Dashboard
// Welcome back
Endpoints
Primary API (OpenAI-compatible):
…
Balance (USD)
…
Requests (30d)
…
Tokens (30d)
Quick start
Replace
$KEY with your token above. Model IDs are on the Models page.My tokens
// These tokens are yours only · keep them secret
New token
Token list
| id | Name | Token | On | Usage | Rate limit | Expires | Created | |
|---|---|---|---|---|---|---|---|---|
| Loading… | ||||||||
Models
// Models currently routable
Shows models with at least one active key and not marked unavailable. Reference pricing and context from models.dev(USD / 1M tokens); not necessarily what you pay here.
Available models
| Model ID | Provider | Context | Max output | Ref. price $/1M in/out |
|---|---|---|---|---|
| Loading… | ||||
Model leaderboard
// Compare real usage to pick the right model
Last 7 days of real requests. Latency and avg output use successful calls only; price is what you pay (discount included). Click headers to sort: lower latency = faster, lower price = cheaper.
Model comparison
| Models | Provider | Latency ▲ | Success rate | Price $/1M in/out | Avg output | Calls (7d) |
|---|---|---|---|---|---|---|
| Loading… | ||||||
Latency: time until the response starts (TTFT for streams, full round-trip otherwise). Price = USD per 1M tokens (discount included), sorted by output price. More samples = more reliable; models with no successes show – for latency.
Docs
// OpenAI / Anthropic compatible · use the same SDKs
API & auth
Chat Completions endpoint:
Auth header:
Authorization: Bearer <your token>
curl
OpenAI SDK
Anthropic compatible (Claude SDK)
This gateway implements Anthropic Messages; Claude SDK / Claude Code can connect directly:
POST
Auth:
x-api-key: <your token> or Authorization: Bearer <your token>. Use the Base URL above for base_url; model IDs are the same as the OpenAI API.Request body is Anthropic format:
messages + max_tokens (required). Responses include X-KeyPool-Provider / X-KeyPool-Model (the upstream model that answered) / X-KeyPool-Requested-Model / X-KeyPool-Fallback (model may differ after fallback).Model ID
| Model ID | Provider | Context | Max output | Ref. price $/1M in/out |
|---|---|---|---|---|
| Loading… | ||||
POST /v1/chat/completions
Base URL (use for OpenAI
baseURL):Request body (OpenAI-compatible; supports
stream:true SSE):By default the gateway falls back to a similar model if yours is unavailable. Set
"fallback": false to disable fallback and fail if the model is unavailable.Response headers
| Response headers | Meaning |
|---|---|
| X-KeyPool-Provider | Provider actually used (e.g. gemini) |
| X-KeyPool-Model | Upstream model ID that actually answered (for free or value, the real model, not the alias; may differ after fallback) |
| X-KeyPool-Requested-Model | Model ID the request asked for (e.g. free or value) |
| X-KeyPool-Fallback | Present (1) when a hop or provider after the first choice answered |
Other endpoints
GET /v1/models — lists routable models (active keys, not marked unavailable). When models.dev data is present, each entry includes pricing (prompt / completion / cache_read / cache_write, USD / 1M tokens; see pricing_unit), context_window, max_output, modalities.GET
Native passthrough (provider-native protocols; gateway picks keys and bills):
/gemini/* · /mistral/* · /openrouter/*
Append the provider path after the prefix, e.g.
/gemini/v1beta/models/...:generateContent. Auth: Authorization: Bearer <your token>.Error codes
Errors are OpenAI-shaped; message is in
body.error.message:| Status | Meaning / what to do |
|---|---|
| 402 | Insufficient balance — top up and retry |
| 429 | Rate limited / quota — back off and retry |
| 4xx | Client error (params / model id / auth) |
| 5xx | Upstream error — retry later |
Billing
Billed per token: input and output settled separately; prices are in admin Model prices. Requests fail when balance is insufficient. See Usage and Balance for history.
Overview
// Key pool health
Pool status
Active
–
Cooldown
–
Disabled
–
| Provider | Active | Cooldown | Disable |
|---|---|---|---|
| Loading… | |||
Console access
Loading…
Devin quota
Loading…
Bifrost migration
bifrost.rowe.lol is mounted on this Worker (custom-domain route in wrangler.toml). It is served by the same Worker and binds the same D1 store as ai.rowe.lol, so its usage, log and alias counts show up in this console.Key pool
// Import to enable · bad keys auto-disabled
Import keys
Supported:
gemini openai mistral openrouter deepseek groq moonshot glm qwen deepinfra fireworks cerebras ai21 cohere elevenlabs stability. Duplicates are skipped; bad keys are disabled on first failed use.Status by provider
| Provider | Active | Cooldown | Disable |
|---|---|---|---|
| Loading… | |||
Need per-key detail?
Model availability
Loading…
Probe each model once; unavailable ones (billing / retired) are greyed out and omitted from
/v1/models.Check all keys · one batch
Runs
POST /admin/check-all-keys: live probes of up to 48 due keys, with the same revive and disable rules as the health check. To sweep the whole pool, use Key list → Check all.Key list
// Per-key view · probe / balance (OpenRouter)
All keys
| Provider | key | Req / fail | Last error | Probe / balance | Added | Actions |
|---|---|---|---|---|---|---|
| Loading… | ||||||
Probe checks live: OpenRouter shows real balance; Gemini/Mistral only report up/down (rate limits count as up). A successful probe can revive disabled/cooldown keys.
Bulk rotate · one provider
Preview lists the rows that would change and each new key's length, never the key itself. Confirm stays disabled until the counts match. Keys are cleared after submit.
Users
// Approval workflow · pending users first
User list
| Role | Status | Created | ||
|---|---|---|---|---|
| Loading… | ||||
Token
// Global tokens · including admin-issued
Issue token
Shown once — copy and store it now
Token list
| id | Name | Role | Owner | Usage | RPM | Expires | On | Created | Actions |
|---|---|---|---|---|---|---|---|---|---|
| Loading… | |||||||||
Model manager
// Visibility overrides · value and free are locked
Models
| Model id | Provider | Visible | Source | Reason | Action |
|---|---|---|---|---|---|
| Loading… | |||||
Hide and Show write an override (
PUT /admin/models/:id/visibility). Clear sends null and returns the model to its rule or default. value and free cannot be hidden.Free order
// Drag to pin the failover order · Save applies it
Pinned order
Drag a row by its handle, or focus a row and press
Alt+↑ / ↓. Pinned hops run first, in this order. Hops that are not pinned follow below in their Terminal-Bench order.Then, not pinned (Terminal-Bench order):
Free lineup (daily)
When
free_lineup_mode is daily, a scheduled job reorders only unpinned hops by Terminal-Bench. The pinned prefix in free_pinned_order is never moved.Value order
// Drag to reorder the value fallback steps · Save applies it
Fallback steps
The
value alias always tries OpenCode Go first, with the value_model target below. That step is not reorderable.1OpenCode Go always first
Then, in this order. Drag a row by its handle, or focus a row and press
Alt+↑ / ↓. Each step runs only when the steps before it failed.Settings
// Stored in D1 · the server validates every save
Trusted IPv4
Default:
Trusted IPv6 /64 prefixes
Default:
Value model
Default:
Hidden name needles
Default: ·
Chat
// Chat with any model · billed per token
Balance – USD
Start a conversation…
Debug chat
// Chat through one key · no pool · not billed
Pick a key to test…
Usage
// Last 30 days · your traffic
Total requests
–
Success rate
–
Total tokens
–
Requests (14 days) OK Failed
By provider
| Provider | Requests | OK | Avg latency | token |
|---|---|---|---|---|
| Loading… | ||||
Recent requests
| Time | Provider | Models | Status | Latency | token | session |
|---|---|---|---|---|---|---|
| Loading… | ||||||
Logs
// Last 30 days · all traffic
Total requests
–
Success rate
–
Total tokens
–
Requests (14 days) OK Failed
By provider
| Provider | Requests | OK | Avg latency | token |
|---|---|---|---|---|
| Loading… | ||||
Top users · 30 days
| # | Users | Requests | Success rate | token | Last |
|---|---|---|---|---|---|
| Loading… | |||||
Click a row to filter logs to that user.
Recent requests Page 1
| Time | Users | Provider | Models | Status | Latency | token | session |
|---|---|---|---|---|---|---|---|
| Loading… | |||||||
Billing
// Balances · top-ups · ledger · pricing
Top up
Balance overview
Total balance USD
–
Users
–
Users with balance
–
| sub | Balance USD | ||
|---|---|---|---|
| Loading… | |||
Recent transactions Up to 100 most recent
| Time | Users | Type | Amount USD | Balance USD | Models | token | Note |
|---|---|---|---|---|---|---|---|
| Loading… | |||||||
Model prices
| Models | Input $/Mtok | Output $/Mtok | |
|---|---|---|---|
| Loading… | |||
Stored as micro-USD per Mtok; input and output billed separately.
Balance
// Your balance and ledger
Current balance
Balance USD
–
Top up
Pay online via Stripe. Balance credits after payment confirms.
Recent transactions
| Time | Type | Amount USD | Balance USD | Models | token | Note |
|---|---|---|---|---|---|---|
| Loading… | ||||||