agentacct v0.7.0

agentacct 0.7.0 — a live terminal dashboard

agentacct tui — usage, cost, and work in the terminal

agentacct now has a live terminal dashboard. agentacct tui shows calendar usage windows (today / last 7 days / last 30 days / all time), a by-client and top-models breakdown, and provider rate-limit bars for each subscription — the 5-hour and weekly windows with their used percentage and a reset countdown — all read from the same authoritative local event log, with no browser, no server, and no credentials. Press u for a dedicated usage screen (a per-day or per-week token and cost time series with a by-model detail, scoped to one client or model with c / m), s to drill into your sessions, r to refresh, and q to quit.

The work, per session — steps and honest check evidence

The sessions drill-down (s) folds a run and its subagents into one row and shows each session's status at a glance: in progress, handed off, done, or blocked, derived from the session's own recorded steps. Open a session to see that work as a list of steps — each step's status and summary, and its machine-check results with pass/fail marks and exit codes, so a failed test is shown as evidence rather than hidden. A signed-out or cancelled provider account, whose last limit reading is frozen forever, is now detected and hidden so the limits panel shows only your live subscriptions.

Estimate a task's share of your weekly plan (beta)

The home panel and the session detail now estimate what fraction of your weekly Claude subscription each task consumed — a plan column, and an "≈ X% of your weekly plan" line. Raw token counts are the wrong unit and cost alone is not enough, because different models burn the plan at very different rates, so the estimate weights per model from a measured baseline and self-calibrates a single per-account factor from your own recorded 7-day usage history (from Codex and the Claude CLI status-line too, not only the desktop app). It is always labeled an estimate, only claims a calibrated scale when the fit is trustworthy, and sharpens as more limit history accrues.

Also new

agentacct limits reports provider rate-limit windows read passively from local files — Codex session rollouts, the Claude desktop plan-usage history, and a lightweight Claude CLI status-line hook — and agentacct now prints a current usage and cost snapshot for a chosen window. Client integrations also re-sync themselves after an agentacct upgrade so a machine's MCP config, hooks, and instructions don't drift, and a new agentacct sync forces that refresh on demand.