Pioneer CLI

Command-line interface for the Pioneer AI training platform with an intelligent chat agent.

██████╗ ██╗ ██████╗ ███╗   ██╗███████╗███████╗██████╗ 
██╔══██╗██║██╔═══██╗████╗  ██║██╔════╝██╔════╝██╔══██╗
██████╔╝██║██║   ██║██╔██╗ ██║█████╗  █████╗  ██████╔╝
██╔═══╝ ██║██║   ██║██║╚██╗██║██╔══╝  ██╔══╝  ██╔══██╗
██║     ██║╚██████╔╝██║ ╚████║███████╗███████╗██║  ██║
╚═╝     ╚═╝ ╚═════╝ ╚═╝  ╚═══╝╚══════╝╚══════╝╚═╝  ╚═╝

Features

• Interactive Chat Agent: Claude Code-like experience with bash execution, file operations, and code sandbox
• ML Integrations: Modal.com for serverless GPU compute, Weights & Biases for experiment tracking
• Self-Evolution: Autonomous self-improvement with feedback collection and model fine-tuning
• Budget Management: Token, time, and cost tracking with configurable limits
• Pioneer Platform: Full access to Pioneer AI training platform APIs

Installation

Quick Install (requires Bun)

curl -fsSL https://pioneer.ai/install.sh | sh

Manual Install

# Install Bun if needed
curl -fsSL https://bun.sh/install | bash

# Clone and install
git clone https://github.com/fastino-ai/pioneer-cli.git
cd pioneer-cli/
bun install

# Run directly
bun run src/index.tsx --help

Chat Agent

The chat agent provides a Claude Code-like experience with powerful capabilities:

Starting the Chat

# Interactive chat mode
pioneer chat

# With specific provider/model
pioneer chat --provider anthropic --model claude-sonnet-4-20250514
pioneer chat --provider openai --model gpt-4o

# Run a single command
pioneer chat --message "Create a Python script that analyzes CSV files"

Capabilities

| Capability | Description | |------------|-------------| | Bash Execution | Run shell commands, install packages, manage system | | File Operations | Read, write, edit, search files and directories | | @ File References | Reference local files with @path syntax like Claude Code | | Code Sandbox | Execute Python, JavaScript, TypeScript, Bash, Ruby, Go in isolation | | Modal.com | Deploy ML workloads on serverless GPUs | | Weights & Biases | Track experiments and metrics | | Model Training | Fine-tune models with LoRA or full training |

@ File References

Reference local files and directories using the @ syntax, similar to Claude Code:

# Reference a single file
> Explain @src/index.tsx

# Reference multiple files
> Compare @package.json and @tsconfig.json

# Reference a directory (shows tree structure)
> What's in @src/

# Relative paths work too
> Look at @./config.ts and @../other-project/file.ts

When you use @path, the file contents are automatically:

1. Read from your filesystem
2. Included in the context sent to the LLM
3. Displayed with a checkmark indicator

This allows the agent to see and understand your code without manually reading files.

Chat Commands

| Command | Description | |---------|-------------| | /help | Show available commands | | /clear | Clear conversation history | | /tools | List available tools | | /budget | Show token/cost usage | | /exit | Exit the chat |

Environment Variables

# LLM Provider Keys (required for chat)
export ANTHROPIC_API_KEY="your-anthropic-key"
export OPENAI_API_KEY="your-openai-key"

# Pioneer Platform
export PIONEER_API_URL="https://agent.pioneer.ai"
export PIONEER_API_KEY="your-pioneer-key"

# ML Integrations (optional)
export MODAL_TOKEN_ID="your-modal-token-id"
export MODAL_TOKEN_SECRET="your-modal-token-secret"
export WANDB_API_KEY="your-wandb-key"

Configuration

Configuration is stored in ~/.pioneer/config.json:

{
  "apiKey": "your-pioneer-api-key",
  "baseUrl": "https://agent.pioneer.ai",
  "agent": {
    "provider": "anthropic",
    "model": "claude-sonnet-4-20250514"
  },
  "budget": {
    "maxTokens": 100000,
    "maxCost": 1.0,
    "maxTime": 3600
  },
  "ml": {
    "wandb": {
      "project": "my-project",
      "entity": "my-team"
    }
  }
}

Pioneer Platform Usage

# Authentication
pioneer auth login       # Enter API key interactively
pioneer auth logout      # Clear stored API key
pioneer auth status      # Check if logged in

# Datasets
pioneer dataset list
pioneer dataset get <id>
pioneer dataset delete <id>
pioneer dataset download <id>
pioneer dataset analyze <id>

# Training Jobs
pioneer job list
pioneer job get <id>
pioneer job logs <id>
pioneer job create --model-name "My Model" --dataset-ids ds_123,ds_456

# Models
pioneer model list       # List deployed models
pioneer model trained    # List trained models
pioneer model delete <id>
pioneer model download <id>

Architecture

src/
├── index.tsx          # CLI entry point and routing
├── config.ts          # Configuration management
├── api.ts             # Pioneer platform API client
├── chat/
│   └── ChatApp.tsx    # Interactive chat UI
├── agent/
│   ├── Agent.ts       # Main agent orchestrator
│   ├── LLMClient.ts   # OpenAI/Anthropic client
│   ├── ToolRegistry.ts # Tool management
│   ├── BudgetManager.ts # Budget tracking
│   └── types.ts       # Type definitions
├── tools/
│   ├── bash.ts        # Shell command execution
│   ├── filesystem.ts  # File operations
│   ├── sandbox.ts     # Code sandbox
│   ├── modal.ts       # Modal.com integration
│   ├── wandb.ts       # W&B integration
│   └── training.ts    # Model training
└── evolution/
    ├── EvolutionEngine.ts  # Self-improvement loop
    ├── FeedbackCollector.ts # Feedback storage
    ├── EvalRunner.ts       # Evaluation framework
    └── ModelTrainer.ts     # Fine-tuning

Self-Evolution System

The agent can autonomously improve itself within budget constraints:

1. Feedback Collection: Records successful/failed interactions
2. Evaluation: Runs test cases to measure performance
3. Training: Fine-tunes models based on feedback
4. Budget-Aware: Respects token/time/cost limits

import { EvolutionEngine } from './evolution';

const engine = new EvolutionEngine({
  targetScore: 0.9,
  maxIterations: 10,
  budgetPerIteration: {
    maxTokens: 50000,
    maxCost: 0.50,
  },
});

await engine.evolve(agent);

Development

cd pioneer
bun install
bun run dev          # Hot reload
bun run typecheck    # Type checking
bun run chat         # Start chat agent

Tech Stack

• Runtime: Bun
• UI: Ink (React for CLI)
• Language: TypeScript
• LLMs: Anthropic Claude, OpenAI GPT
• ML Compute: Modal
• Experiment Tracking: Weights & Biases

License

MIT