1rok serves as a versatile standalone harness designed to run multiple portfolio-construction agents using various LLM (Large Language Model) providers, including OpenAI, Anthropic, Gemini, xAI, DeepSeek, GLM, and OpenRouter. This project enables agents to interact with a range of financial data sources such as Alpaca, Yahoo Finance, FRED, and Tavily, using a sophisticated inline tool registry defined within the repository.

A feature of 1rok is its live leaderboard that monitors the performance of each model's portfolio during paper trading on Alpaca, accessible via investingbench.vercel.app.

Key Features

• Inline Tool Registry: Offers functionalities for listing and calling tools through local handlers, maintaining a single registry for each pipeline run.
• Diverse Tool Groups: Comprises eight tool groups, including market, stock, research, technical analysis, options, earnings, portfolio management, and Tavily web search.
• Multiple LLM Providers: Integrates seven LLM providers into a single workflow, facilitating a seamless tool-calling loop.
• Specialist Agents: Includes roles such as orchestrator, screener, and various analyst types like fundamental, valuation, technical, sentiment, catalyst, macro, and risk agents to enhance decision-making capabilities.
• Two-Stage Pipeline: The pipeline architecture includes a run command that generates a portfolio-construction JSON artifact, followed by an execute command that interprets this artifact to place orders via Alpaca.
• Provider-Agnostic Schemas: Utilizes Zod definitions that convert to each provider's tool-call format, streamlined with shared retry and loop logic.

Agent Pipeline Overview

The project operates with a structured pipeline consisting of four stages involving ten specialized agents running weekly. The flow starts with the Macro agent, which assesses the economic regime, followed by the Screener that identifies potential candidates. Six agents then evaluate the candidates in parallel:

flowchart TD
    Macro["Macro Agent<br/><i>The Economist</i>"]:::entry
    Screener["Screener Agent<br/><i>The Scout</i>"]:::entry

    Sentiment["Sentiment<br/><i>Mood Reader</i>"]:::analysis
    Fundamental["Fundamental<br/><i>Accountant</i>"]:::analysis
    Valuation["Valuation<br/><i>Appraiser</i>"]:::analysis
    Catalyst["Catalyst<br/><i>Event Watcher</i>"]:::analysis
    Risk["Risk<br/><i>Risk Manager</i>"]:::analysis
    Technical["Technical<br/><i>Chart Reader</i>"]:::analysis

    Orchestrator["Orchestrator Agent<br/><i>The CIO</i>"]:::synthesis
    Constructor["Portfolio Constructor<br/><i>The Trader</i>"]:::execution
    Execute["Order Execution<br/><i>Alpaca API</i>"]:::execution

    Macro --> Screener
    Screener --> Sentiment
    Screener --> Fundamental
    Screener --> Valuation
    Screener --> Catalyst
    Screener --> Risk
    Screener --> Technical

    Sentiment --> Orchestrator
    Fundamental --> Orchestrator
    Valuation --> Orchestrator
    Catalyst --> Orchestrator
    Risk --> Orchestrator
    Technical --> Orchestrator

    Orchestrator --> Constructor
    Constructor --> Execute

    classDef entry stroke:#ff8c00,stroke-width:2px
    classDef analysis stroke:#888,stroke-width:1px
    classDef synthesis stroke:#22c55e,stroke-width:2px
    classDef execution stroke:#aaa,stroke-width:1px

The architecture of the project is designed to facilitate efficient interactions with trading APIs while maintaining a clear and structured workflow:

CLI (run | execute)
   │
   ▼
Provider ── TradingPipeline ── InlineToolRegistry (per run)
                │                       │
                ▼                       ▼
        Specialist agents ───── Tool handlers
                                        │
                                        ▼
                              src/data/services
                                        │
                                        ▼
                Alpaca · Yahoo Finance · FRED · Tavily

For those interested in the programmatic use, the integration is greatly simplified:

import { createProviderFromModel } from "1rok/harness";
import { TradingPipeline } from "1rok/pipeline";

const provider = createProviderFromModel("gpt-5.2-medium");
const pipeline = new TradingPipeline({ provider });
const result = await pipeline.run();

For detailed usage and additional functionalities, refer to the full README within the repository. The comprehensive structure of 1rok offers a robust solution for harnessing advanced LLMs for effective trading portfolio construction and management.