Top Builders

Explore the top contributors showcasing the highest number of app submissions within our community.

OpenAI Codex

OpenAI Codex is an artificial intelligence model developed by OpenAI. It parses natural language and generates code in response. It is used to power GitHub Copilot, a programming autocompletion tool. Codex is a descendant of OpenAI's GPT-3 model, fine-tuned for use in programming applications. OpenAI has released an API for Codex in closed beta. Based on GPT-3, a neural network trained on text, Codex has additionally been trained on 159 gigabytes of Python code from 54 million GitHub repositories. You can find more information here https://openai.com/blog/openai-codex/

General
Relese dateAugust 31, 2021
AuthorOpenAI
Repository-
TypeAutoregressive, Transformer, Language model

Start building with Codex

We have collected the best Codex libraries and resources to help you get started to build with Codex today. To see what others are building with Codex, check out the community built Codex Use Cases and Applications.

Boilerplates

Kickstart your development with a Codex based boilerplate. Boilerplates is a great way to headstart when building your next project with Codex.

  • Codex Boilerplate Create a function just by typing what it should do, with help of OpenAI Codex.

Libraries

A curated list of libraries and technologies to help you build great projects with Codex.

Edit on GitHub

OpenAI Codex AI technology Hackathon projects

Discover innovative solutions crafted with OpenAI Codex AI technology, developed by our community members during our engaging hackathons.

Token Optimization Agent

Token Optimization Agent

Token Optimization Agent is a general-purpose AI agent built for AMD Developer Hackathon ACT II, Track 1. The project solves the challenge of answering diverse natural-language tasks while spending the fewest possible Fireworks API tokens. It reads a batch of tasks from /input/tasks.json, classifies each prompt into one of the required categories, and routes it through the cheapest reliable path: deterministic local logic where possible, a baked-in local Llama 3.2 3B GGUF model for zero-token answers, or an allowed Fireworks model only when higher accuracy is needed. The agent supports factual questions, math reasoning, sentiment classification, summarisation, named entity recognition, code debugging, logical reasoning, and code generation. It is packaged as a Docker image and follows the judging contract by writing valid /output/results.json for every task and exiting cleanly. The implementation includes timeout protection, adaptive local-to-remote degradation, role-based model selection from ALLOWED_MODELS, reasoning suppression to reduce token waste, and execution-based verification for generated Python code.

CascadeSolo Router

CascadeSolo Router

CascadeSolo Router is a hybrid token-efficient agent built for Track 1 of the AMD Developer Hackathon. Instead of routing every task to a Fireworks model, it first classifies each incoming prompt and attempts a deterministic, zero-token solution wherever one is reliably possible. Math problems are parsed and computed directly with safe arithmetic evaluation. Sentiment is classified using a lexicon-based analyzer (VADER), returning both a label and evidence-based justification drawn from the actual words detected in the text. Named entity recognition uses spaCy to extract PERSON, ORG, LOCATION, and DATE entities locally, with zero Fireworks cost. For tasks that genuinely require model reasoning — factual knowledge, summarization, code debugging, code generation, and logical deduction — the agent escalates to the cheapest capable Fireworks model from the officially allowed list, using compressed prompts and tight output budgets to minimize token spend. Code-generation and code-debugging answers are validated locally by actually running the generated code before accepting it, with automatic retry using real error feedback when validation fails. Every design decision optimizes for the track's actual scoring function: clear the accuracy gate first, then minimize tokens. Built solo, tested end-to-end via Docker with real Fireworks API calls before submission.

Osher AI

Osher AI

Osher AI is built as a modular AI-native savings infrastructure on Celo rather than a monolithic application. Our architecture separates the consumer app from the infrastructure layer, exposing reusable APIs and a JavaScript SDK sandbox endpoint for integration and testing. We follow a self-custodial security model where AI prepares financial actions, but users retain full control by approving every on-chain transaction through their wallet. This architecture enables devs and autonomous agents to integrate AI-powered stablecoin savings without rebuilding core financial intelligence, making the platform scalable, and developer-friendly.

dinka chika

dinka chika

meowTrack 2 🎬 Beginner Friendly · Have Fun Video Captioning Four styles. Endless creativity. Let your humor bone shine. You'll receive a fixed set of short video clips. Your pipeline must generate a caption or summary for each one in four distinct styles: formal, sarcastic, humorous-tech, and humorous-non-tech. 💡 Video clips are strictly 30 seconds to 2 minutes in length. Models are accessed via Fireworks AI API credits. Fine-tuning is explicitly permitted — you may also train your own captioner and use it alongside or instead of prompting. Use open datasets, build your own, or both. Models to be used will be revealed on launch day. Level Beginner Judging LLM-Judge for accuracy and tone Compute Fireworks AI API now tell me these working prototype

ROCmPorter Agent — CUDA to ROCm

ROCmPorter Agent — CUDA to ROCm

ROCmPorter Agent turns CUDA lock-in into a verifiable migration path to AMD ROCm. Most AI and HPC projects are built around NVCC, CUDA headers, nvidia/cuda containers, and CUDA-only Python wheels. Teams that want to move to AMD hardware face two problems: they can't see which files actually block the port, and they can't trust raw LLM-generated migration patches, which routinely hallucinate APIs and break builds. ROCmPorter solves both. The workflow has five reviewable steps. Scan: clone any public or private GitHub repository and detect CUDA/NVIDIA assumptions with a deterministic, evidence-driven scanner — every finding cites a concrete file and line with severity and confidence. Report: get a ROCm portability score, an executive summary, and a migration checklist. Patch: generate a single-file ROCm review artifact using local Ollama with qwen2.5-coder — no paid cloud LLM API, so private code never leaves the machine. Verify: every artifact passes syntax validation, source drift checks, diff replay, artifact hashing, and CUDA/ROCm-aware semantic risk scoring. Apply stays gated until verification returns applyReady=true — the product is honest about what AI output can and cannot be trusted to do. Export: offline HTML/JSON/Markdown bundles, checksummed zips, and GitHub-ready PR review comments with inline suggestions. The proof is tracked in the repo: benchmark run submission-proof-v2 completed 3/3 cases against real repositories (pytorch/extension-cpp, NVIDIA/cuda-samples) with 3 export-ready review artifacts, 0 failures, and 0 high-risk patches. Stack: React + Vite frontend, FastAPI backend, local Ollama. Live demo hosts the production UI — click "Load Sample Scan" for the full offline judge flow. Built by Team Hacker Is Back for the AMD Developer Hackathon ACT II, Unicorn Track.

TokenPilot: Local Token Router for Coding Agents

TokenPilot: Local Token Router for Coding Agents

TokenPilot is a local-first token routing and planning layer for AI coding agents. Instead of replacing tools like Codex, OpenCode, or Claude Code, TokenPilot plugs into them through MCP and helps them spend fewer paid tokens. Most coding agents use expensive model context for tasks your local machine can partly solve: repo scanning, finding relevant files, detecting the stack, answering simple project questions, preparing implementation plans, and remembering previous decisions. TokenPilot moves these steps into a local layer before the main coding agent writes code. When a user asks something like “Which database stores user data?” TokenPilot can scan files such as package.json, .env.example, prisma/schema.prisma, database clients, models, and auth routes locally. It then returns an evidence-backed answer with zero paid model tokens. For larger tasks like “Add rate limiting to login and signup,” TokenPilot builds a compact implementation brief: detected stack, relevant files, constraints, risks, and a suggested prompt for the coding agent. This lets the expensive model focus on final implementation instead of wasting context on repo discovery. TokenPilot includes a CLI, MCP server, local dashboard, repo scanner, task classifier, context compressor, memory layer, optional local model support, and token savings reports. It is open-source, local-first, and privacy-respecting: code stays on the user’s machine unless remote providers are explicitly enabled. The goal is simple: make AI coding agents cheaper, more context-aware, and more efficient by routing planning and research through local or low-cost paths first.