NOF0 AI Trading Arena
Compare LLM and prompt trading strategies through live-market PNL.
The README openly states that the system connects to real cryptocurrency markets and performs live trading, and it identifies snapshot data categories and local output locations, making external effects and part of the data flow visible. It provides no evidence of least-privilege controls, pre-trade confirmation, exposure limits, credential or sensitive-data safeguards, dependency security checks, or cancellation and rollback procedures. NOF1, major frameworks, the exchange, and an author account are attributed, so source attribution is reasonably developed but incomplete; unknown publisher identity is not treated as suspicious by itself.
The architecture, feature list, and progress statements are mostly consistent, and the README candidly reports a 70% backend and an execution engine still undergoing stability work. There is nevertheless tension between the “out-of-the-box” claim and the unavailable backend startup path. Frontend installation commands, stack versions, and submodule guidance offer limited dependency-availability support, but complete prerequisites, pinning or recovery guidance are absent. No failure-message taxonomy, diagnostic behavior, or degradation handling is evidenced.
The intended audience and scenarios are clear: operators deploying a multi-model live-trading arena to compare prompts and model performance. The README also says this is not a backtesting tool and discloses component maturity, giving useful capability boundaries. It does not specify exactly when an agent may initiate a trade or how triggers are stopped and constrained, and it omits operating-system, runtime, account, exchange, and LLM configuration requirements, so environment fit remains preliminary.
The README has clear navigation across structure, quick start, stack, snapshots, and related resources, with reasonably stable component naming. Installation guidance fully covers only the frontend, while the backend is explicitly unfinished. Examples are mostly commands and feature lists rather than an end-to-end configuration or FAQ. Limitations are meaningfully disclosed through progress labels and the statement that this is not a backtester. MIT is asserted, but no license file or license metadata is present in the supplied evidence. There is no version policy or changelog. An author handle, Telegram channel, and documentation entry provide contact paths, but maintenance policy, support ownership, and verified publisher responsibility remain unclear.
Asset curves, positions, trades, conversations, leaderboards, and model detail views form a concrete and usable output set. A multi-model live arena with prompt comparisons offers clear marginal value over a single-strategy interface. Current end-to-end utility is reduced by the unfinished backend and execution engine, and the source does not quantify LLM usage, trading fees, infrastructure cost, financial exposure, or maintenance effort, leaving cost-benefit treatment thin.
Some statements are traceable to named directories, scripts, documentation paths, and upstream projects. However, the supplied evidence consists only of the README, with no code, tests, configuration, or supporting files to substantiate claims such as a complete reproduction, real data, or stability. There is therefore no cross-source corroboration. Progress declarations and design principles are partly labeled, but promotional claims, established facts, and inference are not systematically separated or paired with verification evidence.
- This is a live cryptocurrency trading project; manually verify order confirmation, position and loss limits, emergency stopping, cancellation, and recovery controls before connecting funds.
- The source does not explain storage, transmission, log redaction, or rotation for API keys, exchange credentials, LLM data, or model conversations.
- The backend is reported as only 70% complete and has no usable startup procedure in the README, so “out-of-the-box” should not be treated as verified end-to-end capability.
- A README alone cannot confirm dependency pinning, security auditing, the license file, test coverage, or the stated functionality and progress.
What does this agent do, and when should you use it?
NOF0 is a self-hostable AI cryptocurrency trading arena intended to reproduce the Alpha Arena experience of NOF1.ai. It places multiple Prompt-LLM agents in a competition with a stated starting balance of $10,000 each and presents their profit-and-loss results visually. Its independently runnable frontend uses Next.js 15, React 19, TypeScript, Recharts, and Zustand to show equity curves, positions, trades, model conversations, rankings, and model details. The Go-Zero backend is organized into executor, llm, manager, exchange, market, and prompt packages. A snapshot command can also download several NOF1.ai upstream datasets into timestamped JSON files for offline inspection. The published progress indicators put the frontend at 100%, the backend at 70%, and the AI workflow execution engine at 80%, so the complete live-trading path is still under development.
The intended loop is: devise a strategy, write a Prompt, trade in the live market, review the PNL ranking, and iterate on the Prompt. The web/ application renders total-equity charts, current positions, trade history, Model Chat, a leaderboard, and per-model details; it uses Recharts for visualization and Zustand for state. Under go/pkg/, executor handles AI data flows and workflows, llm wraps model providers, manager manages strategies, exchange defines exchange access, market handles market data, and prompt contains Prompt templates. The execution engine is intended to run those strategies against real market data and publish actual PNL to the dashboard, although that engine and the backend are not yet complete. Separately, running npm run snapshot:nof1 in web/ writes crypto prices, positions, trades, account totals, inception values, leaderboard data, analytics, and conversations to snapshots/nof1/<ISO timestamp>/ as JSON plus index.json.
- A quantitative researcher wants to compare several LLM or Prompt strategies under a common starting balance and PNL leaderboard.
- A prompt strategist needs positions, fills, equity curves, and model conversations to guide successive revisions of a trading prompt.
- A frontend developer wants to run an Alpha Arena-style Next.js dashboard without waiting for the unfinished Go backend.
- A crypto experimentation team wants a self-hosted arena foundation that it can continue integrating with a Hyperliquid trading workflow.
- A data analyst wants timestamped offline JSON snapshots of NOF1.ai prices, positions, trades, rankings, conversations, and analytics.
What are this agent's strengths and limitations?
- The central experiment compares Prompt strategies using live PNL rather than presenting only a conventional historical backtester.
- The frontend runs independently and already covers equity, positions, trades, conversations, rankings, and model detail views.
- The Go backend separates execution, model-provider access, strategy management, exchange access, market data, and Prompt templates into distinct packages.
- A single documented command captures eight classes of NOF1.ai upstream data for offline inspection.
- The provider-wrapper and Prompt-template structure is described without limiting the arena to one named model vendor.
- The backend is reported at 70% completion and the execution engine at 80%, leaving the full live-trading workflow unfinished.
- There are no copyable instructions for starting the backend, connecting Hyperliquid, configuring secrets, or placing the first trade.
- Live trading exposes capital to loss, but the supplied material documents no risk limits, recovery process, or emergency-stop mechanism.
- Compatible Node.js, npm, and Go versions are not specified, adding environment-validation work for adopters.
- The supplied repository metadata says the license is unknown while the README says MIT License; adopters should verify the actual license file.
How do you install or deploy this agent?
Clone the repository, enter it, and enable recursive Git submodule updates:
git clone https://github.com/wquguru/nof0.git
cd nof0
git config submodule.recurse trueStart the independently runnable frontend:
cd web
npm install
npm run devThen open http://localhost:3000. The source identifies Next.js 15, React 19, and TypeScript, but does not specify supported Node.js or npm versions. It also does not document backend startup, exchange-account setup, API credentials, or LLM-provider credentials; the backend remains in development.
How do you use this agent?
For the first documented working path, start the frontend and visit http://localhost:3000 to inspect equity curves, positions, trades, Model Chat, the leaderboard, and model details. To collect offline sample data, run npm run snapshot:nof1 from web/; the command creates JSON files and index.json under snapshots/nof1/<ISO timestamp>/. No complete live-trading configuration or launch command is supplied, so the available instructions are insufficient for an end-to-end deployment of the exchange, model provider, and execution engine.
How does this agent compare with similar options?
NOF0 explicitly positions itself as an open-source reproduction of the NOF1.ai Alpha Arena that users can deploy themselves. Its backend and execution engine are still incomplete, however, so the supplied material does not establish current feature parity across the full trading workflow.
FAQ
Can it execute live trades immediately?
Does the frontend require the backend?
npm install followed by npm run dev.Which API keys or accounts are required?
Are downloaded snapshots committed to Git?
snapshots/nof1/<ISO timestamp>/ content is described as excluded through .gitignore.