AuditMyAgent — Task ledger and cost transparency for AI agents

See whether your AI agent finished the job, and what it cost in tokens and dollars

Problem

More people hand coding and business tasks to AI agents (local or hosted), yet nobody can say cleanly whether a run actually completed, which tools it called, or why it ended at that price. Someone is explicitly building 'Did your AI Agent accomplish its task? and at what cost?', others complain AI emitted thousands of lines of code with no way to know what each line does, and the flood of multi-agent frameworks (local or hosted) lacks a unified execution and cost view.

Solution

A lightweight agent observability SDK plus a local dashboard: drop a few dozen lines of tracing hooks into your agent loop to record every tool call, input/output summaries, token usage, and estimated cost, then surface a three-column report of task completion, actual spend, and suspicious steps. Data lives in local SQLite and exports as JSON for CI or team sharing.

Users

Indie developers and small teams using LangChain, Claude Code, or self-hosted/hosted multi-agent frameworks; engineers who need to explain AI budgets to their manager.

MVP

A Python/TypeScript SDK that wraps common agent loops; local SQLite logging of every call and token count; a static dashboard showing completion, cost, and tool-call timeline; JSON and CLI report export; built-in cost tables for OpenAI, Anthropic, and local Ollama.

Revenue

Open-source SDK with a paid SaaS dashboard (team collaboration, historical comparison, alerts); updated cost tables as a subscription perk; optional usage-based API pricing.

Why now

As agents move from demos to daily coding, cost and completion are the first metrics that need quantification, but general observability tools like LangSmith are platform-bound, leaving local/hybrid gaps. A functional SDK plus dashboard is a two-week build, and a single hook plus a cost table is a credible wedge for an indie developer.