Sparkco

Sparkco's website is a public log of an open-source research project on the agent stack (runtime containers, harnessing, and messaging between agents). It presents a live feed of AI and prediction-market signals, blog posts on prediction-market trading with CLI agents on Kalshi and Polymarket, a curated directory of 69+ CLI tools, and a list of open-source repositories including SimpleFunctions CLI, Python SDK, MCP examples, and SDK adapters for LangChain, CrewAI, OpenAI Agents SDK, and Vercel AI SDK. The site also describes ongoing research into parametric memory (encoding conversation state into model weights), stating early iterations were negative. The page states tools are shipped as CLIs first rather than MCP.

Key features

• Live feed of prediction-market and AI signals refreshed every 5 minutes
• Curated directory of CLI tools for AI agents, filterable by category
• Open-source SimpleFunctions CLI for Kalshi and Polymarket trading with 42 or 50 commands (page cites both counts)
• Python SDK (simplefunctions-ai) providing a world model across 30k+ prediction markets
• MCP server for prediction-market data usable with Claude, Cursor, and VS Code
• SDK adapters for LangChain, CrewAI, OpenAI Agents SDK, and Vercel AI SDK
• Scaffolding tool (create-prediction-market-agent) for LangChain, CrewAI, OpenAI Agents SDK, or vanilla TypeScript projects
• GitHub Action (world-state-action) for injecting market state into CI/CD workflows
• Dual-agent harness runtime (@spfunctions/harness) with a 5-message protocol
• Open-source research on parametric memory and test-time training for agent memory

Best suited for

• AI agent developers building prediction-market or trading agents
• Traders and developers using CLI tools to monitor Kalshi and Polymarket
• Developers seeking curated CLI tools and framework adapters for agents
• Researchers interested in agent memory and test-time training

Pricing

Not stated on the website