https://aitoolchain.io/tools/agentacct New releases and features in agentacct, tracked by The AI Toolchain. en Sun, 30 Aug 2026 22:18:06 GMT https://github.com/mikehasa/agentacct/releases/tag/v0.10.4 https://github.com/mikehasa/agentacct/releases/tag/v0.10.4 Sun, 30 Aug 2026 22:18:06 GMT 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. • 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. https://github.com/mikehasa/agentacct/releases/tag/v0.10.2 https://github.com/mikehasa/agentacct/releases/tag/v0.10.2 Fri, 28 Aug 2026 17:19:35 GMT agentacct v0.10.2 restores Claude Code usage ingestion and adds a dedicated 'Weekly plan' receipt row. • 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. https://github.com/mikehasa/agentacct/releases/tag/v0.10.1 https://github.com/mikehasa/agentacct/releases/tag/v0.10.1 Thu, 27 Aug 2026 07:38:18 GMT agentacct v0.10.1 adds a user-resolve lane via `POST /v1/disposition`, per-agent plan cards, and codex weekly calibration. • 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. • 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. https://github.com/mikehasa/agentacct/releases/tag/v0.10.0 https://github.com/mikehasa/agentacct/releases/tag/v0.10.0 Wed, 26 Aug 2026 14:10:31 GMT agentacct v0.10.0 adds a local `GET /v1/ingestion` API, checks tallies on `/v1/tasks` rows, and a rebuilt macOS Work Receipts dashboard. • 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. • 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. https://github.com/mikehasa/agentacct/releases/tag/v0.9.0 https://github.com/mikehasa/agentacct/releases/tag/v0.9.0 Mon, 17 Aug 2026 17:09:38 GMT agentacct v0.9.0 adds multi-agent Work Receipts across Claude Code, Codex, OpenCode, and Hermes, plus a signed macOS app. • 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. • 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. https://github.com/mikehasa/agentacct/releases/tag/v0.8.0 https://github.com/mikehasa/agentacct/releases/tag/v0.8.0 Tue, 04 Aug 2026 09:38:21 GMT agentacct v0.8.0 adds a shareable SVG snapshot key, a `plan` column on the sessions list, and store-fallback for global installs. • 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. https://github.com/mikehasa/agentacct/releases/tag/v0.7.0 https://github.com/mikehasa/agentacct/releases/tag/v0.7.0 Mon, 03 Aug 2026 19:29:51 GMT 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`. • 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. • 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. https://github.com/mikehasa/agentacct/releases/tag/v0.6.0 https://github.com/mikehasa/agentacct/releases/tag/v0.6.0 Sat, 01 Aug 2026 09:45:23 GMT agentacct v0.6.0 makes SQLite the default event ledger and adds three new commands to manage the migration • 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. • 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 changes: • 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. https://github.com/mikehasa/agentacct/releases/tag/v0.5.3 https://github.com/mikehasa/agentacct/releases/tag/v0.5.3 Fri, 31 Jul 2026 14:17:41 GMT agentacct v0.5.3 surfaces refused-recording counts, honest partial/stopped task states, and a new `handed_off` section status • 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. https://github.com/mikehasa/agentacct/releases/tag/v0.4.0 https://github.com/mikehasa/agentacct/releases/tag/v0.4.0 Wed, 29 Jul 2026 15:48:25 GMT agentacct v0.4.0 makes `onboard` global by default, writing zero files into your repo. • Changes `agentacct onboard` to install globally by default (zero repo files written); use `agentacct onboard --scope project` for a per-repo install instead.