a feature for Codex and ChatGPT agent workflows:

Allow users to securely connect their own APIs and personal service connectors to ChatGPT/Codex, so the agent can work across the user’s real tools with strict permissions, approval gates, budgets, and audit logs.

Today, developers can use the OpenAI API inside their own applications. I am suggesting the reverse direction as well:

The user should be able to give ChatGPT/Codex controlled access to selected user-owned APIs and services, so ChatGPT/Codex can act as a safe personal operations layer.

Examples:

1. Google Workspace

- Read selected Google Drive folders

- Summarize Google Docs

- Extract deadlines from Gmail

- Add approved events to Google Calendar

- Create a study or project plan based on files, emails, and calendar availability

2. Notion

- Read selected databases

- Find overdue tasks

- Create project summaries from Notion pages

- Update task status after user approval

- Connect notes, tasks, and project documents into one workflow

3. GitHub / Codex workflow

- Read issues and pull requests

- Generate a prioritized Codex task list

- Connect repository state with project documentation

- Update issue checklists after approved Codex work

- Summarize what changed across commits, issues, and docs

4. User-owned backend APIs

- Check app status

- Read logs

- Trigger safe maintenance tasks

- Check local or cloud job status

- Report failures from custom developer tools

- Query a private dashboard and summarize the result

5. OpenAI API orchestration

- Let ChatGPT use the user’s own OpenAI API budget through a secure gateway

- Use smaller API calls as sub-agents for summarization, classification, validation, dataset cleanup, and code review

- Apply budget limits, allowed model lists, max calls per task, and approval thresholds

- Example: “Validate this dataset using my API budget, but stop if estimated cost passes $2.”

6. Desktop or local automation bridge

- Allow ChatGPT/Codex to call a user-approved local bridge for desktop automation

- Example: inspect a project folder and summarize the structure

- Example: prepare a script for an editing or development tool, but ask before changing files

- Example: check a build folder, summarize errors, and suggest the next Codex task

7. Finance dashboards

- Read-only portfolio or transaction analysis

- Risk summaries

- Spending summaries

- No default trading permissions

- Any real financial action should require explicit multi-step confirmation

Important safety requirements:

- Users should not paste API keys directly into chat

- Connections should use OAuth or a secure user-owned gateway

- Every connector should support granular permissions:

  • read-only

  • write with approval

  • limited write scope

  • selected folders/databases/repositories only

  • daily cost or usage budget

- Dangerous actions must require confirmation:

  • deleting files

  • sending emails

  • changing permissions

  • making purchases

  • placing financial trades

  • modifying production systems

- Every action should create an audit log:

  • user request

  • selected tool/API

  • parameters

  • result

  • timestamp

- Rollback or undo should be supported when possible

- The user should always be able to disable a connector instantly

Why this would be valuable:

This would turn ChatGPT/Codex from a conversation interface into a secure personal operations layer for real workflows.

It would help:

- students manage school files, deadlines, notes, and study plans

- developers coordinate GitHub, docs, local builds, Codex tasks, and logs

- creators manage scripts, assets, editing workflows, and publishing tasks

- small businesses coordinate Google Workspace, Notion, finance dashboards, and internal APIs

- advanced users build safe automation workflows across their own tools

This would also make Codex much more powerful because Codex could work not only inside a codebase, but across the surrounding developer workflow: issues, docs, calendars, local tools, custom APIs, and project trackers.

The key idea is not unlimited control.

The key idea is safe delegated control:

ChatGPT/Codex plans the work, the user approves risky actions, controlled APIs execute the task, and every action is logged.

In short:

Please consider allowing users to connect secure, permission-scoped APIs and personal service connectors to ChatGPT/Codex, so it can safely operate across the user’s own tools and workflows.

Thank you.

Hi @Curoc1

I really appreciate the detailed write-up here. The idea of letting ChatGPT/Codex work across user tools with approvals, budgets, and clear permissions comes through well.

I’ll pass this along to the team, this is thoughtful and gives useful context for how people would actually want to use it.

~Smith