agentacct

Summary

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.

Command line

Values are placeholders taken from each option’s declared default. Nothing is executed here — the output shown is a recording of a run that already happened.

Release history

1. v0.10.4 Aug 30, 2026 · issue 012

   agentacct v0.10.4 expands secret redaction to xAI, Groq, and Google OAuth tokens, and maps more agent tool names out of the 'other' bucket.

   └──▷ GET THIS VERSION
   copy

   $ git clone --branch v0.10.4 https://github.com/mikehasa/agentacct.git
   # already have the repo? check out this version:
   $ git checkout v0.10.4

   • ›Extends value redaction to cover xAI (xai-), Groq (gsk_), and Google OAuth 2.0 access tokens (ya29.), and reports Anthropic keys (sk-ant-) under their own named class instead of a generic category.
2. v0.10.2 Aug 28, 2026 · issue 009

   agentacct v0.10.2 restores Claude Code usage ingestion and adds a dedicated 'Weekly plan' receipt row.

   └──▷ GET THIS VERSION
   copy

   $ git clone --branch v0.10.2 https://github.com/mikehasa/agentacct.git
   # already have the repo? check out this version:
   $ git checkout v0.10.2

   • ›Adds a dedicated 'Weekly plan' row to the Work Receipt (macOS app, CLI, and TUI) showing each task's calibrated share of the client's weekly Claude plan — displays a named calibration state instead of a fabricated number when uncalibrated.
3. v0.10.1 Aug 27, 2026 · issue 009

   agentacct v0.10.1 adds a user-resolve lane via POST /v1/disposition, per-agent plan cards, and codex weekly calibration.

   └──▷ GET THIS VERSION
   copy

   $ git clone --branch v0.10.1 https://github.com/mikehasa/agentacct.git
   # already have the repo? check out this version:
   $ git checkout v0.10.1

   • ›Adds POST /v1/disposition endpoint plus 'Mark reviewed', 'Resolve', and 'Reopen' controls to give a red finding or blocker an honest exit; a human resolution reads 'Finding resolved' or 'Blocker resolved' — never machine verification.
   • ›Surfaces the newest recorded blocker's own words under the task headline instead of a generic statement, so 'Why a task is blocked' is immediately readable.
   • ›Expands files, commands, tools, and check rows in-place inside a Work Receipt, making a thousand-command receipt legible without a second page.
   • ›Adds a per-agent plan card to the Dashboard — one row per recording agent showing only what that agent can prove, replacing the single merged account view.
   • ›Adds per-Task weekly-plan share (calibrated-or-nothing, client-scoped) to the cost line, summary strip, rail, and receipts table.
   +3 moreshow less

   • ›Adds a dedicated 'Stopped' tab for ended_open receipts, keeping them out of other lifecycle tabs.
   • ›Adds color families to decision words (blue for done-ish claims, amber for inferred stop, green reserved for machine-verified) with a legend popover.
   • ›Adds an in-record back control (and Esc) to return to the receipts list; the Work tab now always lands on the table; receipts default to latest-first with attention/cost sort one click away.
4. v0.10.0 Aug 26, 2026 · issue 009

   agentacct v0.10.0 adds a local GET /v1/ingestion API, checks tallies on /v1/tasks rows, and a rebuilt macOS Work Receipts dashboard.

   └──▷ GET THIS VERSION
   copy

   $ git clone --branch v0.10.0 https://github.com/mikehasa/agentacct.git
   # already have the repo? check out this version:
   $ git checkout v0.10.0

   └──▷ TRY IT
   Poll the new /v1/ingestion endpoint to check per-source import health and sync state from a script or shell.
   copy

   $ curl -s -H 'Authorization: Bearer <token>' http://127.0.0.1:8765/v1/ingestion | jq .

   Fetch /v1/tasks and inspect the new per-row check tallies to surface tasks with failing checks.
   copy

   $ curl -s -H 'Authorization: Bearer <token>' http://127.0.0.1:8765/v1/tasks | jq '[.[] | select(.checks_failed > 0) | {id, checks_total, checks_passed, checks_failed}]'

   • ›Adds bearer-gated GET /v1/ingestion local API endpoint — the /v1 twin of the legacy /ingestion/health snapshot — advertised from GET /v1/version.
   • ›Adds checks_total, checks_passed, and checks_failed fields to /v1/tasks rows, sourced from the same reducer as the full Receipt so list and detail views are always consistent.
   • ›New macOS app Dashboard surfaces work evidence: recent tasks with decision badges and evidence-tier pips, a needs-review card, live active work, the provider plan ring, and daily fresh-token history.
   • ›New Work surface in the macOS app provides a receipts table with lifecycle filter tabs, evidence-tier pips with checked/checkable ratios, a checks rail with failure annotations, cost, and recency — plus a Receipt record page with a dimensions ledger carrying provenance chips, inline gap annotations, a checks card, and an evidence-coverage card with a counted tier legend.
   • ›New Sources pane in the macOS app renders /v1/ingestion data: per-source import state and recency, continuous-sync watcher state, actionable issues, a verifier shelf, and a local-only scope card.
   +1 moreshow less

   • ›New Stamped Tile brand mark ships as a deterministic generator script for the app icon, a top-bar lockup, and a menu-bar template mark.
5. v0.9.0 Aug 17, 2026 · issue 009

   agentacct v0.9.0 adds multi-agent Work Receipts across Claude Code, Codex, OpenCode, and Hermes, plus a signed macOS app.

   └──▷ GET THIS VERSION
   copy

   $ git clone --branch v0.9.0 https://github.com/mikehasa/agentacct.git
   # already have the repo? check out this version:
   $ git checkout v0.9.0

   └──▷ TRY IT
   Scope onboarding to the current repo only, writing no global files, when you want per-project isolation.
   copy

   $ agentacct onboard --scope project

   Poll the local JSON API for machine-readable task data from a shell script or external tool.
   copy

   $ curl http://127.0.0.1:8765

   • ›Adds agentacct onboard auto-detection and instrumentation of all four supported coding agents (Claude Code, Codex, OpenCode, Hermes) in a single run.
   • ›Adds agentacct tui live terminal dashboard showing tasks, evidence tiers, cost, and checks across all four agents.
   • ›Adds agentacct onboard --scope project for a per-repo install instead of the default global install.
   • ›Starts a loopback-only local JSON API at http://127.0.0.1:8765 during onboarding for machine-readable polling by shells and scripts.
   • ›Supports four coding agents — Claude Code, Codex, OpenCode, and Hermes — through a single client-agnostic Task and Work Receipt model.
   +5 moreshow less

   • ›Derives Receipt Actions (commands run, files edited, tools used) from each agent's own on-disk transcript store at import time for agents whose hooks do not fire (Codex, OpenCode), with no hook required.
   • ›Adds independent check evaluation for OpenCode using recorded exit codes to promote steps from self-checked to independently-checked.
   • ›Labels Actions provenance per session — distinguishing live client-hook observations from transcript-scan derivations — so scan-derived data is never presented as hook-observed.
   • ›Ships a signed and notarized macOS app (agentacct-0.9.0.dmg) bundling the CLI; requires macOS 14+.
   • ›Adds a weekly Claude plan cost-fraction estimate per task, calibrated from the user's own recorded limit history rather than raw token counts.
6. v0.8.0 Aug 4, 2026 · issue 009

   agentacct v0.8.0 adds a shareable SVG snapshot key, a plan column on the sessions list, and store-fallback for global installs.

   └──▷ GET THIS VERSION
   copy

   $ git clone --branch v0.8.0 https://github.com/mikehasa/agentacct.git
   # already have the repo? check out this version:
   $ git checkout v0.8.0

   • ›Read commands (tui, now, limits) now fall back to the machine-wide store when run from a directory with no project store, so a global install no longer requires --store-dir; a misconfigured --store-dir or AGENTACCT_STORE_DIR now surfaces its error instead of silently reading the global store.
   • ›Press p from any screen in agentacct tui to save a shareable SVG snapshot of the current view, written under the store (not the working directory), renderable in any browser or on GitHub.
   • ›The full sessions list (press s in agentacct tui) now includes the same weekly-plan plan % column shown on the home panel.
   • ›The per-session weekly-plan % figure is now displayed only when it has been calibrated from your own recorded limit history; otherwise the session detail shows a 'calibrating from your own usage' note and the list shows —, replacing the previous shipped universal baseline.
7. v0.7.0 Aug 3, 2026 · issue 009

   agentacct v0.7.0 adds a live TUI dashboard, session step drill-downs, plan-consumption estimates, and new CLI commands tui, limits, now, and sync.

   └──▷ GET THIS VERSION
   copy

   $ git clone --branch v0.7.0 https://github.com/mikehasa/agentacct.git
   # already have the repo? check out this version:
   $ git checkout v0.7.0

   └──▷ TRY IT
   Launch the live terminal dashboard to monitor usage, costs, and rate-limit windows across all agents without leaving the terminal.
   copy

   $ agentacct tui

   Check current provider rate-limit windows and remaining quota read from local files, without opening a browser.
   copy

   $ agentacct limits

   Print a current usage and cost snapshot for a quick status check from a script or shell alias.
   copy

   $ agentacct now

   • ›New agentacct tui command launches a live terminal dashboard showing calendar usage windows (today / last 7 days / last 30 days / all time), by-client and top-models breakdowns, and provider rate-limit bars with used-percentage and reset countdowns — no browser or credentials needed.
   • ›Within agentacct tui, press u for a per-day or per-week token and cost time series with by-model detail, scoped to one client or model with c / m keys.
   • ›Within agentacct tui, press s to drill into sessions, showing each run and its subagents folded into one row with step-level status (in progress, handed off, done, blocked) and machine-check results including pass/fail marks and exit codes.
   • ›New agentacct limits command reports provider rate-limit windows read passively from local files — Codex session rollouts, Claude desktop plan-usage history, and a lightweight Claude CLI status-line hook.
   • ›New agentacct now command prints a current usage and cost snapshot for a chosen window.
   +3 moreshow less

   • ›New agentacct sync command forces an on-demand re-sync of MCP config, hooks, and instructions across client integrations.
   • ›Adds a plan column and per-task '≈ X% of your weekly plan' estimate to the home panel and session detail, weighted per model from a measured baseline and self-calibrated from your own 7-day recorded usage history (including Codex and Claude CLI).
   • ›Signed-out or cancelled provider accounts with frozen limit readings are now automatically detected and hidden from the limits panel.
8. v0.6.0 Aug 1, 2026 · issue 009

   agentacct v0.6.0 makes SQLite the default event ledger and adds three new commands to manage the migration

   └──▷ GET THIS VERSION
   copy

   $ git clone --branch v0.6.0 https://github.com/mikehasa/agentacct.git
   # already have the repo? check out this version:
   $ git checkout v0.6.0

   └──▷ TRY IT
   Confirm the SQLite ledger is a faithful copy of the old flat file before retiring it — useful right after an upgrade when both files still exist.
   copy

   $ agentacct event verify-log

   Remove the events.jsonl backup once you've verified the SQLite copy is complete and no old daemons are still writing to it.
   copy

   $ agentacct event drop-flat-ledger --confirm

   Keep a long-running daemon on the legacy flat-file path while you test the SQLite upgrade on a separate machine.
   copy

   $ AGENTACCT_EVENT_LOG_AUTHORITATIVE=0 agentacct tui

   • ›Switches the default event ledger from events.jsonl to events.sqlite3, eliminating full-file re-parses on every read and write.
   • ›Adds AGENTACCT_EVENT_LOG_AUTHORITATIVE=0 environment variable to opt back into the legacy flat-file mode after upgrading.
   • ›Adds agentacct event verify-log command to prove the SQLite copy matches a flat file line-for-line.
   • ›Adds agentacct event drop-flat-ledger --confirm command to retire the events.jsonl backup once a store has fully cut over.
   • ›Adds agentacct canonical rebuild-store command to rebuild the SQLite usage index from the authoritative ledger.
   +1 moreshow less

   • ›Existing events.jsonl stores auto-adopt SQLite on first open, keeping the flat file as a backup; straggler writes from not-yet-restarted daemons are drained into the new log so no events are lost during rolling upgrades.
   └──▷ BREAKING ON UPGRADE

   • !The owner-gated agentacct evidence rebuild subcommands (snapshot / build-candidate / activate / rollback) have been removed entirely; the rebuild subsystem that replayed the v1 ledger from events.jsonl is gone.
9. v0.5.3 Jul 31, 2026 · issue 009

   agentacct v0.5.3 surfaces refused-recording counts, honest partial/stopped task states, and a new handed_off section status

   └──▷ GET THIS VERSION
   copy

   $ git clone --branch v0.5.3 https://github.com/mikehasa/agentacct.git
   # already have the repo? check out this version:
   $ git checkout v0.5.3

   • ›Adds a handed_off section status so an agent can record a clean stop when continuing in a new session, stored as a terminal state rather than a live one.
   • ›Local logs now surface counts and reason codes for every recording call agentacct refused — derived at read time from existing on-disk data, covering refusals that predate this release, storing only counts and reason codes (never the offending value or path).
   • ›Tasks now expose a partial verification count (e.g. '3 of 5 steps verified') instead of a single verified/unverified flag.
10. v0.4.0 Jul 29, 2026 · issue 009

    agentacct v0.4.0 makes onboard global by default, writing zero files into your repo.

    └──▷ GET THIS VERSION
    copy

    $ git clone --branch v0.4.0 https://github.com/mikehasa/agentacct.git
    # already have the repo? check out this version:
    $ git checkout v0.4.0

    └──▷ TRY IT
    Opt into a per-repo install instead of the new global default, scoping hooks and state to a single project.
    copy

    $ agentacct onboard --scope project

    • ›Changes agentacct onboard to install globally by default (zero repo files written); use agentacct onboard --scope project for a per-repo install instead.