Cost Guard — spend limits for AI agents

Cost Guard records and limits what your agents consume — requests, LLM calls, runtime — and says “unknown” instead of guessing when the real bill isn’t connected. A Team edition (shared limits across agents, one view for every agent’s spend) is being scoped. Tell us if you’d pay for it.

What works today (open source)

• Limits and records on what the agent side can actually observe: request counts, LLM calls, runtime, CPU/memory time
• When the real API bill is not connected, spend is marked unknown — no guessed numbers, no zero-filling
• The recorder and replay tooling ship today in the 20131-agentruntime repo

Team edition — planned, not built

• Shared spend limits across all your agents (planned)
• Every agent’s recorded spend visible on one screen (planned)
• Alerts when a run crosses its recorded limit (planned)

Price: UNKNOWN — we’re asking before we set it. This page records interest only. No payment is collected and nothing is promised.