Tuneloop
Turns your raw AI coding agent transcripts into a dashboard that shows exactly what each feature or PR actually cost, which tools kept failing, and where the agent kept redoing the same work.
🔗 Visit Tuneloop
Description
If you use an AI coding agent regularly, you generate a pile of session transcripts but almost never look back at them — which means you have no real visibility into what any given feature actually cost in tokens, or which parts of your workflow are wasteful. Tuneloop is local analytics software that ingests those transcripts and turns them into a dashboard: cost attribution per shipped artifact (a specific PR or feature), session outcome rates, and tool/skill usage with error rates.
It links transcripts to pull requests both explicitly and by content-matching, tracks cost at the block level across tasks, assesses how autonomously the agent operated, and can detect recurring re-work patterns — a signal that the agent keeps redoing the same task rather than actually finishing it. There's a `tuneloop query` command for read-only SQL access to the analytics data directly, and enrichment supports multiple providers (Anthropic, OpenAI, AWS Bedrock, Ollama, OpenRouter, Groq). The core tool is open source under the MIT license, with paid enrichment features running roughly $6 per ~100 sessions.
💬 Our review
The short version: this fills a real blind spot — most developers using AI coding agents have no idea which tasks are actually cost-effective versus which ones the agent quietly struggles with and redoes — and having that data as a local dashboard rather than scattered across raw transcripts makes it genuinely actionable.
Against manually reviewing transcripts (the default for most people, meaning effectively never), Tuneloop automates the analysis and surfaces patterns like recurring re-work that are hard to notice by eye across dozens of sessions. Against a hosted analytics SaaS for AI agent usage, Tuneloop runs locally and keeps the core tool free and open source, with the paid tier only for enrichment — a reasonable trade for developers who want to keep transcripts on their own machine. The nearly-free enrichment pricing ($6 per ~100 sessions) means the cost barrier to trying the paid tier is minimal.
💰 Pricing
📊 Global score
🤖 AI-enriched data
Outil de base open source (MIT), gratuit. Fonctions d'enrichissement payantes : environ 6$ pour ~100 sessions
Pros
Attribution de coût par PR/fonctionnalité livrée
Détecte les schémas de re-travail répétitif de l'agent
Accès SQL en lecture seule aux données d'analyse
Cons
Nécessite d'avoir déjà un volume de transcripts à analyser
Enrichissement multi-fournisseurs payant, même si peu cher
Projet indépendant récent