Codex ecosystem dominates: codex-chatgpt-web leads at 237 stars/day
We don't tell you what's popular — popularity lags and is gameable. We tell you what's gaining speed right now, and whether it's worth your attention. 6 accelerating AI repos earned today's slot.
Top mover
This is a local MCP/Responses-API shim that drives the ChatGPT web UI with Playwright, exposing it to Codex as a native model so prompts, tool calls, streaming, and images route through your ChatGPT Pro subscription instead of Codex quota. It's gaining traction because Codex rate limits are the binding constraint for heavy users, and this sidesteps them by reusing an existing paid seat. The obvious trade-off: it depends on browser automation against a UI OpenAI can change or block at any time, and it likely violates ToS, so it's fragile and account-risky for production use.
🤖 Agents & automation
OpenExecutive wraps 8 specialist agents behind a single coherent 'executive' persona, routing tasks across a FastAPI backend and Next.js frontend with RAG over your own docs. It's gaining traction because it packages the multi-agent pattern into a deployable product shape rather than a notebook demo, and the Claude/Anthropic stack signals real LLM orchestration. The caveat: 'coherent persona' over 8 agents is hard to keep consistent, and the repo's actual routing/state logic is the thing to audit before trusting it with real decisions.
Tooling & infra
This is an Electron app plus Chrome extension that exposes local MCP servers as tools to ChatGPT, adding Goal/Compact/Resume session management and durable multi-agent workflows across macOS, Windows, and Linux. It matters because ChatGPT's native tooling is cloud-bound, so a local-first MCP bridge unlocks filesystem, shell, and custom tool access without routing through a hosted backend. The caveat: it depends on ChatGPT's web UI and Chrome extension APIs, so it can break on upstream changes and carries real local-execution security surface.
A 5MB Tauri wrapper that bundles the DeepSeek Harness CLI with preset plugins and zero environment setup across Windows/macOS/Linux. It matters because it removes the Python/Node dependency friction that keeps most agent harnesses out of non-developer hands, and the star velocity reflects demand for one-click local agent tooling. Caveat: it is a packaging/distribution layer, not a new harness — you inherit upstream DeepSeek Harness limitations, plugin quality, and update lag.
codex-router is a LiteLLM-based proxy that lets OpenAI's Codex CLI talk to external models like Kimi (via guided OAuth/API setup) and DeepSeek, with migration and rollback tooling for config changes. It's gaining traction because Codex users want cheaper or region-accessible models without abandoning their existing CLI workflow. The caveat: it depends on third-party providers' API stability and Codex's own config surface, so breakage is likely as either side changes.
Also accelerating
This repo bundles adversarial 'jailbreak' prompts and a test harness aimed at bypassing safety/instruction constraints in GPT-series models, specifically targeting Codex-style code generation. It is gaining traction because developers want to probe model guardrails, benchmark refusal behavior, and understand prompt-injection surfaces for red-teaming and robustness testing. The caveat: it is a prompt collection, not a library or tool, so its value decays as OpenAI patches the underlying model behavior, and using it may violate provider terms of service.