For OpenAI Codex CLI users

Persistent memory for Codex.

The OpenAI Codex CLI writes every session to ~/.codex/sessions as a rollout log, then never reads it back. Start a new prompt and the agent is blank again — the plan, the file map, the failed approaches are all on disk but unused. Tramya reads those rollout logs, turns them into durable memory, and hands the relevant slice back to Codex over MCP so each new session continues the last one instead of restarting it.

Why Codex forgets

AGENTS.md is not memory.

Codex reads an AGENTS.md at the repo root for standing instructions, and that file is useful — but it is static. You write it by hand, it says the same thing on every run, and it knows nothing about what happened in your last three sessions: the migration you half-finished, the dependency that broke the build, the interface you agreed on but never wrote down. Codex CLI keeps a full transcript of each session, yet codex resume only reloads one conversation at a time and only if you remember which one. Everything else is dead weight on disk.

The result is a tax you pay every prompt: re-describe the architecture, re-point at the right files, re-explain the convention you settled yesterday. The knowledge existed. Codex just had no way to consult it.

What Tramya adds

A memory layer between the logs and the agent.

Tramya is a local companion that parses your Codex rollout logs continuously, pulls out the decisions, the touched files, the errors and the outputs, and keeps each one attached to its repository, branch and worktree. It serves that memory through an MCP server that Codex calls natively: get_project_context at the start of a run returns a tight brief, save_memory at the end persists whatever is worth carrying forward.

Because Tramya indexes every agent on the machine rather than Codex alone, a decision you made in Claude Code is available to Codex the same day, with its provenance intact. The index is a SQLite file in ~/.tramya/; search runs offline and no source code leaves your disk unless you explicitly turn on encrypted Cloud sync.

Set it up

Three steps to a Codex that remembers.

1. Install the app

Download the signed Tramya app (macOS, Windows, Linux). On first launch it runs a local companion on 127.0.0.1:4317 and finds your existing ~/.codex/sessions history automatically.

2. Register the MCP server

Click “Install in my agents”. Tramya writes the MCP entry into your Codex config, so the memory tools are live on the next codex run — no hand-edited JSON.

3. Let sessions accumulate

From then on each Codex run reads the context the previous ones left and saves its own. The longer you use it, the sharper the recall — especially across worktrees and long task queues.

Frequently asked questions

How do I make the OpenAI Codex CLI remember previous sessions?

Install Tramya, then click “Install in my agents” to register its MCP server for Codex. Tramya indexes your ~/.codex/sessions rollout logs and exposes them through get_project_context, so a new Codex run inherits the decisions and files from earlier ones without you resuming a specific transcript.

Is this different from Codex's built-in AGENTS.md?

Yes. AGENTS.md is a static file you maintain by hand. Tramya's memory is generated from what actually happened in your sessions — decisions, errors, touched files — and updates itself as you work, so it reflects the current state of the project rather than a fixed set of instructions.

Does it read my Codex logs without sending them anywhere?

Correct. Parsing and indexing happen locally, and the SQLite index lives in ~/.tramya/. Nothing is uploaded unless you enable Cloud sync, in which case the vault is encrypted on your device (AES-256-GCM) before it leaves.

What does it cost for a single Codex user?

The Free tier is free forever on one machine and already covers Codex. Local Lifetime (49 € one-time) unlocks every connector; Pro and Cloud add multi-device use and encrypted sync.