I built TokenFence after a coding agent hit a retry loop overnight and kept calling GPT-4o until I noticed the bill. The problem: there's no native way to say "this workflow should cost no more than $2." Account-level limits exist, but nothing per-agent or per-workflow. TokenFence wraps your LLM calls with a hard dollar cap. When it's hit, you choose: kill the run or auto-downgrade to a cheaper model to finish within budget. Two lines of code. Works with OpenAI, Anthropic, any provider. We have ~380 npm installs and ~300 PyPI installs per week — all organic, from developers who found it after getting burned. This is the first time we've told anyone it exists. Would love honest feedback on: - Is $49/mo right for teams? Too high, too low? - What assertion types would make this useful in CI? - Anyone running this in production? GitHub: https://github.com/u4ma-kev/toke... If this solves a real problem for you, an upvote helps us reach more developers dealing with the same thing.

TokenFence

Per-workflow budget caps and kill switch for AI agents

Per-workflow budget caps and kill switch for AI agents