Loopers-OSS open-source AI firewall cuts LLM spend by 68%

Loopers-OSS launched an open-source wire-level AI firewall that intercepts bloated MCP tool outputs, offloads large payloads to encrypted storage and routes intermediate steps to cheaper models. In tests across 60 agent turns it cut input tokens by 75.79% and the total LLM bill by 68.35%, serving repeated read-only tool calls from memory in 3.8 ms.

• 75.79% fewer input tokens and 68.35% lower LLM bill across 60 agent turns
• 94.6% cost reduction on initial discovery turns via fast-tier models
• Repeated read-only tool calls served from memory in 3.8 ms
• Deterministic policy gating, semantic drift thresholds and budget kill switches