Reliability Dashboard

Both gateways · cost, fallback, every request

Open chat

Polls every 2.5s. This view — spend, request log, approving or rejecting tickets — is restricted to admins.

Checking sign-in… Log in

Admin access required

This dashboard — spend, request log, and the ability to approve or reject tickets — is restricted to admins.

Claude apps gateway governs who is allowed to call Claude for sensitive LLM Gateway actions — every request carries a real signed-in admin identity, with spend visibility and an audit trail.

Claude apps gateway

Identity-attributed & audited — account requests
Time Requested by Ticket Drafted action Status
No account requests yet.

LiteLLM gateway routes requests across model vendors — Claude, GPT, a local model — picking the right tier for the job and failing over automatically if a provider goes down.

LLM gateway (LiteLLM)

Cost tiers & fallback chains — routine questions

Spend vs budget

$0.000000 / $0.00

Fallback & retry events

● nominal
0

fallback events

0 total retries · 0 requests logged

Request log

Time Query Type Tier Model Latency Cost Status
No requests yet — send a message in the chat.

Role Defaults

Starting model access, budget, and limits applied automatically when granting access by role. Edit anytime — changes only apply to new grants, not existing users.

Role Can request Model access Budget cap Session limit Daily limit Monthly limit
Loading…
Edit role defaults

Users

Role, account ticket permissions, model access, and spend vs budget per user. Session limits reset when a user starts a new chat; daily limits reset at midnight UTC; monthly limits reset on the 1st of the month UTC.

cheap-tier = Claude Haiku, for short/routine questions · strong-tier = Claude Sonnet / GPT-4o, for long or complex questions.

Name Username Role Can request Model access Spend vs budget Session limit Daily limit Monthly limit
Loading…
Grant access

Model access, budget, and limits above are pre-filled from the selected role's defaults — adjust if this person needs something different.

Edit limits