Give Codex a persistent memory
Swylink gives OpenAI Codex intelligent context from your full AI stack. Structured metadata and semantic search keep Codex aligned with every tool.
Connect OpenAI Codex in 4 steps
- 01Installnpx swylink@latest init
- 02Authenticatenpx swylink auth
- 03Connectnpx swylink connect
- 04CodeOpen OpenAI Codex & go
Where Swylink writes the OpenAI Codex config
~/.codex/config.tomlRunning npx swylink connect automatically detects OpenAI Codex and writes the MCP bridge configuration to this path. No manual editing required.
Why OpenAI Codex needs persistent context across tools
OpenAI Codex is a terminal-based, autonomous coding agent that excels at executing complex tasks independently. You give Codex a high-level instruction, and it plans and implements the solution — creating files, writing tests, refactoring code — all without manual intervention. This autonomous nature is its greatest strength, but it also means that decisions made during Codex sessions are often ephemeral.
When Codex autonomously decides to use a specific design pattern, chooses one dependency over another, or restructures a module for better testability, those decisions live only in the terminal output. The next time you open Cursor or Claude Code, those tools have no knowledge of what Codex decided or why. You end up re-explaining architecture choices or, worse, your other AI tools make contradictory decisions.
Swylink captures the decisions Codex makes during autonomous runs and persists them as structured, searchable context. Dependency choices with rationale, architecture patterns selected, implementation strategies, and file organization decisions — all captured with semantic embeddings. When you switch to another tool, a simple search like "why did Codex choose this ORM" returns the exact decision with full reasoning. Codex's autonomous power is preserved while its knowledge becomes shared across your entire AI stack.
Context from Codex autonomous sessions made searchable
Codex's autonomous coding generates unique context that Swylink captures: dependency selection decisions with evaluated alternatives, architecture patterns chosen during autonomous implementation, test strategy decisions including which testing frameworks and coverage approaches were selected, file and module organization rationale, error handling strategies established during implementation, and build configuration choices. This context becomes instantly searchable across all your AI tools — Cursor, Claude Code, Windsurf, and Copilot all benefit from Codex's autonomous decisions.
Frequently asked questions about Swylink and OpenAI Codex
Does Swylink work with Codex's autonomous mode?
Yes. Codex supports MCP, and Swylink provides save_context and search_context tools that Codex calls during autonomous execution. Decisions made during unattended runs are captured automatically, so the context is available even for sessions you did not monitor in real time.
Where is the MCP config for OpenAI Codex?
Codex reads MCP configuration from ~/.codex/config.toml. Running npx swylink connect detects Codex and writes the correct Swylink server configuration to this file. The TOML format differs from the JSON used by other IDEs, but the connect command handles this automatically.
Can Swylink help when Codex and Claude Code make conflicting decisions?
Yes. Because both tools share the same Swylink workspace, you can search for prior decisions before starting a new session. If Codex decided on a specific architecture pattern, Claude Code can find that decision via semantic search and build on it rather than making a contradictory choice. Swylink acts as the single source of truth for cross-tool decisions.