export declare const HELP_TEXT = "ai-usage -- local token usage and cost across coding agents\n\nUsage: ai-usage <command> [options]\n\nCommands:\n  report                Verdict, where usage went, what to change (same as usage_report)\n                        Defaults to the last 7 days.\n  status                Collectors, data stores, database path, record counts, last sync\n  sync                  Run the collectors and store what they find\n  stats                 Totals for a period, split by client   (same numbers as the usage_summary MCP tool)\n  models                Per-model tokens and cost              (same as model_usage)\n  clients               Per-client tokens and cost             (same as client_usage)\n  projects              Per-project tokens and cost            (same as project_usage)\n  sessions              Recent sessions                        (same as recent_sessions)\n  session <id>          One session in detail                  (same as session_usage)\n  daily                 Per-day breakdown                      (same as daily_usage)\n  breakdown             Two or more dimensions at once         (same as usage_breakdown)\n  counterfactual        These tokens on another model          (same as counterfactual_cost)\n  budget                Spend against a target, with run rate and projection\n  limits                Pro/Max 5-hour and weekly limit usage   (same as plan_usage)\n  statusline            One line for Claude Code's status bar; also keeps the\n                        plan-limit readings `limits` reports. Reads Claude Code's\n                        status-line JSON on stdin. Ends with the session's context\n                        size (ctx 77k), yellow from --ctx-warn, red from --ctx-limit.\n  hook                  Claude Code UserPromptSubmit hook: warns Claude and you when the\n                        session's context is at or above --ctx-limit, silent otherwise.\n                        Reads the hook JSON on stdin, opens no database, always exits 0.\n  export                One row per stored turn, as CSV or JSON Lines\n  import <file>         Merge JSON Lines from another machine (idempotent)\n  prune                 Remove records older than a date (dry run without --yes)\n  vacuum                Compact the database file after a prune\n  verify                Re-read the source data and diff it against the local database\n  version               Print the installed version\n  help                  Show this text\n\nPeriod options (default: all time):\n  --today               Just today, local time\n  --days N              Last N days, from local midnight\n  --since <ISO>         Explicit start (inclusive)\n  --until <ISO>         Explicit end (exclusive)\n  --compare previous    stats only: also report the equal-length window before this one,\n                        with the delta. Needs a period of fixed length: --days, --today,\n                        or --since AND --until. Open-ended (--since alone, --until alone,\n                        all time) has no window before it and is rejected, not ignored.\n  --grain <g>           daily only: hour | day | hour-of-day (default day).\n                        hour-of-day collapses every day onto one 24-slot local clock.\n  --by a,b              breakdown only: up to 3 of client, model, provider, project,\n                        session, day, hour, hour-of-day. Crosses them in one query.\n\nScope options (repeatable, or comma-separated; each matches ANY value given):\n  --client <name>       claude-code | opencode | codex\n  --model <id>          Restrict to these models      (--model a,b or --model a --model b)\n  --project <path>      Restrict to these projects (their working directories)\n  --target-models a,b   counterfactual only: models to price the selected tokens AGAINST.\n                        Not a filter -- `--model` chooses which turns, this chooses the\n                        rates. (`--models` is an accepted alias.)\n\nList options (sessions, models, projects, clients):\n  --limit N             Rows to return\n  --offset N            Rows to skip, for paging. Output tells you the next offset.\n  --sort <key>          tokens | reported-cost | estimated-cost | records | sessions | recent\n                        Default: recent for sessions, tokens elsewhere. There is no plain\n                        `cost`: reported and estimated cost are never summed, so ordering by\n                        one sorts every row priced on the other basis as $0.\n  --no-subagents        Exclude subagent/sidechain turns (included by default)\n  --all-stores          Read every detected data store, not only the one the client itself uses\n  --full                Ignore saved sync cursors and re-read everything\n  --json                Emit JSON instead of text\n  --no-color            statusline only: no ANSI colour (NO_COLOR=1 does the same)\n  --ctx-warn <n>        statusline, hook: context size where `ctx` turns yellow. Takes 150k,\n                        1.5m or 150000. Default 150k; AI_USAGE_CTX_WARN sets it too.\n  --ctx-limit <n>       statusline, hook: context size where compacting is advised (red, and\n                        the hook speaks up). Default 250k; AI_USAGE_CTX_LIMIT sets it too.\n\nScripting options:\n  --field <path>        Print ONE value from the JSON result and nothing else, so a shell\n                        can use it without jq. e.g. --field overall.cost.estimated\n  --fail-over <n>       Exit 1 when --field's value exceeds n. Requires --field: there is\n                        no default, because reported and estimated cost are never summed\n                        and a default would ignore every record priced the other way.\n  --format <f>          export only: csv (default) | jsonl. --csv is shorthand for csv.\n\nLifecycle options:\n  --before <ISO>        prune only: remove records BEFORE this instant (exclusive), so\n                        --before 2026-01-01 removes 2025 and keeps New Year's Day.\n  --yes                 prune only: actually delete. Without it, prune is a dry run.\n\nBudget options (budget command):\n  --amount N            The target, in USD. Required.\n  --basis <b>           reported | estimated. Required -- the two are never summed, so a\n                        budget with no stated basis is a budget against nothing.\n  --period <p>          month (default) | week. Calendar periods only: a projection needs\n                        a period end to aim at, which a rolling window has not got.\n                        `budget` exits 1 when spend already exceeds the target.\n\nExamples:\n  ai-usage sync\n  ai-usage stats --today\n  ai-usage stats --days 7\n  ai-usage models --days 30 --client claude-code\n  ai-usage projects --days 30\n  ai-usage stats --days 7 --compare previous\n  ai-usage daily --days 7 --grain hour\n  ai-usage daily --days 30 --grain hour-of-day\n  ai-usage breakdown --by project,day --days 30\n  ai-usage breakdown --by model,day --days 7 --sort estimated-cost\n  ai-usage sessions --limit 5\n  ai-usage sessions --sort estimated-cost --limit 5     # the costliest, not the latest\n  ai-usage projects --sort estimated-cost --limit 10\n  ai-usage sessions --limit 100 --offset 100            # page two\n  ai-usage models --model claude-opus-5,claude-sonnet-5\n  ai-usage counterfactual --today --target-models claude-sonnet-5,claude-haiku-4-5\n  ai-usage counterfactual --model claude-opus-5 --target-models claude-sonnet-5\n  ai-usage budget --amount 500 --basis estimated\n  ai-usage budget --amount 20 --basis reported --period week\n  ai-usage export --days 30 > usage.csv\n  ai-usage export --format jsonl > laptop.jsonl   # then, on the desktop:\n  ai-usage import laptop.jsonl\n  ai-usage prune --before 2026-01-01              # dry run\n  ai-usage prune --before 2026-01-01 --yes && ai-usage vacuum\n  ai-usage stats --today --field overall.cost.estimated\n  ai-usage stats --today --field overall.cost.estimated --fail-over 25   # exit 1 if over\n  ai-usage limits\n  ai-usage verify\n\nStatus line: add this to ~/.claude/settings.json, after `npm i -g ai-usage-mcp`:\n  \"statusLine\": { \"type\": \"command\", \"command\": \"ai-usage statusline\" }\nClaude Code passes it your plan-limit usage; `ai-usage limits` reports what it kept.\n\nContext warning: every turn re-sends the whole conversation, so a long session gets costly.\nAdd this beside statusLine to be told before it does (the status line shows `ctx 77k`):\n  \"hooks\": { \"UserPromptSubmit\": [ { \"hooks\": [\n    { \"type\": \"command\", \"command\": \"ai-usage hook\", \"timeout\": 10 } ] } ] }\n\nNotes:\n  `status` also checks npm for a newer version, at most once a day. It sends no\n  data, gives up after 1.5s offline, and `AI_USAGE_NO_UPDATE_CHECK=1` disables it.\n\n  `sync` also downloads LiteLLM's public price list, at most once a day, to price\n  Claude models released after the built-in table. It only fills models the table\n  lacks, sends no data, and `AI_USAGE_NO_PRICING_REFRESH=1` disables it (as does\n  `AI_USAGE_NO_UPDATE_CHECK=1`).\n\n  Cost from OpenCode is what OpenCode reported. Cost for Claude Code and Codex is\n  an API-equivalent estimate from a versioned pricing table, because neither\n  records a cost -- on a subscription your marginal cost per request is $0. The\n  two are never added together.\n";