PromptLedger

PromptLedger is a web app (with optional self-hosted agent) for LLM ops teams who need provable governance over prompts and model calls. It captures prompts, variables, tool calls, model versions, latency, token spend, and outputs across environments, then ties them to releases and experiments. The core value is an immutable audit trail: who changed what prompt, when, why, and what it did to quality and cost. It also provides regression checks by replaying a fixed evaluation set against new prompt/model versions and flagging drift. Unlike generic logging, it focuses on prompt lifecycle management: approvals, change requests, rollbacks, and compliance exports (SOC2-friendly evidence). It integrates via SDKs and OpenTelemetry so teams don’t have to rewrite their stack. Realistically, this wins only if it’s dead-simple to adopt and clearly reduces incidents and spend within the first week.