Memory for Codex CLI

Memory for Codex CLI — AGENTS.md + MCP Memory | Awareness

Give Codex CLI persistent memory: AGENTS.md workflow rules, local Markdown memory, and shared context with Claude Code and Cursor. Free, no account.

Install locally

npx @awareness.market/setup --ide codex

Local-first. No account required. The setup CLI writes the rules and MCP config for this IDE.

The problem

Codex CLI starts every session from zero.

AGENTS.md says how; memory says what

AGENTS.md is a static instruction file. It cannot tell Codex what the last session decided, which approach failed, or what is still open. Awareness adds that recall layer and writes it back as the agent works.

Recall happens before planning

The injected rules make the agent load recent decisions and constraints at session start, so it does not re-derive what the last session already settled.

Local-first by default

Memory lives in Markdown and SQLite under your project. No account, no API key, and nothing leaves the machine unless you opt into cloud sync.

One brain across IDEs

The same daemon serves every editor and agent. A decision recorded in one tool is available in the next, with project isolation by directory.

How it works

Recall before work. Record after.

01

Install the memory layer

Run the setup CLI. It detects the IDE, writes the MCP config, and injects the workflow rules that make the agent recall before it plans.

02

Keep working as usual

No new commands to remember. The agent initializes memory at session start and records decisions as it works.

03

Switch IDEs without losing context

Open the same project in another editor. The task board and memory are already there.

Awareness injects workflow rules into Codex CLI's configuration, so the agent initializes memory at session start, recalls before work, and records decisions after every change. Retrieval runs on local SQLite with hybrid semantic + full-text search; it makes zero LLM calls.

What gets remembered

Knowledge, not a transcript.

  • Decisions and the reasons behind them
  • Constraints, pitfalls, and failed approaches
  • Open questions and the next concrete step

Compare

Where Awareness fits.

CLAUDE.md / AGENTS.md alone

Static files someone has to maintain. Awareness is written by the agent as it works, and recall is automatic.

Cloud memory APIs

Managed and remote. Awareness is local-first by default: Markdown + SQLite on your machine, no account.

Editor chat history

Locked to one tool and one session. Awareness is shared across IDEs and survives restarts.

Frequently asked questions

Does Awareness replace AGENTS.md?

No. Keep your AGENTS.md for conventions and guardrails. Awareness adds the memory layer: recall before planning, record after changes.

How does Codex connect?

The setup CLI writes ~/.codex/config.toml with an MCP server entry and injects the workflow rules. You keep using the same CLI.

Can Codex share memory with Claude Code?

Yes. Both use the same local daemon and project directory, so the task board and memory are shared.

Does this require an account?

No. Local-first memory is free forever and needs no account or API key. Cloud sync and team features are optional.

Where is the memory stored?

In a .awareness/ directory inside your project: Markdown files plus a SQLite index. It is portable and git-friendly.

How is this different from a rules file?

A rules file is static. Awareness records what actually happened — decisions, pitfalls, open questions — and the agent recalls it before planning.

Give your agent a memory that compounds.

Local-first, no account, no credit card. Start in 30 seconds and keep the data on your machine.