You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Proposal: show Muse token usage and cost on the Usage page
#20926
Muse turns never appear on Usage. Every other built-in provider has a
usage reader (Codex, Claude Code, Grok, Pi, and the rest), but muse is not a
usage provider kind, so someone who runs Muse in T3 sees no token counts
or cost for it. This is separate from Usage → Limits subscription windows
(#18274, #17397): this is about the per-day token/cost history the other
readers feed.
Proposal
Add a transcripts usage reader for the Muse driver, in the same shape as
the Pi and Grok readers:
Read Muse's own session.jsonl logs (one model_completed event per
model step, carrying the step's token counts), keyed per record so copied
logs count once.
Add muse to the usage provider kinds and provider lists on server, web
and mobile, plus a line in the user usage guide.
Pricing goes through the existing path only. Standard Muse Wonder has a
public rate and prices from the existing table. Muse Wonder Full has no
public per-token rate (it is Muse subscription only), so it would show
tokens unpriced unless the user sets a price in Model prices. No rate is
hardcoded.
Scope: one new reader file plus the provider-kind plumbing and tests, about
480 changed lines in a draft I have locally. No new settings, no default
changes, no network calls.
Two things I would call out up front:
Muse reports cache_write_tokens next to the input counts. I cannot
confirm from the logs whether writes are already included in input_tokens; every step in my logs has zero writes, so token totals
today are unaffected either way, but the split may need revisiting once
writes appear.
The reader follows the account's XDG_DATA_HOME for the log location,
with the usual default, matching how the other readers locate their logs.
Ask
Is this direction and scope acceptable for a PR? Happy to adjust (for
example tokens-only with no pricing hooks, or a smaller first step) before
opening one.
reacted with thumbs up emoji reacted with thumbs down emoji reacted with laugh emoji reacted with hooray emoji reacted with confused emoji reacted with heart emoji reacted with rocket emoji reacted with eyes emoji
Uh oh!
There was an error while loading. Please reload this page.
Problem
Muse turns never appear on Usage. Every other built-in provider has a
usage reader (Codex, Claude Code, Grok, Pi, and the rest), but
museis not ausage provider kind, so someone who runs Muse in T3 sees no token counts
or cost for it. This is separate from Usage → Limits subscription windows
(#18274, #17397): this is about the per-day token/cost history the other
readers feed.
Proposal
Add a
transcriptsusage reader for the Muse driver, in the same shape asthe Pi and Grok readers:
session.jsonllogs (onemodel_completedevent permodel step, carrying the step's token counts), keyed per record so copied
logs count once.
museto the usage provider kinds and provider lists on server, weband mobile, plus a line in the user usage guide.
public rate and prices from the existing table. Muse Wonder Full has no
public per-token rate (it is Muse subscription only), so it would show
tokens unpriced unless the user sets a price in Model prices. No rate is
hardcoded.
Scope: one new reader file plus the provider-kind plumbing and tests, about
480 changed lines in a draft I have locally. No new settings, no default
changes, no network calls.
Two things I would call out up front:
cache_write_tokensnext to the input counts. I cannotconfirm from the logs whether writes are already included in
input_tokens; every step in my logs has zero writes, so token totalstoday are unaffected either way, but the split may need revisiting once
writes appear.
XDG_DATA_HOMEfor the log location,with the usual default, matching how the other readers locate their logs.
Ask
Is this direction and scope acceptable for a PR? Happy to adjust (for
example tokens-only with no pricing hooks, or a smaller first step) before
opening one.
All reactions