forces a live quota probe of all Claude / z.ai / Codex accounts
· auto every {{ov.quota_poll_minutes||'?'}}m
⚠ Grant health
{{staleGrants.length}} Claude account(s) haven't refreshed in >{{ov.grant_stale_hours}}h — the grant may be dying; re-auth before it forces a prompt:
{{staleGrants.map(a=>a.name).join(', ')}}
{{prov}} {{vendorStatus(prov).text}} {{(ov.providers[prov]||[]).length}} account(s) · pool free {{cw.label}} {{cw.avail}}%
⚠ No usable {{prov}} account
— {{stuckClients(prov).length}} client(s) are stranded on a dead/depleted grant with nowhere to fail over
({{stuckClients(prov).join(', ')}}). Re-auth or enable an account, or have that machine run
authpool-agent reauth.
{{a.name}}
⚠ dup of {{(a.duplicate_of||[]).join(', ')}} {{statusText(a)}}{{a.has_creds?'creds stored':'NO creds'}} ·
cached · {{ago(dataAsOf(a))}}
probed {{ago(dataAsOf(a))}}
· token {{ago(a.token_expires_at,true)}}
⚠ Codex token expired {{ago(a.token_expires_at)}} — the gauge below is a frozen snapshot from the last Codex turn
({{a.quota_as_of?ago(a.quota_as_of):'unknown'}}); it only refreshes when Codex is used on a connected client.
⛔ blocked — a client's CLI hit a rate/quota-limit message on this account, so the server pulled it
from rotation (reactive report; {{resetIn(a.blocked_until)}}).
A healthy live probe clears it early —
{{a.provider}} has no live gauge to verify a false alarm (a benign "rate limit" phrase in CLI output trips it too) —
{{w.label}}
stale · reset {{ago(w.reset)}}
{{w.pct}}% · {{resetIn(w.reset)}}
{{gaugeEmpty(a)}}
usage credits {{a.extra_usage.is_enabled?'(on)':'(off)'}}
{{a.extra_usage.utilization_pct}}%
used {{a.extra_usage.used_credits}} / {{a.extra_usage.monthly_limit}}
monthly tools (web search / reader)
{{a.mcp_uses.utilization_pct}}% · {{resetIn(a.mcp_uses.resets_at)}}
used {{fmt(a.mcp_uses.used)}} / {{fmt(a.mcp_uses.limit)}} · never blocks the account (separate tools budget)
plan: {{planOf(a)}}
usage resets banked: {{a.reset_credits}}
reserves: 5h {{a.max_5h_pct||'—'}} · 7d {{a.max_7d_pct||'—'}}
grant:
{{a.last_refresh?'refreshed '+ago(a.last_refresh):'not yet'}}
stale >{{ov.grant_stale_hours}}h
reclaimed {{ago(a.last_reclaim_at)}}
drift {{ago(a.last_drift_at)}}
login: {{a.identity||((a.provider==='z.ai'||a.provider==='kimi')?'— (unknown — set the account email via caps; API keys carry no identity to learn)':a.provider==='claude'?'— (unknown — learns automatically on the next poll/re-auth, or set it via caps)':'— (unknown — learns automatically from the stored credentials on the next poll, or hit ⟳ refresh)')}}
· id {{a.cred_fingerprint}}
⚠ duplicate: {{dupWhy(a)}} as {{(a.duplicate_of||[]).join(', ')}} — both rows are ONE underlying account (identical usage); re-auth one of them into the correct login
used by: , {{c.label||c.key_prefix}}still on {{clientDrift(c,a)}}no clients
📌 forced: , {{c.label||c.key_prefix}}
models: main {{(a.models||{}).ANTHROPIC_MODEL||'—'}} · opus {{modelOf(a,'OPUS')}} · sonnet {{modelOf(a,'SONNET')}} · haiku {{modelOf(a,'HAIKU')}} · forced
no accounts — add one below
| account | status | gauge | reserves | actions | |
|---|---|---|---|---|---|
| {{a.name}} · {{a.identity}} | {{statusText(a)}}⚠ dup | {{gaugeSummary(a)}} | 5h {{a.max_5h_pct||'—'}} / 7d {{a.max_7d_pct||'—'}} | ||
|
{{w.label}}
{{w.pct}}% · {{resetIn(w.reset)}}
usage credits {{a.extra_usage.is_enabled?'(on)':'(off)'}} — used {{a.extra_usage.used_credits}} / {{a.extra_usage.monthly_limit}}
{{gaugeEmpty(a)}}
| |||||
| no accounts — add one below | |||||
Global GLM model mapping (Claude Code; overrides per-account)
opus {{(ov.glm_global&&ov.glm_global.models||{}).ANTHROPIC_DEFAULT_OPUS_MODEL||'—'}} ·
sonnet {{(ov.glm_global&&ov.glm_global.models||{}).ANTHROPIC_DEFAULT_SONNET_MODEL||'—'}} ·
haiku {{(ov.glm_global&&ov.glm_global.models||{}).ANTHROPIC_DEFAULT_HAIKU_MODEL||'—'}} ·
force {{ov.glm_global&&ov.glm_global.force?'yes':'no'}}
Global Kimi model mapping (Claude Code; overrides per-account)
main {{(ov.kimi_global&&ov.kimi_global.models||{}).ANTHROPIC_MODEL||'—'}} ·
opus {{(ov.kimi_global&&ov.kimi_global.models||{}).ANTHROPIC_DEFAULT_OPUS_MODEL||'—'}} ·
sonnet {{(ov.kimi_global&&ov.kimi_global.models||{}).ANTHROPIC_DEFAULT_SONNET_MODEL||'—'}} ·
haiku {{(ov.kimi_global&&ov.kimi_global.models||{}).ANTHROPIC_DEFAULT_HAIKU_MODEL||'—'}} ·
force {{ov.kimi_global&&ov.kimi_global.force?'yes':'no'}}
Moderato+/Allegretto preset: map everything to k3 (and raise CLAUDE_CODE_AUTO_COMPACT_WINDOW to 1048576 for the 1M context).
Add / replace account credentials
Paste the whole .credentials.json (claude), auth.json (codex), oauth_creds.json (gemini), or the raw API key (z.ai / kimi — sk-kimi-… from kimi.com/code/console, NOT a platform.kimi.ai key):
Clients
| label | status | os / host | ver | installed |
|---|---|---|---|---|
| {{c.label||c.key_prefix}} | {{c.online?'online':'offline'}} | {{c.os||'?'}} / {{c.hostname||'?'}} | {{c.client_version||'?'}}old | {{installedStr(c)}} |
Realtime model usage
moving-average depletion forecast · auto-refreshes every {{predEvery}}s
Global hard limit
a model is abandoned & treated as depleted at this %
{{pred.global_hard_limit_pct}}%
No gauge readings yet — add accounts with a usage gauge (Claude / Codex / z.ai) and let them poll.
{{a.name}} {{a.provider}}
limit {{a.effective_hard_limit_pct}}% ·
cached · {{ago(dataAsOf(a))}}
probed {{ago(dataAsOf(a))}}
{{w.label}} {{predText(w.status)}}
{{w.utilization_pct}}% / {{w.threshold}}%
{{w.rate_pct_per_hour!=null?('+'+w.rate_pct_per_hour+'%/h'):'rate n/a'}}
{{predEta(w)}}
resets {{fmtDur(w.reset_seconds)}}
Usage est. cost, last {{usageData.days}}d
{{fmt(totals.req)}}requests
{{fmt(totals.tok)}}total tokens
{{fmt(totals.input)}}input
{{fmt(totals.output)}}output
{{fmt(totals.cache_read)}}cache read
${{totals.cost.toFixed(2)}}est. cost
Daily tokens
By provider
Model usage ({{msShare?'share of daily ':'daily '}}{{msMetric==='cost'?'est. cost':msMetric}}{{msProv==='all'?'':' · '+msProv}} — the model "market share" over time)
y-axis
view
vendor
range
machines
users
no per-model usage recorded in this range / filter yet
Activity heatmap ({{heatMetric}} per day × hour, {{heatTzLabel}} local time — live data)
range
metric
timezone
machines
accounts
models
{{(h-1)%6===0?(h-1)+'h':''}}
less
more peak {{fmt(heat.max)}} {{heatMetric}}/hour · {{heatDaysN}} days
Subscription utilization (daily peak % per account · red = limit hit)
window
range
No utilization recorded yet — it accumulates as the server polls each account's gauge (Claude / Codex / z.ai). Build up over a few days.
{{a.name}} {{a.provider}}
peak {{a.peak}}% · {{a.hits}} limit-hit day(s) · limit {{a.limit}}%
0%
100% / limit · grey = no reading · {{utilDaysN}} days
Machine → account (current assignments)
| machine | status | claude | codex | gemini |
|---|---|---|---|---|
| {{m.machine}} | {{m.online?'online':'offline'}} | {{(m.accounts||{}).claude||'—'}} | {{(m.accounts||{}).codex||'—'}} | {{(m.accounts||{}).gemini||'—'}} |
| no active assignments | ||||
By machine
| machine | provider | requests | tokens | est. cost |
|---|---|---|---|---|
| {{r.machine}} | {{r.provider}} | {{fmt(r.requests)}} | {{fmt(r.tokens)}} | ${{(r.cost||0).toFixed(2)}} |
| no usage yet | ||||
By provider
| provider | requests | input | output | cache | est. cost |
|---|---|---|---|---|---|
| {{r.provider}} | {{fmt(r.requests)}} | {{fmt(r.input)}} | {{fmt(r.output)}} | {{fmt(r.cache_read)}} | ${{(r.cost||0).toFixed(2)}} |
By CLI × vendor
| CLI | vendor | requests | tokens | est. cost |
|---|---|---|---|---|
| {{r.cli||'?'}} | {{r.provider}} | {{fmt(r.requests)}} | {{fmt(r.input+r.output+r.cache_write+r.cache_read)}} | ${{(r.cost||0).toFixed(2)}} |
By day
| day | requests | tokens | est. cost | |
|---|---|---|---|---|
| {{r.day}} | {{fmt(r.req)}} | {{fmt(r.tok)}} | ${{r.cost.toFixed(2)}} |
By model
| provider | model | requests | input | output | cache | est. cost |
|---|---|---|---|---|---|---|
| {{r.provider}} | {{r.model||'?'}} | {{fmt(r.requests)}} | {{fmt(r.input)}} | {{fmt(r.output)}} | {{fmt(r.cache_read)}} | ${{(r.cost||0).toFixed(2)}} |
Cost is an estimate from the pricing table; the real budget on subscriptions is the % gauges on Overview.
Historic usage (ccusage backfill — all time; no per-account; not summed with the live figures above)
{{fmt(historicTotals.tok)}}historic tokens
${{historicTotals.cost.toFixed(2)}}est. cost
{{(usageData.historic_by_day||[]).length}}day×provider rows
By model (all time)
| provider | model | input | output | cache | est. cost |
|---|---|---|---|---|---|
| {{r.provider}} | {{r.model||'?'}} | {{fmt(r.input)}} | {{fmt(r.output)}} | {{fmt(r.cache_read)}} | ${{(r.cost||0).toFixed(2)}} |
By day (all time)
| day | provider | tokens | est. cost |
|---|---|---|---|
| {{r.day}} | {{r.provider}} | {{fmt(r.input+r.output+r.cache_write+r.cache_read)}} | ${{(r.cost||0).toFixed(2)}} |
No historic data yet — it's backfilled from ccusage hourly by each agent (client ≥1.0.24) and on client install/update (needs Node + ccusage on that machine). Today's row keeps growing through the day.
Clients / API keys
Client installer
Give each user this download + a per-computer API key (mint one below).
Then on their machine: extract, pip install ".[client]", and
{{enrollCmd}}
New API key (copy now — shown once):
{{newKey}}
| label | prefix | OS / host | version | last seen | status | |
|---|---|---|---|---|---|---|
| {{c.label||'—'}} | {{c.key_prefix}}… | {{c.os_version||c.os||'?'}} · {{c.hostname}} | {{c.client_version||'?'}} update failed ⟳ {{updPhase(c)}} outdated | {{c.online?'online':'offline'}}{{c.last_seen?ago(c.last_seen):'never'}} | {{c.revoked?'revoked':(c.enabled?'enabled':'disabled')}} | ⟳ updating… |
Admin users
| user | status | last login |
|---|---|---|
| {{u.username}} | {{u.disabled?'disabled':'active'}} | {{u.last_login||'—'}} |
Audit log
| time | actor | event | provider | detail |
|---|---|---|---|---|
| {{tshort(r.ts)}} | {{r.actor}} | {{r.event}} | {{r.provider||''}} | {{r.detail?JSON.stringify(r.detail):''}} |