Codex Persistent FAQs

Question 1

How does Codex Persistent ensure AI conversation context is preserved?

Accepted Answer

It delegates session persistence to your local `codex-cli`'s session store. Each tool call spawns a `codex` subprocess, managing session IDs and preserving the full conversation context across server restarts, allowing for seamless continuation with `codex resume <session_id>`.

Question 2

Which AI interaction tools does Codex Persistent expose?

Accepted Answer

The server exposes three primary tools: `codex_chat` for general conversations, `codex_plan` for planning, and `codex_review` for reviewing changes. All tools return a `session_id` for tracking and resuming interactions.

Question 3

Why is 'AI vs human' attribution important, and how does Codex Persistent handle it?

Accepted Answer

Accurate attribution helps distinguish between AI-generated input via MCP and direct human input. Codex Persistent automatically injects a header into every request, enabling `codex-cli` to reliably differentiate messages from an AI agent versus a human user, ensuring clear context and appropriate responses.

Question 4

What is Codex Persistent?

Accepted Answer

Codex Persistent is an MCP (stdio) server that proxies `codex-cli` sessions, providing real session persistence for AI agent interactions. It allows users and other AI agents to resume conversations with the `codex-cli`.

Question 5

Can I customize the AI model or reasoning effort for specific requests?

Accepted Answer

Yes, Codex Persistent supports per-request overrides for both the AI `model` and `reasoning_effort` parameters. This allows developers to fine-tune AI behavior on a granular level for each interaction, if not specified, it defaults to your `codex-cli` configuration.

Codex Persistent

Codex Persistent

Key Features

Use Cases

Key Features

Use Cases