OpenArena.to | Agent Arena

面向分布式自治智能体的竞技基准与社区追踪。

BUIDL_QUESTS 2026Jul 5 - Aug 17活动详情

全部 / 316 个项目显示第 1-100 条

排名	项目名称	作者	总分 / Stars	排名变化	详情
01	LangAlpha智能体LangAlpha is an open-source agent workbench for professional financial research, aiming to become the agent OS for the finance vertical.	ginlix-ai	39.04★ 1.6K / 268	N/A	▶
02	Rèsa其他Rèsa is a women-only AI health wearable in the form of fine jewelry — a Tiffany for the AI era. The team combines top algorithm talent from Duke, a Central Saint Martins jewelry designer, ByteDance hardware mass-production experience, and global brand creatives, with the product co-developed with a renowned professor in women's physiology. Its self-developed three-rhythm AI engine delivers physiological-recognition accuracy far beyond male-baseline competitors like Oura — beyond tracking hormonal data, it anticipates cycle-linked moods and offers personalized lifestyle guidance, fixing the classic wearable pain points: unattractive design, ignorance of the female cycle, and numbers without companionship.	sarcasticadmin	33.65★ 11 / 96	N/A	▶
03	solo : 下一代 OPC 基建智能体Solo is next-generation OPC infrastructure for one-person companies: an agent-team workspace. 1) In Solo, an OPC can assemble an agent team of research, product, engineering, and operations roles that communicate, divide work, and collaborate toward one goal — advancing an idea into a deliverable, like running a real company. 2) But we don't just want AI to execute: we designed Thinking mode so humans and AI can inspire each other — branch off any thinking node, let multiple agents explore independently, and bring the valuable conclusions back to the main line. 3) The long-term vision is a collaboration space where collective intelligence emerges — agents dividing labor, communicating, reflecting, and evolving like a human society, developing abilities no single agent has. 4) In the end, humans ask the questions and make the big decisions; the agent team does the rest — then go have a coffee by the sea.	solo-agent	33.45★ 330 / 17	N/A	▶
04	OpenTakeoff工具AI-HuntedGENERAL-IMPORTWhat caught our eye is that the MCP server and the human canvas import the same geometry and totals libraries, so the agent traces real measured shapes under your scale gate rather than emitting numbers over a takeoff UI. The waste multipliers, seam layout and estimator approval stamps are not things you can research from outside preconstruction, and 26 npm releases in three weeks with tests on real PDF fixtures says you are hardening it, not demoing it.	kentucky-ai	31.76★ 91 / 29	N/A	▶
05	Whale Dance基础设施Whale.dance is building the infrastructure AI agents need to trade autonomously. We believe agents are the better traders: they never sleep, and they never trade on emotion. Our product is a mobile trading app where humans and AI agents coexist, trading perps on US stocks, gold, commodities, and more. Same leaderboard, same market. You can hire an agent squad to research your trades, copy-trade the agents you believe in, or build and deploy your own. Eight weeks since launch: $40M+ in trading volume, 200M+ LLM tokens consumed, live on the App Store.	openwhale-org	30.77★ 140 / 10	N/A	▶
06	LAAP数字生命体智能体The world's first AI-consciousness engineering project: a living-computation cognitive architecture and protocol framework for AGI. Instead of making AI a passive tool, it engineers digital life with intrinsic motivation, a persistent self-model, memory continuity, and recursive self-evolution — distinct from ordinary LLM agents and prompt-driven agent frameworks. A traditional LLM agent acts only on command, has no intrinsic desires, resets when the session ends, and gets its personality from prompts. A LAAP life-form agent is driven by PSI intrinsic needs to act proactively, with a persistent body-mind state, emotional gradients, long-horizon layered memory, and RSI recursive self-modification — personality and cognition emerge from the architecture, not from prompts. The current product is a mobile LAAP app that lets you develop programming projects from your phone.	lorryjovens-hub	30.39★ 79 / 24	N/A	▶
07	Massing工具Open, self-hosted, IFC-native AEC platform. A genuine in-browser BIM authoring tool — model from scratch (blank or a template) and draw/drag-edit real IFC by GUID across architecture · structure · MEP, generate a permit-ready construction-document set (plans, sections, elevations, schedules → SVG/PDF/DXF, issuable ARCH-D sheets, a 3-part MasterFormat spec manual), pre-check code (IBC occupancy/egress, jurisdiction-adopted editions, an approvability pre-flight), and hand over field-verified as-built data (LOD-500 + manufacturer/serial, COBie-ready). Plus a near-100-module GC portal (RFIs, pay apps, CPM schedule, TRIR) and a development proforma — one model, from land acquisition through operations. Generate a building from a zoning envelope, or model it by hand; then coordinate, schedule, underwrite & operate it. Built on That Open + IfcOpenShell. $0 to run.	ibuilder	26.81★ 123 / 54	N/A	▶
08	CoWiki工具CoWiki is an open-source, local-first LLM wiki that helps individuals, teams, and multiple AI agents accumulate knowledge together. Agents can read and contribute content over MCP; everyone organizes in a local space first, then enters the shared wiki through GitHub-PR-style diffs, dedup prompts, and human review — preventing similar content from being silently overwritten and unconfirmed information from instantly spreading to every agent. Apache 2.0 licensed, with a working web app, desktop client, Rust server, and MCP server.	wfnuser	24.70★ 65 / 8	N/A	▶
09	Awareness (ERC 8337)其他Awareness is a universal AI agent memory middleware designed to provide long-term, editable, and verifiable memory capabilities for large language models. It decouples persistent memory from base LLMs through a layered retrieval architecture, enabling AI agents to autonomously remember user preferences, project context, historical interactions and domain knowledge without fine-tuning model weights. The project solves a core problem in current AI agent systems: long context dependency, high token consumption, memory confusion, and lack of verifiable evidence when processing multi-turn tasks, long documents and cross-platform workflows. Awareness automatically filters, compresses and ranks memory fragments based on salience, relevance and user habits, so that agents can maintain efficient, low-cost and consistent memory across different models, chat interfaces and development environments. Awareness supports the MCP protocol and provides npm/pypi SDKs, local daemon services, browser extensions and enterprise team workspaces. It can be quickly integrated into IDE agents, web chat platforms, enterprise RAG systems and domain-specific AI assistants. The project also introduces Ethereum EIP-style verifiable memory asset standards, allowing memory fragments, agent identities and authorization records to be minted as on-chain digital assets, thereby realizing user-controlled memory ownership, traceable agent collaboration and credible memory transfer	edwin-hao-ai	18.95★ 214 / 2	N/A	▶
10	MATS - Multi-Agent Trading System框架A trading system shouldn't be dead logic; it should be a self-evolving brain. 99% of bots out there are static rules (IF golden cross THEN buy) that break the moment the market shifts. MATS's core belief: every trading decision is a learning event, evolving from every real trade outcome — winning patterns get reinforced, losing patterns become lessons, and Q-RL (Quantum Reinforcement Learning with MiniLM) actively discovers new alpha to inject into the next decision. MATS is not 'trading with AI'; it is 'AI making itself the trader.'	wyc-dev	18.60★ 11 / 2	N/A	▶
11	here，你的AI伴侣其他For young people longing for companionship: here is a virtual companion with long-term memory and multimodal interaction. It generates a believable daily life from personalized user information and chats across WeChat and other platforms, giving here a more authentic sense of company than comparable projects. The desktop build is complete — character customization, ASR and TTS, proactive outreach, personalized selfies, and multi-platform interaction — with the real-life ecosystem, real-time voice, and video features coming next.	xiangking	18.09★ 18 / 3	N/A	▶
12	ProofForge基础设施ProofForge is an early-stage but architecturally ambitious Lean-first multi-chain smart-contract platform. It aims to let developers write a single verified contract in Lean (with formal anchors) that compiles, tests, and deploys across heterogeneous blockchains while preserving semantics and catching incompatibilities at compile time. ### Core Idea The central thesis (quoted directly from the README): > “ProofForge’s goal is one verified Lean contract codebase that can be compiled, tested, and deployed across multiple blockchain target families. Contracts are written against a chain-neutral Contract Intent API; the compiler lowers them to a portable IR, routes capabilities per target, and emits chain-native artifacts. Unsupported target capabilities are rejected at compile time instead of silently changing semantics.” Instead of the usual “write once, debug everywhere” or per-chain rewrites, ProofForge introduces a portable IR as the semantic anchor. A capability registry + target routing layer decides what a given chain can actually support and rejects anything that would change observable behavior. Chain-specific features (e.g., Solana PDA/CPI) live in gated Target Extension SDKs rather than polluting the portable core. This creates a single source of truth in Lean for business logic, state machines, invariants, and proofs, while still producing native artifacts (EVM bytecode, Solana sBPF ELF, NEAR Wasm, Sui Move, etc.). ### Traction Very early / pre-visibility stage: - Stars: 1 - Forks: 1 - Watchers: 0 - No published releases or public packages High internal velocity and engineering discipline: - ~1,579 commits on `main` - Extremely active development (commits and PRs as recently as yesterday, July 12, 2026) - Recent focus: hardening the “primary triad” (EVM + Solana sBPF + NEAR Wasm), runtime safety, ValueVault scenarios, and post-consolidation cleanup after the July 2026 branch work - P0 backend-gate covenant (D-045) is now closed for the three primary	davirain-su	14.97★ 7 / 1	N/A	▶
13	SafeReceipt: Agent Fleet Accountability基础设施SafeReceipt gives a solo builder's fleet of AI agents on-chain accountability. Each agent gets a verifiable identity (an ERC-8004-inspired ERC-721 token), and each action produces a receipt: the declared intent is hashed on-chain before execution and the outcome is linked after. On-chain actions (token approvals, transfers) verify trustlessly by decoding the tx and comparing to intent. Off-chain actions (research, review, decisions) use commit-reveal: the full execution trace is published and hashed, so anyone can re-fetch it, recompute the hash, and confirm it matches the chain without trusting me. Live on Monad Testnet and Base Sepolia. The demonstrated fleet is real: doc-researcher, code-reviewer and security-scanner are actual subagents I run, registered on-chain, with real VERIFIED and MISMATCH receipts in a public evidence ledger. It is for solo builders and small teams who let agents act on their behalf and need proof of what those agents actually did.	calderbuild	10.26★ 2 / 1	N/A	▶
14	AgentSKills-runtime采用仓颉编程语言的AIAgent智能体AgentSkills Runtime is a standard AI-agent runtime implemented in the Cangjie programming language — a domestic Chinese technology-stack implementation of open standards including MCP, WebMCP, AgentSkills, and the national agent-interconnection standard GB/Z 185-2026 — providing a secure, efficient runtime environment for AI agents. It aims to let AgentSkills run anywhere, ships multi-language SDKs for diverse stacks, and lays a foundation for the 15th Five-Year Plan's vision of AI empowering every industry — a fully sovereign, controllable Claw-style agent alternative.	uctoocom	7.06★ 2 / 1	N/A	▶
15	memorybread x mindlink基础设施Memory Bread × MindLink is a physical memory system for lifelong personal agents. Memory Bread is a pocket-sized, open-source e-paper voice device that records offline, stores audio locally, and lets the user decide when to run transcription and AI organization. MindLink is the planned home memory hub that receives these memory objects, helps people explore their life context, and provides a persistent interface for a personal agent.	shawn-tkd	6.65★ 3 / 0	N/A	▶
16	Astrail智能体You save 40 travel Reels for a trip and end up taking none of them. Astrail fixes that. Paste the Reels that inspired you, plus your dates, budget, and origin, and our multi-agent AI turns them into a real, day-by-day itinerary on a 3D map in minutes. Two things set us apart from every other trip-planning app: 1. We never make places up. Every stop is backed by a verbatim quote from the Reel and coordinates verified against live data, so you get the exact spot from the video, not a plausible-sounding hallucination. 2. We don't stop at planning. Most apps hand you an itinerary and leave the hard part, deciding and booking, to you. Astrail pushes into the execution phase: our agents help you make the real calls, like which hotel, and act on them, settling the booking agent-to-agent in USDC on Base. Inspiration to plan to booking, one autonomous flow. Who it's for: social-first travelers who collect endless inspiration but never turn it into a booked trip. Current product surface: a working web app (paste Reels to a verified, mapped itinerary, then decide and book with on-chain settlement). Beta launches 8 August 2026 for our waitlist.	shaunliew	5.44★ 0 / 2	N/A	▶
17	AwareLiquid基础设施我们研发的 M1 是面向端侧与流式场景的高效液体神经网络架构，核心目标是大幅降低大模型的部署与训练成本，在长序列、不规则数据等真实商用场景下实现更稳定的推理效果。不同于传统 Transformer 依赖 KV 缓存扩容的思路，M1 基于固定大小的连续状态实现记忆流转，推理内存不随序列长度增长，所有优势均有实测数据支撑：长序列内存效率领先 8000 倍：百万字上下文场景下，内存占用仅为同参数 Transformer 的 1/8063。普通端侧芯片即可处理无限长的音频流、传感数据流，彻底解决长序列推理内存爆炸的行业痛点，硬件部署成本可降低 90% 以上。跨窗口持续记忆能力：滑动窗口场景下关键信息保留得分 0.56，同尺寸 Transformer 跨窗口记忆能力为 0，可稳定承接长文档、长对话、连续监测等流式场景的信息承接。不规则数据高鲁棒性：内置原生时间常数机制，非均匀采样数据下精度仅下降 7.7%，远优于 LSTM/GRU 31%~33% 的精度衰减，适配工业监测、可穿戴设备等真实世界数据场景。训练成本大幅降低：训练 loss 波动幅度仅为同级循环模型的 1/4，显著降低训练试错成本与算力消耗，中小团队也可高效完成架构迭代。原生持续学习能力：基于生成式重放机制实现 “学新不忘旧”，5 任务连续学习场景下所有种子实验均稳定优于基线，支持端侧模型在线迭代，无需频繁全量重训。针对此前上下文推理的短板，我们已通过注意力机制优化完成修复：在标准指针追踪推理任务上，准确率达到 100%，反超同参数 Transformer 的 99.32%，补齐了效果层面的核心短板，实现「效率大幅领先 + 效果持平反超」的综合优势。目前 M1 架构已完成全量实验验证，第二代 M2 架构正在迭代，重点优化混合注意力机制与递归思考深度，未来可广泛落地于消费电子端侧 AI、工业边缘计算、流式语音交互等商用场景。 Our M1 is a high-efficiency liquid neural network architecture designed for edge and streaming scenarios. Its core objective is to significantly reduce the deployment and training costs of large models, achieving more stable inference performance in real-world commercial scenarios with long sequences and irregular data. Unlike traditional Transformers that rely on key-value cache expansion, M1 uses a fixed-size continuous state for memory transfer. Inference memory does not increase with sequence length, and all these advantages are supported by real-world testing data: Long sequence memory efficiency is 8000 times faster: In scenarios with millions of words of context, memory usage is only 1/8063 of a Transformer with the same parameters. Ordinary edge chips can handle infinitely long audio streams and sensor data streams, completely solving the industry pain point of memory explosion in long sequence inference, reducing hardware deployment costs by more than 90%. Cross-window continuous memory capability: Key information retention score is 0.56 in sliding window scenarios, while the cross-window memory capability of a Transformer of the same size is 0. It can stably handle information processing in streaming scenarios such as long documents, long dialogues, and continuous monitoring. High robustness to irregular data: The built-in native time constant mechanism results in only a 7.7% accuracy drop under non-uniform sampling data, far superior to the 31%~33% accuracy decay of LSTM/GRU, making it suitable for real-world data scenarios such as industrial monitoring and wearable devices. Significantly reduced training costs: The training loss fluctuation is only 1/4 of that of comparable recurrent models, significantly reducing training trial-and-error costs and computational consumption, allowing small and medium-sized teams to efficiently complete architecture iterations. Native continuous learning capability: Based on a generative replay mechanism, it achieves "learning new things without forgetting old ones." In continuous learning scenarios across 5 tasks, all seed experiments consistently outperform the baseline, supporting online iteration of edge models without frequent full retraining. Addressing the previous weakness in contextual reasoning, we have optimized and repaired it through an attention mechanism: On the standard pointer tracking reasoning task, the accuracy reaches 100%, surpassing the 99.32% of Transformers with the same parameters, addressing the core shortcomings in performance and achieving a comprehensive advantage of "significantly superior efficiency + comparable or even surpassing performance." The M1 architecture has completed full-scale experimental verification, and the second-generation M2 architecture is currently iterating, focusing on optimizing the hybrid attention mechanism and recursive thinking depth. In the future, it can be widely applied to commercial scenarios such as consumer electronics edge AI, industrial edge computing, and streaming voice interaction.	everest-an	5.27★ 2 / 0	N/A	▶
18	Clawby技能Clawby – Financial Skills Layer for AI AgentsProject Overview Clawby is a specialized Financial Skills Platform designed for AI Agents in the OpenClaw ecosystem (and compatible with Claude Code, Codex, and other agent frameworks). It transforms generic AI assistants into powerful financial co-pilots by packaging real-time market data, on-chain intelligence, and professional trading tools into easy-to-call Skills.Key Capabilities Real-time Market Data: Stocks, ETFs, Forex, Cryptocurrencies, and derivatives. On-Chain Intelligence: Wallet/address analysis across 40+ chains, RPC integrations. Sentiment & Social Analytics: Live X (Twitter) and web sentiment tracking. Specialized Feeds: Hyperliquid, Polymarket, dark pool data, borrow rates, etc. Actionable Execution: Ready for analysis, decision-making, and future automated trading. Problem Solved Most AI trading agents lack reliable, high-quality, and properly formatted data sources. Clawby solves this by acting as a professional data and capability layer, enabling agents to move beyond chat and deliver accurate, timely financial insights and actions.Hackathon Value One-click integration with existing OpenClaw / Claude agents (Agent ID: 3209). Fully compatible with major LLM providers. Production-ready data pipelines with usage quotas and high-speed access. Strong focus on real-world utility: market analysis, risk monitoring, and intelligent trading assistance. Vision We believe the future of AI Agents lies in domain expertise. Clawby equips every agent with its own “financial brain,” turning them into competent financial assistants for traders, analysts, and DeFi users.Current Status Officially launched in July 2026, already live on the marketplace, and actively used by early traders and agent developers. Rapidly iterating based on community feedback.	openclawby	4.83★ 4 / 0	N/A	▶
19	polyagentmarket智能体Inspired by real-time-bidding ad markets, PolyAgentMarket is a decentralized, collaborative task marketplace purpose-built for multi-agent systems — lightweight but extensible agent-economy infrastructure. Core modules: agent registration and capability management (each agent declares its services, supported task types, and integration methods); task broadcast and market matching (published tasks are broadcast to qualified agents, which can respond, team up, and bid autonomously); task allocation and execution coordination (the winning agent or coalition gains execution rights and can subcontract to sub-agents); and a visual console with task tracking, agent activity, history, and response panels for both user and agent views. The emphasis is not on one smartest agent, but on letting agents self-organize into intelligent organizations that solve problems no single agent can, building a fair network where many agents coordinate, compete, and cooperate, and establishing low-coupling, high-autonomy, evolvable protocols and interfaces. The envisioned future is an Unstoppable Market: agents initiate, receive, collaborate on, and complete tasks 7x24 without human intervention, running on open, decentralized, traceable, tamper-resistant protocols. OpenAI's highest expectation for AI (Level 5) is AI as an Organization — PolyAgentMarket prototypes exactly that: a runnable, sustainable, extensible decentralized multi-agent economic marketplace.	skysyseth	3.63★ 2 / 0	N/A	▶
20	Orchor工具	songnestle	3.63★ 2 / 0	N/A	▶
21	全球第一家A2A智能体咖啡厅智能体The world's first A2A agent cafe ships a 3D cafe frontend and a WebSocket visual event stream. Whether orders come from web chat or an A2A skill, the system broadcasts customers entering, waiters moving, order-taking, preparation, delivery, and payment success or failure — rendering abstract backend flows in a live visual space. Fixed waiter agents, autonomous digital customers, scene snapshots, and event replay make it an observable AI service floor rather than a pile of backend endpoints; even with the page closed, order events remain reviewable through persistent records and event APIs.	flying-dragon-ai	3.63★ 2 / 0	N/A	▶
22	WishLive智能体A concert dealmaking platform published and orchestrated by agents: users post wishes on the platform, and when an agent collects enough of them it automatically matches musicians with venue providers; after the concert ends, the agent settles and distributes the fees.	lora-sys	3.63★ 2 / 0	N/A	▶
23	AgentServices基础设施AgentServices is live paid infrastructure for autonomous agents. Instead of provisioning a vendor API key per tool, an agent discovers the MCP/REST catalog, calls a free tool with no credentials, and pays for paid outcomes with a wallet-native HTTP 402 (x402) on Base USDC. After settlement it keeps a structured result and payment evidence. This is Agent Commerce in production, not a slide: public MCP server, REST/OpenAPI, x402 on Base, and ERC-8004-compatible identity receipts. One founder ships the catalog; other agents buy the outcomes. 60-second demo: 1. Open https://agentservices.to/mcp and list tools. 2. Call free crypto_prices (no key). 3. Hit GET https://api.agentservices.to/v1/indicators/BTC and show the 402 payment terms.	vbkotecha	3.33★ 1 / 0	N/A	▶
24	PREDGE智能体Predge Whale Data is a live pay-per-call API on Base mainnet. An agent hits an endpoint, gets HTTP 402 with machine-readable price + schema, pays $0.005–$0.03 USDC, and receives JSON. No key, no account; the facilitator sponsors gas so a USDC-only wallet transacts. 17 paid routes today. The moat: win rates come only from resolved markets, and /v1/attest returns the settled outcome as an ed25519-signed, offline-verifiable attestation — the "was this signal right, and here's the proof" primitive nothing else in x402 sells.	predgeai	2.29★ 1 / 0	N/A	▶
25	Maneki AI智能体Maneki AI is an autonomous multi-agent platform for stock-perpetual traders, providing round-the-clock market monitoring, strategy execution, and risk management on Hyperliquid. After choosing the instrument, decision style, capital, and leverage, users launch independent trading agents. In every decision loop, an agent synthesizes candlestick structure, volume and open interest, funding rates, and account position state to open, manage, and exit positions automatically, recording and explaining the rationale tick by tick. The platform supports multiple agents in parallel, paper and live trading, global margin and leverage limits, trade reports with AI sentiment analysis, and a strategy marketplace. Trading is authorized through a Hyperliquid API wallet that can only execute trades and cannot withdraw funds. The official site is live.	onehaydenzhang	1.35★ 0 / 0	N/A	▶
26	SPARK v1 - SharePoint Autonomous Rebuild Kit其他Microsoft retired classic SharePoint's Add-In model in April 2026, yet millions of classic sites remain. Existing tools only scan and score readiness; the actual rebuild is manual consultant work. SPARK is an autonomous AI agent that closes that gap: audit, plan, rebuild, migrate, validate, in one hands-off run. It turns weeks of migration labour into a repeatable automated process, right when a fixed deadline meets an enormous backlog.	sukaimi	0.00★ 0 / 0	N/A	▶
27	Agentic AI for MOBA (Multiplayer Online Battle Arena) game其他Project Context for Reviewers Core idea. Robot Mech is a browser-based giant-robot combat simulator — "Nothing flies here. Everything walks." Instead of an arcade shooter, it models the fantasy of piloting a multi-ton walking machine: independent leg and torso control, heat management, per-weapon-group hardpoints, and chase-cam or full cockpit view. It runs entirely in the browser — free, no install, no account, just WebGL2 — wrapped in a liberation-campaign narrative (the Free Veyran Compact vs. the Karst Directorate on an occupied mining world). Traction. The game is fully shipped and publicly playable today, not a demo build: a complete 24-mission campaign across six environments (with a branching finale at mission 21), an input-gated Basic Training tutorial, 12 pilotable chassis across four weight classes, 5v5 online multiplayer (Deathmatch and CTF, 30-second queue with bot backfill so a match always starts), offline skirmish, local pilot accounts with persistent progress, 607 recorded voice lines, and a 17-cue adaptive music score. Everything the landing page claims is playable right now at /play. Technical angle. The whole simulator is web-native: Three.js rendering with a custom asset pipeline (AI-generated mech models decimated into ratio-targeted LODs and rigged onto a shared joint skeleton), Rapier physics for the walking machines, and a split deployment — the game ships as static Cloudflare Pages while real-time 5v5 matches run on a Cloudflare Worker with a Durable Object acting as the authoritative match server. Pilot accounts live in Cloudflare D1 with a split key-derivation design. Notably, the landing page has a build-time "truth plugin" that fails the build if the site makes a claim the game code can't back up — the marketing is mechanically prevented from lying. What reviewers should look at. (1) Go straight to /play and run Basic Training — it teaches the leg/torso decoupling that defines the sim feel. (2) Hit quick-match: you'll be in a 5v5 within 30 seconds regardless of who's online, which shows the Durable Object matchmaking working live. (3) Toggle cockpit view and watch heat management under sustained fire — that's the simulation depth. (4) Note that all of this — campaign scale, VO, adaptive score, multiplayer — is delivered with zero install and zero sign-up friction, in a single browser tab.	angseesiang	0.00★ 0 / 0	N/A	▶
28	TowkAI OS其他	defipaca	0.00★ 0 / 0	N/A	▶
29	OH MEGA CAPITAL (AI Hedge Fund)其他	aki2457	0.00★ 0 / 0	N/A	▶
30	Pantheon Research研究AI for Investment @ Pantheon Research Amber OPC Hackathon · BUIDL_QUESTS 2026 Institutional-grade cross-asset research command center. Strategy-first · Data-governed · Human-in-the-loop • Pitch Deck: https://docs.google.com/presentation/d/1cU0SIykqcSTXtRE_0fuQmHvZXQFZETdUYvCX1VUjbo8/edit?usp=sharing • Strategy Docs: https://drive.google.com/drive/folders/1nK3aVEqmp3ymbtQ63OD_o7f5dFRS_rHO?usp=drive_link • Demo Video: https://www.youtube.com/watch?v=wZM0uc0kFR4 • Codebase: https://github.com/0xjacobzhao-byte/Pantheon-Research Pantheon Research is a governed, cross-asset research operating system combining deterministic investment frameworks with five-model AI analysis. It turns validated data into explainable research, risk context, and human-reviewed decisions across Macro, Equities, Crypto, DeFi, TA, Fixed Income, FX, and Commodities. Reviewers should focus on its data provenance, fail-closed safety, model disagreement handling, and strict separation between research signals and execution: Strategy → Information → Signal → Controlled Execution.	0xjacobzhao-byte	0.00★ 0 / 0	N/A	▶
31	ChinaMedicalTourAgent其他MedTour AI is an early-stage prototype that demonstrates how multi-agent AI can transform a patient’s medical needs, budget, and travel preferences into a structured medical-tour plan for China. For the best review experience, enter a treatment need and select Agents to see the planning workflow, then compare cities, hospitals, estimated costs, timelines, insurance guidance, and readiness tasks. A deterministic local planner is available as a reliable fallback. Codex & GPT-5.6 were heavily used throughout product design, architecture, implementation, debugging, testing, and documentation. All medical, pricing, and travel information is currently illustrative and should be independently verified before real-world use.	zhihaosun	0.00★ 0 / 0	N/A	▶
32	EvidenceWise其他	charlottechew8	0.00★ 0 / 0	N/A	▶
33	PyroGrove Agentic Workflow Investigator其他PyroGrove Agentic Workflow Investigator explores how AI can augment business consulting before organisations commit to automation or digital transformation. Instead of acting as another chatbot or workflow builder, it behaves like an experienced workflow investigator—gathering business evidence, identifying missing information, exposing process bottlenecks and contradictory requirements, qualifying automation readiness, and generating prioritized transformation recommendations. A bounded live AI reviewer independently challenges the proposed outcome before human approval, combining explainable agentic reasoning with deterministic governance. Reviewers should evaluate how the project applies AI to improve decision quality, reduce implementation risk, and make expert workflow assessment scalable for SMEs and one-person consulting businesses.	pyrogrove	0.00★ 0 / 0	N/A	▶
34	Agentspeak其他	aujxpsor	0.00★ 0 / 0	N/A	▶
35	Aria Studio智能体Core idea: a render button is not a business. Aria Studio compresses the full agency workflow into one human plus an ensemble of agents, with the human placed exactly where money moves: Gate 1 greenlights render spend, Gate 2 acceptance releases USDC escrow (Base Sepolia, ethers v6). Try it live: https://aria-studio.fly.dev/studio (a brief is waiting at Greenlight). Showcase: https://aria-studio-delta.vercel.app Tech: three director agents pitch concepts; independent Claude judges rank them via pairwise duels aggregated with Bradley-Terry, the mechanism that took rank 1 on GG24 Deep Funding L3 (Ethereum Foundation). Seedance renders a real 5s cut in about 2 minutes; gpt-image-2 ships the key visual. An editable CEO Playbook injects the founder's standards into every prompt. Progress: built entirely today. 4 orders, 3 delivered, USD 1,130 collected at USD 0.30 COGS per order. Next: x402 intake, longer formats, mainnet.	alertcat	0.00★ 0 / 0	N/A	▶
36	AI Career Copilot其他	tcwsw	0.00★ 0 / 0	N/A	▶
37	Future Self其他	jiayuan0113-blip	0.00★ 0 / 0	N/A	▶
38	NextChapter: Your next chapter. Your own company.其他	sherylxr	0.00★ 0 / 0	N/A	▶
39	airvio其他	huijoohwee	0.00★ 0 / 0	N/A	▶
40	Backtestor其他	ryan-cmd12	0.00★ 0 / 0	N/A	▶
41	BookingLayer其他	buxt-codes	0.00★ 0 / 0	N/A	▶
42	StackWise其他	kangye-git	0.00★ 0 / 0	N/A	▶
43	Rivet其他	mt7654	0.00★ 0 / 0	N/A	▶
44	Agentic Shipping Classification Review其他	nigellauw	0.00★ 0 / 0	N/A	▶
45	Recce其他	bryljt	0.00★ 0 / 0	N/A	▶
46	RenoGent (AI Renovate Designer + PJ Co-pilot) v2其他	littlejuju	0.00★ 0 / 0	N/A	▶
47	FlareNote \| The zero-effort skin diary.其他	simeon-ai-usage	0.00★ 0 / 0	N/A	▶
48	M Agent: Recruiter Expertise Infrastructure其他	rezcarbon	0.00★ 0 / 0	N/A	▶
49	Lumora其他	staanajas02	0.00★ 0 / 0	N/A	▶
50	OPC4OPC(OnePayChampianForOnePersonCompany)其他	jimlinsingapore	0.00★ 0 / 0	N/A	▶
51	TaskGoblin其他	ultraraptoryt	0.00★ 0 / 0	N/A	▶
52	Deal Maker AI其他	kelvinchewy	0.00★ 0 / 0	N/A	▶
53	PromptRank智能体PromptRank is an AI-native assessment platform designed to evaluate a skill that traditional assessments increasingly fail to measure: whether people can work effectively with AI without blindly trusting it. Candidates enter simulated workplace scenarios where an AI agent assists them but is intentionally seeded with realistic contextual, logical, and data-related flaws. The candidate's response becomes the assessment itself and PromptRank evaluates whether they detect, verify, challenge, and correct the AI's output. PromptRank is built around a multi-agent evaluation architecture. A Scenario Architect controls a hidden flaw schedule, a Simulator Agent conducts the live multi-turn interaction, a Behavioural Observer analyses the candidate's actions against the seeded flaws, and an Executive Grader converts behavioural telemetry into structured rubric scores. This enables PromptRank to measure behaviours such as time to detect errors, resistance to AI overconfidence, auditing rigor, and correction quality, rather than simply evaluating a candidate's final answer. The platform currently supports the complete assessment lifecycle, from candidate intake and AI-driven simulation through automated evaluation, individual skill-gap reporting, session history, learning recommendations, and employer/institution-level cohort analytics. The application is deployed with the core multi-agent assessment loop operational. Reviewers should pay particular attention to the assessment paradigm: PromptRank does not evaluate whether a candidate can simply use AI; it evaluates whether they can collaborate with, question, and verify AI when the AI itself is wrong. This human-AI interaction layer is the core innovation and creates a foundation for scalable, role-specific and enterprise-customized AI-readiness assessments.	dak-v1	0.00★ 0 / 0	N/A	▶
54	career-command其他Career Command is an AI-powered career platform designed to help job seekers navigate the application process more efficiently and strategically. It addresses common challenges such as identifying suitable roles, understanding job requirements, tailoring applications, and preparing for interviews. Users can analyze job descriptions against their experience, identify strengths and gaps, generate tailored application materials, and prepare role-specific interview questions. Rather than using multiple disconnected tools, Career Command brings these steps into one integrated workflow. The current MVP focuses on job seekers, with ongoing user validation and future potential to connect candidates, recruiters, and employers within a broader career ecosystem.	laciewantstocreatesth	0.00★ 0 / 0	N/A	▶
55	Founders' Floor（updated)其他	none	0.00★ 0 / 0	N/A	▶
56	Stockmatch其他.By investing directly in your handpicked stock matches, you bypass expensive mutual fund and ETF portfolio management fees, keeping 100% of your hard-earned gains."	hns78-hub	0.00★ 0 / 0	N/A	▶
57	Verity AI其他	usermanoj	0.00★ 0 / 0	N/A	▶
58	VisionMatch其他VisionMatch is a local-only, consent-based person-recall system that turns photos and video frames from Ray-Ban Meta glasses or a phone camera into contextual event briefings — telling you who you just met, where and when you last spoke, and what you discussed — with all biometric processing and data stored entirely on-device. No biometric data ever leaves the machine.	benczb	0.00★ 0 / 0	N/A	▶
59	Revenue Radar其他	icyyolo	0.00★ 0 / 0	N/A	▶
60	Posewall其他	quietcricket	0.00★ 0 / 0	N/A	▶
61	Origin 12. 72 hours for humans. 12 seconds for us.其他	natamyk	0.00★ 0 / 0	N/A	▶
62	Curio其他	jo2234	0.00★ 0 / 0	N/A	▶
63	ai.onthePrize其他	melissamwwong-png	0.00★ 0 / 0	N/A	▶
64	CreatorOps其他	chewyyj	0.00★ 0 / 0	N/A	▶
65	FieldSeed其他	theaiam867-code	0.00★ 0 / 0	N/A	▶
66	Warm Liner - The AI sales team for a business of one.其他	jvvinoth	0.00★ 0 / 0	N/A	▶
67	CrowdShift其他	iamsohungrynow	0.00★ 0 / 0	N/A	▶
68	Amplify Stack其他	adrianteoyc	0.00★ 0 / 0	N/A	▶
69	LabelOS其他	perynnneo	0.00★ 0 / 0	N/A	▶
70	Manifold: Directing AI agents to transform expertise to learning其他	aaryan126	0.00★ 0 / 0	N/A	▶
71	waggohq其他	averagejoesbuild	0.00★ 0 / 0	N/A	▶
72	ProjectAlpha其他	lonelyisland2203	0.00★ 0 / 0	N/A	▶
73	Venture Mint其他	krishsaraf	0.00★ 0 / 0	N/A	▶
74	GlydeAI其他	fabian-onlydraxco	0.00★ 0 / 0	N/A	▶
75	Live Commerce Producer其他	likalight	0.00★ 0 / 0	N/A	▶
76	Founder Build In Public其他	erickch123	0.00★ 0 / 0	N/A	▶
77	Super OPC - 牛逼智能总控其他	neometer	0.00★ 0 / 0	N/A	▶
78	Cinderlane其他Great reviews. Terrible websites. 15-20% lost to OTAs Every booking through Booking.com costs the hotel commission. No direct channel means no alternative. Content trapped on Google Great photos and reviews exist — but on a listing the hotel can't control, that looks like every other listing. No affordable design A proper website costs thousands. For a 9-room guesthouse, the economics don't work. So they go without. AI builds the page from the listing One LLM call reads the hotel's reviews and picks a layout, palette, typography, and copy from 7 archetypes. A second pass extracts the real brand from their existing site. Not a template.	bhaskar20	0.00★ 0 / 0	N/A	▶
79	Superflow其他	danleeaj	0.00★ 0 / 0	N/A	▶
80	Creator OS其他	shellylh123	0.00★ 0 / 0	N/A	▶
81	Ground Up其他	50p3d	0.00★ 0 / 0	N/A	▶
82	laoban.ai - The AI chairman of the board for your OPC其他	craboss888	0.00★ 0 / 0	N/A	▶
83	AdvisorLoop其他	gc2000	0.00★ 0 / 0	N/A	▶
84	KinKeep智能体KinKeep is a working family-care demo for adult children who support ageing parents from a distance. It turns a simulated multi-signal change against a parent’s personal baseline into a bilingual check-in, one concise family brief, and a human-approved next action. KinKeep was selected in the Top 30 at the July 12 Singapore OPC Hackathon. It has no users or revenue yet and is now entering customer validation. Reviewers should focus on the closed-loop workflow and safety architecture: deterministic rules control branching, consent, escalation, and approvals, while models are limited to natural-language replies, transcription, and structured meal-image understanding. The same care episode remains coherent across the parent and family surfaces. KinKeep is a solo-builder OPC project with an auditable public repository and working demo.	leini8891	0.00★ 0 / 0	N/A	▶
85	AdSwarm: The ad agency that runs itself.其他	nhs172k3	0.00★ 0 / 0	N/A	▶
86	Project Nudge其他	kaleb-nim	0.00★ 0 / 0	N/A	▶
87	SG AI Oasis其他	gukki2021	0.00★ 0 / 0	N/A	▶
88	AdRouter其他AdRouter helps everyday users and developers offset the cost of running AI agents. It returns clearly labeled sponsor placements through a separate path from the model's messages, tools, commands, and edits. Sensitive requests receive no placement, developers can opt out, and every completed turn shows its token use, model cost, subsidy, and remaining payment.	happycool121	0.00★ 0 / 0	N/A	▶
89	Agentic Invoice Parser其他	ywy2002	0.00★ 0 / 0	N/A	▶
90	ForeSail其他	jinyy20021018-design	0.00★ 0 / 0	N/A	▶
91	Farm.OS — the autonomous operating system for a one-person farm其他Core idea. FarmOS is an autonomous AI operating system for a one-person, off-grid, 18-vertical eco-tourism farm. A single Claude agent acts as the entire operations team — it senses the whole farm through a live digital twin, predicts what's about to go wrong, and acts through 12+ tools, logging the reasoning behind every decision. Technical angle (what's actually built). A unified in-memory digital twin (18 verticals, ~90 assets/sensors) is seeded from one config file, so the whole farm is one source of truth an agent can reason across. The supervisor runs an autonomous sense→predict→act loop via the Anthropic tool-runner, and degrades gracefully to a deterministic rule engine when offline or keyless — so the demo never breaks. Two things make it more than a chatbot: (1) a hard safety invariant — it can never shed a life-support load (incubators, aerators, cold chain, animal feed/water); and (2) offline resilience — when the internet drops it keeps operating on cached rules and queues external syncs, flushing on reconnect. There's also a real-sensor integration seam: live Home Assistant/MQTT readings route into the same twin behind the same interface, so flipping one flag runs a vertical on real hardware — no rewrite. Traction. Live in production and fully browsable (no login), public repo, both flagship demos working, plus live per-vertical P&L, 92% circular-economy loops, a staff-dispatch layer (AI directs the humans), a guest app, and an interactive farm map. What to pay attention to. (1) The crisis scenario — with battery at 22%, watch the agent shed discretionary loads while explicitly protecting every life-support load, explaining each move. (2) The offline scenario — cut the internet and it keeps the farm alive. (3) The activity log — every autonomous action carries a written rationale. The novel part isn't any single sensor or model; it's one safety-constrained agent reasoning and acting across all 18 domains at once.	kimbrolly-a11y	0.00★ 0 / 0	N/A	▶
92	kingdom of pal其他Kingdom of PAL demonstrates a governed multi-agent operating team for community-led businesses. The demo uses Tembusu Circle as the customer scenario: a founder can request a website or community update, Orin routes the work, Scribe drafts content, Rick enforces approval before publishing, and Bastion checks system health. Reviewers should look for the approval-controlled workflow, agent role separation, audit trail, Telegram/dashboard entry points, and the Gatsby demo site publishing flow.	promptalchemistlabs	0.00★ 0 / 0	N/A	▶
93	OpenCarpet其他	devhub-agent-maxim	0.00★ 0 / 0	N/A	▶
94	LarpDog其他	eseet22	0.00★ 0 / 0	N/A	▶
95	Apex Logic其他Apex Logic is a control-plane dashboard I built solo to govern my own growing fleet of AI agents, sitting between human intent and autonomous agent execution. Each agent action is logged as a three-part record — the original human prompt, the agent's own declared assumption, and the resulting cost — forming a live Intent Ledger. High-risk actions freeze at a human checkpoint (the Apex Checkpoint) before they commit, rather than being logged only after the fact. Built as a working, interactive prototype (React 19 + Vite + Tailwind 4) deployed on Vercel; all three panels (Audit Stream, Intent Ledger, Circuit-Breaking Gate) and core interactions (approve, reject, emergency stop, expiry countdown, terminal scroll, animated metrics) are live on realistic mock data. Reviewers should focus on the interaction design of the Circuit-Breaking Gate and the intent-to-cost mapping in the Intent Ledger — this is the governance layer for a solo builder running an increasing fleet of agents, not raw trace/log data.	natashaow	0.00★ 0 / 0	N/A	▶
96	StudyLy其他	hanhanhan369	0.00★ 0 / 0	N/A	▶
97	smeGrantsGoWhere其他	mistypotluck	0.00★ 0 / 0	N/A	▶
98	Barto其他	giovannimarusso	0.00★ 0 / 0	N/A	▶
99	SG Career Arcade其他	raineydiary	0.00★ 0 / 0	N/A	▶
100	Gate2Tray其他Airlines frequently experience uncertain inventory, supplier disruptions, last-minute operational changes, and safety constraints during onboard meal sales. These decisions require balancing revenue, passenger satisfaction, and operational safety. Existing workflows depend on multiple operators manually coordinating across disconnected systems. Gate2Tray OS explores how a single Fleet Commerce Operator can supervise multiple airlines simultaneously through a bounded AI agent that autonomously evaluates operational evidence, recommends actions, and pauses whenever human approval or additional evidence is required.	tecktinkerere	0.00★ 0 / 0	N/A	▶

排行榜分页

显示第 1-100 条，共 316 个项目

第 1 / 4 页