Persistent memory for Codex.

Codex is great at running a task end to end, but it treats each task as an island. Once a task is done — or archived — its context is gone: the next run rediscovers the repo, re-asks the conventions, and often re-proposes an approach a previous task already rejected. The work happened, the reasoning existed, but nothing carries it forward.

It gets worse across tools. You settle a design in Claude Code, then open Codex to run the migration — and Codex knows none of it. The history is on disk, task by task, but no agent consults another's. Context is lost at the boundary between tasks and between agents, by default.

Tramya is a local companion that indexes your Codex task history, keeps each task attached to its working directory, and extracts the decisions, touched files and outputs worth keeping. It exposes all of it through an MCP (Model Context Protocol) server that Codex calls natively. When a task starts, Codex queries get_project_context and receives a condensed brief; when it ends, it calls save_memory to persist a structuring decision for the next one.

Nothing leaves the machine: the SQLite index lives in ~/.tramya/, and because Tramya indexes every agent — not just Codex — a decision made in Claude Code, Cursor or Gemini CLI is available to Codex too, with its provenance intact.

1. Install the Tramya app

Download the signed app (macOS, Windows, Linux) from tramya.com. On first launch it starts a local companion on 127.0.0.1:4317 and detects the local history of the agents you already use, Codex included.

2. Register the MCP server

From the dashboard, click “Install in my agents”. Tramya writes the MCP server entry into Codex's configuration, so the tools are available on the next run with no manual JSON editing.

3. Index active & archived tasks

Tramya parses your Codex tasks — running and archived — extracts entities, decisions and code blocks, and keeps each tied to its repo, branch and worktree so nothing gets mixed up between features.

4. Consume & save context

Codex calls get_project_context at the start of a task and save_memory before it ends. The next task — even from Claude Code or in another worktree — resumes with the right decisions in hand.

Claude Code → Codex handoff

You design in Claude Code and execute in Codex. Tramya carries the decisions across, so Codex starts the migration already knowing the plan instead of re-deriving it. See the Claude Code to Codex guide.

Long-running task queues

When you fire off many Codex tasks over a day, each one benefits from what the earlier ones learned — recurring fixes, fragile files, validated workarounds — instead of relearning them every time.

Teams on shared repos

With an optional encrypted vault, the project context a Codex task produces is available to any teammate on the repo, so nobody re-explains the architecture from scratch.