Home

Codexify

Codexify connects an ordinary ChatGPT conversation to tools running on your computer: files, shell commands, Git, diffs, skills, and other MCP servers. ChatGPT supplies the reasoning; Codexify executes tool calls locally. This workflow uses the ChatGPT conversation's allowance rather than starting a Codex or Work agent. OpenAI's account, model, and service limits still apply.

The recommended connection uses an outbound OpenAI Secure MCP Tunnel. A private owner interface adds persistent chats, queued messages, attachments, goals, browser-assisted continuation, and analytics.

Codexify runs commands with the permissions of your OS account. Workspaces and Git worktrees organize access and edits; they are not an OS sandbox. Read the Security model before connecting sensitive projects.

Start here

Task	Guide
Install or upgrade	Installation
Connect your first conversation	Quick start
Configure tunnels and the ChatGPT plugin	Connecting to ChatGPT
Open the private chat interface	Agent chat and goals
Enable offline continuation and automatic recovery	Browser automation
Explore usage and share a stats card	Analytics
Find a command or setting	CLI reference, Configuration
Diagnose a problem	Troubleshooting, FAQ

Work on projects

Every conversation selects an authorized project, a registered worktree, or a private scratch workspace. The setup picker can switch projects in an existing conversation; the agent must then read the new brief before continuing. Git worktrees isolate concurrent edits. A full task handoff transfers ownership to a new ChatGPT conversation without replacing the workspace.

• Multi-project mode and workspace selection
• Worktree isolation
• Diff checkpoints
• Context, memory, and task continuation
• AGENTS.md and skills

Operate and extend

• Tools reference — native tools, optional features, and app-only controls.
• How it works — transport, task identity, execution, and persistence.
• Bridging MCP servers — reuse configured stdio and Streamable HTTP servers.
• Diagnostics and audit logging — current logging defaults, redaction, and how to turn logging off.
• Security model — credentials, shell authority, browser permissions, and file resources.

Current baseline

These guides describe v2.3.9, reviewed on 22 October 2026. Existing explicit settings survive upgrades, so an older installation can legitimately differ from the defaults described here. ChatGPT's labels and account permissions can change independently of Codexify.

Latest release · Source repository · Project website

---

Reviewed against v2.3.9 on 22 October 2026. Source: README.md · src/registry.rs.