agentacct — Track what your AI coding agents actually do, what it costs, and if their work is verified.

Analyzed by Sai Pavan Gopularam · AI · Observability · View on GitHub

Stars: 734
Forks: 79
Commits last 30 days: 100
Health: Active (100 commits this month)
Language: Python
License: MIT

What It Is

Imagine a detailed expense report and timesheet combined, but for your AI coding agents. `agentacct` processes the session logs from tools like Claude Code or Codex, creating a 'Work Receipt' for each task. It shows you every step the agent took, what files it changed, which tests ran, and how much time and tokens were spent.

This matters because agents often claim 'done' without providing proof. `agentacct` cuts through the hype, showing concrete evidence (or lack thereof) for every action. It helps you trust your agents, understand their efficiency, and control costs by seeing exactly what they're doing on your machine, without sending data to the cloud.

agentacct GitHub repository card

License Verdict

MIT License — Build and Sell Freely — Commercial Use Approved • No Copyleft Restrictions

The MIT License is highly permissive, allowing you to use, copy, modify, merge, publish, distribute, sublicense, and sell copies of the software. You can incorporate `agentacct` into proprietary software and sell it commercially without needing to open-source your own code, provided you include the original license.

How to Use It

Install `agentacct` via `pipx` to get a local-first dashboard. After installation, onboard it to detect your agents and set up its local data store. Then, launch the terminal user interface (TUI) to start monitoring your agent's work.

Prerequisites:

Python 3.11+
macOS or Linux (WSL for Windows)

Estimated setup time: 10 minutes.

pipx install agentacct
agentacct onboard
agentacct tui

What I'd Build With This

AI Agent Cost & Performance Dashboard for Freelancers (micro-saas)

Build a lightweight web dashboard that connects to `agentacct`'s local API (http://127.0.0.1:8765) to provide a more visual and shareable overview of agent activity. Freelance developers or small agencies could use this to generate client-ready reports on agent-assisted project work, proving value and justifying costs. Offer features like multi-project aggregation and simple cost breakdowns.

Effort: 2 Weeks Build Time · Target: Freelance Developers, Small Agencies · Pricing: $29/mo

Team-Wide AI Agent Observability Platform (saas)

Develop a SaaS platform that aggregates `agentacct` data from multiple developer machines within a team or organization. This would offer engineering managers a holistic view of agent usage, costs, and verification status across their entire team. Features could include anomaly detection for high-cost agent runs, compliance checks for agent-generated code, and team-level performance analytics.

Effort: 3 Months Build Time · Target: Engineering Managers, DevOps Teams · Pricing: $99/user/month

Custom Agent Audit & Compliance Integration (enterprise)

Offer consulting and custom development services to integrate `agentacct`'s core logic into large enterprise development workflows and internal audit systems. This would be for companies needing to ensure AI agent outputs meet strict regulatory or internal quality standards, providing an immutable, local-first audit trail of agent actions and costs. Focus on integrating with existing CI/CD, security, and governance tools.

Effort: 6 Months Build Time · Target: Large Enterprises, Regulated Industries · Pricing: $50,000+ per project

Sai Pavan Gopularam's Take

This is a fantastic tool for bringing transparency to AI agent usage, which is a massive black box for most developers. The 'Work Receipt' concept is brilliant for building trust and verifying agent output, which is crucial for adoption. I'd lean into offering a cloud-synced version for small teams, charging $49/month for shared visibility and reporting across multiple machines.

Watch Out For

Early Alpha Software: The project is explicitly labeled 'early alpha' and 'would rather show you a gap than a guess,' meaning some functionality might be incomplete or evolve rapidly. Expect potential breaking changes or missing features.
Local-First Only: By design, `agentacct` is local-first, with no cloud features, telemetry, or login. While great for privacy, this means no built-in multi-machine sync or centralized team dashboard out of the box.
Cost Estimates, Not Invoices: The tool provides cost estimates based on pricing tables, marked with `≈ agentacct — What It Is & What You Could Build With It . It cannot access your actual provider invoices, so the reported costs are approximate and not definitive billing records.
Limited Agent Integration: While it supports several agents, the level of detail and verification varies. Claude Code and Codex have the fullest receipts, while others might have less comprehensive data. Always check the coverage matrix for your specific agent.
No Conversation Recording: For privacy reasons, `agentacct` records tool names, files, commands, and tokens, but never your prompts, the model's responses, or full transcripts. This limits its use for detailed conversational analysis.

I break down trending repos like agentacct every week — join the newsletter.

Browse all free repo breakdowns