tastesteak/codeseex

CodeSeeX is a small desktop app that sits between OpenAI's Codex client and the DeepSeek API so a Codex user can run DeepSeek V4 models inside the Codex workflow. It runs a local server on port 8787 by default that mimics the OpenAI Responses API. Codex talks to that local endpoint, and CodeSeeX forwards the request upstream to DeepSeek. The README is explicit that the project is unofficial and not affiliated with OpenAI or DeepSeek, and that the user must follow each provider's terms. Two DeepSeek V4 models are exposed: deepseek-v4-pro and deepseek-v4-flash, both with a 1 million token context window declared in the model catalog. The bridge keeps Codex tool features working, including Apply Patch, MCP, Skills, and Plugins, and adds its own built-in tool layer: a Web Search tool, a workspace search, file reading, and patch support. There is also a desktop manager UI that shows service state, balance, usage estimates, daily logs, and tool activity. Setup involves downloading a release for Windows, macOS, or Linux, starting CodeSeeX, and copying the generated config.toml from the proxy settings page into the Codex configuration. The TOML defines a custom model_provider pointed at http://127.0.0.1:8787/v1 with wire_api set to responses. API credentials are read from the existing Codex auth setup rather than stored separately by CodeSeeX. Runtime files live under ~/.codeseex, including the model catalog JSON that Codex reads, daily JSONL log files, and optional community tool packages. Community tools are off by default and the README warns that enabling them runs local code from that tool package, so trust matters. The troubleshooting section covers balance query failures, missing DeepSeek models in Codex, and fetch-failed errors that are usually caused by a wrong base_url or a port collision. CodeSeeX is licensed under AGPL-3.0-only, which the README notes means anyone who modifies it and offers it as a network service must publish the modified source. The project is JavaScript and ships build scripts for all three desktop platforms; GitHub Actions can produce the desktop artifacts automatically.