About Agentacct
agentacct answers the question of what your coding agents have actually been doing. It reads the local session logs of tools such as Claude Code, Codex, Kimi Code, OpenCode and Hermes, then shows recent activity, tool calls, recorded steps, checks, tokens and estimated cost across projects and clients.
Each task has a Work Receipt that brings together the participating sessions, the recorded outcome, estimated cost, claims and check results, with an activity timeline from attempt to result. An agent's own statement that it is done is labeled Reported, while Verified requires passing checks after the last recorded change, and failed attempts remain visible in the history. Work groups point at a project folder to see which agents worked there and what they did most recently.
There is no account, no cloud and no telemetry; nothing leaves your computer. It needs Python 3.11 or newer on macOS or Linux, with Windows supported through WSL, or a signed, notarized macOS app for macOS 14 and later. agentacct is written in Python and licensed under MIT.
Key features
- Dashboard built from local agent session logs
- Per-task Work Receipts with outcomes
- Token usage and estimated cost
- Reported versus Verified result states
- Support for several coding agents
- Signed macOS app and Python install
Good fit for
- Reviewing what AI coding agents did
- Tracking token spend across agent tools
Agentacct: questions and answers
- What is Agentacct used for?
- agentacct is a local-first dashboard that shows what coding agents did and what it cost, built from your Claude Code, Codex and OpenCode session logs. It is a good fit for reviewing what AI coding agents did and tracking token spend across agent tools.
- Is Agentacct open source?
- Yes. Agentacct is open source under the MIT licence. Its source code is on GitHub at mikehasa/agentacct and is written mainly in Python.
- Is Agentacct free?
- Yes. Agentacct is open source, so the software itself is free to use. Paid plans are also available, starting at $9.99 per month.
- What are some alternatives to Agentacct?
- Similar open-source tools in the Developer Tools category include Munder Difflin, Agent Beacon and Mission Control. SaaS products in the same category include CNVS, Headroom and JustShipIt.
- Is Agentacct actively maintained?
- Yes. The most recent commit to Agentacct was on 3 October 2026, and the latest release is v0.12.11, published on 3 October 2026. The project has 757 stars on GitHub.
Open-source alternatives to Agentacct
See all
Munder Difflin
Developer Tools
Free and open source multi agent harness. Run Claude Code, Codex, Gemini CLI and nine more
MIT★ 9.5k
Agent Beacon
Developer Tools
The cross-harness, self-improving memory layer for AI agents.
MIT★ 3.5k
Mission Control
Developer Tools
Run agent operations from one self-hosted control plane with tasks, activity, costs, appro
MIT★ 6.5k
Atlas
Developer Tools
Atlas is a native macOS workspace for coding agents. Install Claude Code, Codex, OpenCode
Apache-2.0vs Remote★ 9.4k
Storybloq
Developer Tools
Project memory and workflows for Claude Code and Codex. Keep stories, plans, handovers, an
OSS★ 788CTX
Developer Tools
CTX Fit finds the cheapest AI coding setup that reliably works on your repository, then ap
MIT★ 590
SaaS alternatives to Agentacct
See all
CNVS
Developer Tools
Infinite canvas that runs Claude, Cursor and Codex agents in parallel, controlled by voice
SaaS
Headroom
Developer Tools
Menu bar app that compresses prompt bloat, logs and boilerplate sent to Claude Code and Codex
SaaS
JustShipIt
Developer Tools
Managed infrastructure with a workspace and kanban board where AI agents work through your backlog
SaaS
Starkomand
Developer Tools
AI coding workspace for Mac that keeps agents, issues, diagrams and previews on one canvas
SaaS
Tacit
Developer Tools
Saves every Claude Code session, extracts decisions and rationale, and makes sessions searchable
SaaS
Termi Protocol
Developer Tools
Control room that shows AI coding agents like Claude Code, Codex and Gemini working at desks in a 3D room
SaaS

