agentacct

agentacct is an open-source, local-first Agent Work Intelligence tool for developers who use coding agents like Claude Code and Codex. It reads the session logs those clients already write on your machine, joins them with recorded work steps, and surfaces the result in a live terminal dashboard (agentacct tui). The project is published under the MIT License and is currently in public alpha, with v0.8.1 released in August 2026.

What It Is

agentacct sits in the category of LLM observability and cost-tracking tooling, specifically targeting the coding-agent workflow. Rather than requiring a cloud account or browser, it operates entirely on your local machine: state is stored in plain local files, nothing binds a network port, and there is no phone-home telemetry. The tool never stores or requests a provider API key. Its core job is to answer two questions: what did my coding agent actually do? and what did it cost?

How the Evidence Pipeline Works

agentacct keeps two evidence streams separate and joins them on real client IDs:

• Usage truth is imported from the client's own local session files. Tokens are labeled client_reported; costs come from a local pricing table and are clearly marked as estimates, never provider invoices.
• Work meaning comes from sections and events the agent records over MCP while it works (agentacct_record_section, agentacct_record_machine_check), plus machine checks like test runs.
• The join links the two through session/transcript IDs and labels every attribution exact, high, medium, or low. Missing attribution is shown as a gap rather than a guess.

For Claude Code, an installed hook bridge captures real session/transcript IDs at session start and on every tool call. For Codex, the link is evidenced from the client's own session logs at import time.

Terminal Dashboard Features

Running agentacct tui opens a live, in-terminal view with:

• Usage and cost panels — tokens per agent, model, and day, with provider rate-limit bars and reset countdowns
• Sessions list — per-session tokens, cost, status, and step counts
• Session detail — per-step work ledger showing pass/fail check evidence with exit codes; steps are labeled Verified (passing test) or Agent reported (agent's own claim)
• Weekly plan cost estimate (beta) — estimates what fraction of a weekly Claude plan each task consumed, calibrated from the user's own recorded limit history; shows "still calibrating" rather than a guess until enough data is available
• Snapshot export — press p to save a shareable SVG of the current view

Setup Path

agentacct requires Python ≥ 3.11 on macOS or Linux (Windows is supported only via WSL). Installation is via pipx or uv:

pipx install agentacct
agentacct onboard
agentacct tui

onboard detects local coding-agent logs, sets up a global store, and runs a first usage sync. It also starts a managed background sync and a local JSON API on http://127.0.0.1:8765 for shell and script polling. A --scope project flag keeps state in the repo's .agent-sentinel/ directory instead of the global store. A agentacct demo command runs a safe local walkthrough in a throwaway temporary store with no provider keys or paid API calls.

Client Support and Honest Scope

The README explicitly scopes per-client support: Claude Code and Codex have live-observed usage paths; Hermes, OpenCode, OpenClaw, and Cursor have narrower, explicitly-scoped paths. The agentacct capabilities agents command prints the current capability matrix for the user's machine. The project's stated design principle is that missing attribution beats wrong attribution.

Update: v0.8.1

The latest release is v0.8.1, published on 2026-08-04. The repository was created in July 2026 and has seen active development, with the last push recorded on 2026-08-07. The project is self-described as early alpha, with interfaces subject to change. The public alpha checklist and live smoke evidence are documented in the repository's docs/ directory.