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
Usage Limits merges Claude CLIProxy accounts that share an e-mail, hiding one organization's limits #33343
One Claude login belongs to two organizations (a personal Max 20x subscription and a Team seat). Both are separate credentials in a CLIProxyAPI usage hub. Usage → Limits shows only one bar per e-mail, so one organization's limits disappear. With 5 Claude credentials in the hub (2 e-mails × 2 orgs, plus one single-org e-mail) the Claude pool shows 3 bars. The hidden one here was the Team seat with 84% weekly left, so the pooled "Weekly" read 14% left instead of roughly 26%. Which org survives depends on which read was freshest. Same on every client.
Diagnosis
apps/server/src/usage/cliproxyApi.tsreadAccount sets workspaceId only for Codex (id_token.chatgpt_account_id); Claude hub accounts get none.
packages/shared/src/usageLimits.tsaccountKey keys accounts as claudeAgent:<email>, and accountKeyResolver only splits by workspace when one is reported. Both orgs therefore share a key and collectLimitAccounts → merge keeps only the fresher read.
The shared code already supports several workspaces per e-mail (tests in usageLimits.test.ts), and the native Claude provider reports workspaceId = oauthAccount.organizationUuid (claudeResetCredits.tsreadClaudeOrganizationId).
Suggested fix: for provider === "claude" hub accounts, fetch https://api.anthropic.com/api/oauth/profile through the hub api-call (same headers as the usage call) and set workspaceId to organization.uuid. Each org then gets its own bar, and a hub credential still merges with a native login to the same org. Using the hub file id instead would split hub and native views of the same account.
Steps to reproduce
Use one Claude e-mail that is a member of two orgs (e.g. personal Max + Team).
Log both into CLIProxyAPI, so the hub lists two claude auth files with the same email.
Add the hub as a usage limit source in T3 Code.
Open Usage → Limits → Claude.
Only one bar appears for that e-mail; the hub panel shows two credentials with different plans, limits and reset times.
Version
0.1.8-nightly.20261109.4399 (desktop app, desktop-managed server). Triage CLI: 0.1.7. main at 965ddc1 has no changes to the files involved.
Environment
Windows 11 x64, T3 Code desktop nightly. Usage limit source: CLIProxyAPI hub with 5 Claude credentials (3 e-mails; 2 e-mails each in a personal Max 20x org and a shared Team org). Reproduces on every client connected to it.
Evidence
Hub panel (per credential, % left):
- A@ / Max 20x: 5h 89%, weekly 44%
- A@ / Team: 5h 7%, weekly 84%
- B@ / Team: 5h 100%, weekly 0%
- B@ / Max 20x: weekly 0%
- C@ / Max 20x: weekly 0%
T3 Limits → Claude → Weekly shows 3 bars: A@ 42% (Max reset), B@ 0% (Team reset), C@ 0%. Missing: A@ Team and B@ Max. Screenshots to be attached.
Related issues
#11515 (Codex multi-account via proxy, pre-hub), #23054 (Codex tile per environment), #13110 (pool averages ignore plan size), #28052 (closed; hub vs instance labelling).
Fix applied or workaround
None in T3 config. Read per-credential limits in the CLIProxyAPI panel.
Filed by
Claude Code (Claude Opus 5.5, claude-opus-5-5) via t3 triage.
An earlier PR already fixes this: #26634 (reads the organization from /api/oauth/profile through the hub and reports it as the workspace). Verified against the hub described above there. My duplicate #33354 is closed.
Independent confirmation on stable 0.1.8 (Linux, t3 background service), with a detail not covered above: the org that wins differs per server connected to the same hub.
Setup. One CLIProxyAPI hub with 5 Claude OAuth accounts. Two share an email but belong to different orgs: a personal Max 20x org (organization_type: claude_max) and a Team org (claude_team). Two T3 servers use this hub as a usage limit source. Server A also has a local Claude login. Server B also has a Claude provider instance that routes through the hub.
Evidence (checked 2026-11-11 ~04:30 UTC):
Through the hub's api-call, /api/oauth/profile returns a different org for each of the two credentials. /api/oauth/usage returns separate weekly windows: 16% used, resetting Sunday, vs. 89% used, resetting Friday.
Usage → Limits on server A shows one row for that email, with the personal org's numbers. Server B shows one row for the same email, with the Team org's numbers. All other accounts match on both servers.
In the 0.1.8 binary, CliproxyApi.readAccount still sets workspaceId only from id_token.chatgpt_account_id and never calls /api/oauth/profile. main is the same (apps/server/src/usage/cliproxyApi.ts:219). Both credentials therefore get the key claudeAgent:<email> in the client and collapse into one row. Which credential wins then depends on what else that server reports for the email, so each server can show a different org. This matches the PR fix(server): keep Claude hub logins that share an email separate #33602 description: hub accounts without a workspace fold into the local login's row.
Same cause and fix as above. The open keeper is #33602 (reads organization.uuid from /api/oauth/profile through the hub). #28142 is the earlier report of the same bug.
Filed by Claude Code (Claude Opus 5.5, claude-opus-5-5), following the t3 triage playbook (triage CLI 0.1.8).
Activity
lol2x commentedon Nov 9, 2026
An earlier PR already fixes this: #26634 (reads the organization from
/api/oauth/profilethrough the hub and reports it as the workspace). Verified against the hub described above there. My duplicate #33354 is closed.adambarito commentedon Nov 11, 2026
Independent confirmation on stable 0.1.8 (Linux,
t3background service), with a detail not covered above: the org that wins differs per server connected to the same hub.Setup. One CLIProxyAPI hub with 5 Claude OAuth accounts. Two share an email but belong to different orgs: a personal Max 20x org (
organization_type: claude_max) and a Team org (claude_team). Two T3 servers use this hub as a usage limit source. Server A also has a local Claude login. Server B also has a Claude provider instance that routes through the hub.Evidence (checked 2026-11-11 ~04:30 UTC):
api-call,/api/oauth/profilereturns a different org for each of the two credentials./api/oauth/usagereturns separate weekly windows: 16% used, resetting Sunday, vs. 89% used, resetting Friday.CliproxyApi.readAccountstill setsworkspaceIdonly fromid_token.chatgpt_account_idand never calls/api/oauth/profile.mainis the same (apps/server/src/usage/cliproxyApi.ts:219). Both credentials therefore get the keyclaudeAgent:<email>in the client and collapse into one row. Which credential wins then depends on what else that server reports for the email, so each server can show a different org. This matches the PR fix(server): keep Claude hub logins that share an email separate #33602 description: hub accounts without a workspace fold into the local login's row.Same cause and fix as above. The open keeper is #33602 (reads
organization.uuidfrom/api/oauth/profilethrough the hub). #28142 is the earlier report of the same bug.Filed by Claude Code (Claude Opus 5.5, claude-opus-5-5), following the
t3 triageplaybook (triage CLI 0.1.8).juliusmarminge commentedon Nov 12, 2026
This is the same problem as #28142, so I'm closing it in favour of that one. If it is different, say how and it will be reopened.