Configure AI providers: OpenRouter, OpenAI, and Anthropic

Providers are defined under the top-level providers key in config.yml. Each entry is a named block that describes how MonoRelay connects to a specific upstream API — its base URL, authentication keys, timeout, and optional features like retry logic or request cloaking. You can enable or disable any provider at runtime; the configuration is hot-reloaded without a server restart.

Provider configuration tabs

OpenRouter
NVIDIA NIM
OpenAI
Anthropic
ChatGPT Web

OpenRouter aggregates hundreds of models behind a single OpenAI-compatible endpoint. It is the recommended default provider.

config.yml

providers:
  openrouter:
    enabled: true
    base_url: "https://openrouter.ai/api/v1"
    keys:
      - key: "sk-or-v1-your-openrouter-key"
        label: "main"
        weight: 1
      - key: "sk-or-v1-your-backup-key"
        label: "backup"
        weight: 1
    headers:
      HTTP-Referer: "https://github.com/Excurs1ons/MonoRelay"
      X-Title: "MonoRelay"
    rate_limit_cooldown: 60
    timeout: 120
    models:
      include: []
      exclude: []

NVIDIA NIM serves open-source and NVIDIA-tuned models at integrate.api.nvidia.com.

config.yml

providers:
  nvidia:
    enabled: true
    base_url: "https://integrate.api.nvidia.com/v1"
    keys:
      - key: "nvapi-your-nvidia-key"
        label: "main"
        weight: 1
    rate_limit_cooldown: 30
    timeout: 120
    models:
      include: []
      exclude: []

Direct OpenAI API access. Disable this if you route OpenAI-branded models through OpenRouter instead.

config.yml

providers:
  openai:
    enabled: false
    base_url: "https://api.openai.com/v1"
    keys:
      - key: "sk-your-openai-key"
        label: "main"
        weight: 1
    rate_limit_cooldown: 60
    timeout: 120

Direct Anthropic API access. Uses the native Anthropic endpoint rather than the OpenAI-compatible wrapper.

config.yml

providers:
  anthropic:
    enabled: false
    base_url: "https://api.anthropic.com"
    keys:
      - key: "sk-ant-your-anthropic-key"
        label: "main"
        weight: 1
    rate_limit_cooldown: 60
    timeout: 120

The chatgpt_web provider connects through ChatGPT’s web interface rather than the developer API. It requires an access token from an authenticated ChatGPT session.

config.yml

providers:
  chatgpt_web:
    enabled: false
    provider_type: "web_reverse"
    keys:
      - key: "your-chatgpt-access-token"
        label: "main"
        weight: 1
    rate_limit_cooldown: 120
    timeout: 120
    web_reverse:
      pow_difficulty: "00003a"
      conversation_only: false
      history_disabled: true
      proxy_url: ""
      chatgpt_base_url: "https://chatgpt.com"
      model_mapping:
        "gpt-3.5-turbo": "text-davinci-002-render-sha"
        "gpt-4o": "gpt-4o"
        "gpt-4o-mini": "gpt-4o-mini"

The ChatGPT web reverse proxy depends on the internal ChatGPT web API, which is undocumented and may break without notice after ChatGPT updates.

Common provider fields

Field	Type	Default	Description
`enabled`	boolean	`true`	Whether this provider is active
`base_url`	string	—	Upstream API base URL
`keys`	object[]	`[]`	List of API keys; see key fields
`headers`	object	`{}`	Extra HTTP headers sent with every request
`rate_limit_cooldown`	integer	`60`	Seconds a key stays on cooldown after a `429` response
`timeout`	integer	`120`	Request timeout in seconds
`models.include`	string[]	`[]`	Explicit allowlist of models; empty means all models
`models.exclude`	string[]	`[]`	Models to exclude from this provider

Retry configuration

Retry logic is disabled by default. Enable it per-provider to automatically retry failed requests with exponential backoff.

config.yml

providers:
  openrouter:
    retry:
      enabled: true
      max_retries: 3
      retry_on_status: [429, 500, 502, 503, 504]
      retry_on_errors:
        - "rate_limit_error"
        - "timeout_error"
        - "upstream_error"
        - "service_unavailable"
      backoff_factor: 1.5
      backoff_max: 60

boolean

default:"false"

Enable automatic request retries.

number

default:"3"

Maximum number of retry attempts per request.

number[]

default:"[429, 500, 502, 503, 504]"

HTTP status codes that trigger a retry.

string[]

Error type strings that trigger a retry (e.g., "rate_limit_error", "timeout_error").

number

default:"1.5"

Multiplier for exponential backoff delay between attempts.

number

default:"60"

Maximum backoff delay in seconds.

Ignore configuration

When certain upstream error types should be silently swallowed rather than surfaced to the client, use the ignore block.

config.yml

providers:
  openrouter:
    ignore:
      enabled: true
      ignore_errors:
        - "content_filter"
        - "content_moderation"
        - "policy_violation"

boolean

default:"false"

Enable error ignoring.

string[]

Error type strings to suppress. Requests that trigger these errors return an empty successful response.

Request cloaking

Cloaking lets you override HTTP headers so outbound requests appear to originate from a different client. This is most useful with the chatgpt_web provider or when a provider restricts access by User-Agent.

config.yml

providers:
  openrouter:
    cloaking:
      user_agent: "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36"
      referer: "https://chat.example.com"
      origin: "https://chat.example.com"
      accept: "application/json"
      accept_language: "en-US,en;q=0.9"

string

Override the User-Agent header sent to the upstream provider.

string

Override the Referer header.

string

Override the Origin header.

string

Override the Accept header.

string

Override the Accept-Language header.

ChatGPT web reverse-proxy fields

The web_reverse block is only used when provider_type: "web_reverse".

string

default:"\"00003a\""

Proof-of-work difficulty string required by the ChatGPT web gateway.

boolean

default:"false"

When true, only the conversation endpoint is used (no streaming backend).

boolean

default:"true"

Prevent conversations from appearing in the ChatGPT history sidebar.

string

default:"\"\""

Optional HTTP/SOCKS proxy URL for outbound ChatGPT web requests.

string

default:"\"https://chatgpt.com\""

Base URL for the ChatGPT web interface.

object

Maps OpenAI model names in the request to internal ChatGPT web model identifiers. For example, "gpt-3.5-turbo" maps to "text-davinci-002-render-sha".

Get Started

Configuration

Admin Dashboard

Authentication

Configure AI providers: OpenRouter, OpenAI, and Anthropic

Provider configuration tabs

Common provider fields

Retry configuration

Ignore configuration

Request cloaking

ChatGPT web reverse-proxy fields

​Provider configuration tabs

​Common provider fields

​Retry configuration

​Ignore configuration

​Request cloaking

​ChatGPT web reverse-proxy fields

Provider configuration tabs

Common provider fields

Retry configuration

Ignore configuration

Request cloaking

ChatGPT web reverse-proxy fields