agentacct
See what your coding agents did and what it cost. Breaks each task down into work steps — tools used, files changed, tests run, time and tokens spent. Local-first dashboard for Claude Code, Codex, OpenCode, and more. No login, no telemetry.
What is it?
What it is
A local-first dashboard for tracking coding agent token usage, costs, and work evidence by analyzing session logs and agent-recorded tasks.
Why it exists
To provide transparent, privacy-preserving observability of agent activity and costs without cloud dependencies or telemetry.
Who should use it
Developers using AI coding agents like Claude Code or CodexAI engineers needing local-first observability for LLM agentsUsers concerned with privacy who want to keep agent logs and costs on their own machineDevelopers looking to track token usage and estimated costs per agent session
Who should avoid it
Users on Windows without WSLUsers requiring cloud-synced dashboards or remote telemetryUsers needing exact financial invoice data (only provides estimates)Users looking for a production-ready, non-alpha tool
How it works
A quick walkthrough in plain English
How agentacct works
Step 1 of 3
You interact with it
Open agentacct, send a request, or connect it to your stack.
Features
Advantages
- Complete privacy: data never leaves your machine
- No need to share API keys or credentials
- Transparent cost estimates clearly labeled
- Honest attribution: shows gaps instead of false links
- Works with existing agent logs without modification
- Easy installation via pipx or uv tool
- Local dashboard accessible via browser
- Supports both global and project-scoped installations
- Provides detailed work evidence (test results, machine checks)
- Early alpha with active development and clear documentation
- MIT licensed, free to use
Disadvantages
- Early alpha software; interfaces may change
- Currently supports Claude Code and Codex with full features; other agents have limited support
- Cost estimates are not exact invoices
- Requires Python >=3.11 and macOS/Linux (Windows only via WSL)
- Requires manual onboarding per machine or repo
- May show attribution gaps when links cannot be proven
- No hosted dashboard or cloud backup option
- Limited to agents that produce compatible session logs
Installation
native
pipx install agentacct
FAQ
How do I install and start agentacct on my machine?
Install via pipx (or uv/venv) with `pipx install agentacct`, then run `agentacct onboard` once per machine (global by default). This detects your local coding‑agent logs, sets up a global store, runs an initial sync, and starts the dashboard at http://127.0.0.1:8765. Open a new agent session after onboarding so that session becomes the first recorded Task.
Is my data private and does agentacct send anything to the cloud?
Yes. agentacct is local‑first: all state is stored in plain local files, the dashboard binds only to 127.0.0.1, and there is no phone‑home telemetry, no account, no cloud sync, and it never stores or requests provider API keys.
How do I view the dashboard after onboarding?
After running `agentacct onboard`, the managed dashboard starts automatically and is accessible at http://127.0.0.1:8765 in your browser. You can also control it manually with `agentacct start`, `status`, `stop`, and `repair`.
What steps are required to completely uninstall agentacct?
Run `agentacct stop` to halt the managed sync and dashboard, then `agentacct uninstall-autostart` if you enabled autostart, followed by `pipx uninstall agentacct`. Finally, remove the onboarding‑added state: for a global install delete `~/.local/state/agentacct/state` (or keep it for history) and clean agentacct entries from `~/.claude.json`, `~/.claude/settings.json`, `~/.claude/hooks/`, and `~/.codex/config.toml`. For a per‑project install delete the repo’s `.agent-sentinel/` directory and the related config entries.
Can I try agentacct without affecting my real coding‑agent data?
Yes. The command `agentacct demo` launches a safe, temporary walkthrough in an isolated store, using no provider keys and making no paid API calls, so you can explore the UI and features risk‑free.
Featured in Videos
YouTube tutorials and walkthroughs for agentacct
Alternatives
Similar projects ranked by category, topics, and text overlap.
