Your Agent Is Leaking Tokens. Build a Decision Ledger on a Free Server

说明与用途

使用免费MonkeyCode服务器构建一个记录每个决策到SQLite账本并在达到令牌预算时停止的代理。

问题

An LLM‑powered agent can repeatedly call the same tool, rephrase, and retry, causing silent token budget leaks without errors.

结果

A lightweight, self‑auditing agent that logs every decision and enforces a token budget, preventing silent leaks on a free server.

工作流

• Provision a free Linux server via MonkeyCode dashboard and install Python 3.10+ and requests
• Export MC_API_KEY, MC_BASE_URL, MC_MODEL, and MC_BUDGET_TOKENS as environment variables
• Create agent.py: initialize SQLite ledger, define call_llm(), a placeholder search_tool(), and main() loop
• In main(), loop while used tokens < BUDGET: call LLM, parse JSON action, execute tool, record decision and token usage in ledger
• Verify token counts from LLM response and update used counter
• Stop loop when budget exceeded or LLM returns stop action

证据

MonkeyCode’s free tier includes a 10M token allowance and a free server option.

来源

The final artifact is a SQLite ledger with a hard token budget.

来源