## for OpenAI Codex
Long-term memory for Codex
AGENTS.md handles standing instructions. Basic Memory handles everything that compounds.
AGENTS.md is only the short-term layer
Static instructions cover what every session needs to know. The reasoning behind every decision, the gotchas, and the project history need somewhere durable to live.
- —Codex starts fresh every time. No memory of your codebase, past decisions, or what you tried last week.
- —You spend time re-establishing context instead of shipping code.
- —AGENTS.md loads in full every turn. It is not the place for accumulated knowledge.
- —Long-running projects lose continuity because there is no shared knowledge layer.
Basic Memory changes that
Give OpenAI Codex a persistent, searchable knowledge base it can read and write to.
Retrieved on demand, not loaded every turn
Codex searches Basic Memory for the few notes relevant to the current task instead of loading one giant file. AGENTS.md stays small; Basic Memory holds the rest.
Reads and writes the same graph
Decisions, patterns, and project context accumulate as plain Markdown notes. Searchable by full-text, semantic, and graph navigation. Codex reads them when relevant and writes new ones as work progresses.
Plain Markdown you own
Every note is a file on your disk. Read, edit, and version with Git. The same vault works in Claude Code, Cursor, ChatGPT, and any MCP client.
Native MCP integration
Codex supports MCP servers natively. Basic Memory connects directly. No wrappers, no middleware.
One command to get started
Quick Setup
codex mcp add basic-memory bash -c "uvx basic-memory mcp"One command. `uvx` runs Basic Memory in an isolated environment with no install step — handy for trying it out. For a permanent install, run `pip install basic-memory` or `brew install basic-memory` and use `bm mcp` instead of `uvx basic-memory mcp`. For cloud, swap in the remote MCP URL. Verify with `codex mcp list`. Full setup guide →
What users are saying
“Basic memory is the missing 'wow' factor in AI chatbots—now I can't imagine Claude or Claude Code without it.”
Caleb
Caleb Picker Consulting
“Basic Memory turned my scattered notes and half-finished ideas into something coherent. It remembers what matters and builds on it — so I don't have to start from scratch every time I sit down to work.”
@Nellins
Creator
Start building with OpenAI Codex
Try Basic Memory free for 7 days. Your notes stay yours either way.
7-Day Free Trial
$15/seat/month
Graduated team pricing- AI collaboration via MCP
- Shared workspaces for teams and agents
- Team member management and audit logs
- Full-text search across notes and projects
- Desktop, mobile, and web access
- Private, exportable Markdown files
- 50,000 notes per seat
- 1,000 note updates per seat per day
- Cancel anytime, your data stays yours
Open source available for power users
View on GitHubFrequently asked questions
- How do I give Codex persistent memory?
- Connect Basic Memory as an MCP server in your Codex configuration. Codex can then search and update your knowledge base. Architecture decisions, conventions, project context. In every session instead of starting cold.
- How is Basic Memory different from AGENTS.md?
- AGENTS.md is a static instruction file; Basic Memory is a living knowledge base. Codex retrieves relevant notes on demand via MCP search rather than loading one fixed file, and it can write new knowledge back. Use AGENTS.md for standing instructions and Basic Memory for everything you learn along the way.
- Can Codex and Claude share the same memory?
- Yes. That is the point of building on MCP. Codex, Claude Code, Cursor, and ChatGPT can all read and write the same Basic Memory knowledge base, so switching tools never means losing context.
