Skip to main content

​
Overview

Nanobot supports multiple LLM providers through a unified configuration system. The provider registry automatically detects the correct provider based on model names, API keys, or explicit configuration.

​
ProvidersConfig

The root configuration object for all LLM providers.
​
custom
ProviderConfig
default:"ProviderConfig()"
Custom OpenAI-compatible endpoint (direct, bypasses LiteLLM).
​
anthropic
ProviderConfig
default:"ProviderConfig()"
Anthropic (Claude) provider configuration.
​
openai
ProviderConfig
default:"ProviderConfig()"
OpenAI (GPT) provider configuration.
​
openrouter
ProviderConfig
default:"ProviderConfig()"
OpenRouter gateway configuration.
​
deepseek
ProviderConfig
default:"ProviderConfig()"
DeepSeek provider configuration.
​
groq
ProviderConfig
default:"ProviderConfig()"
Groq provider configuration.
​
zhipu
ProviderConfig
default:"ProviderConfig()"
Zhipu AI (GLM) provider configuration.
​
dashscope
ProviderConfig
default:"ProviderConfig()"
DashScope (阿里云通义千问) provider configuration.
​
vllm
ProviderConfig
default:"ProviderConfig()"
vLLM local deployment configuration.
​
gemini
ProviderConfig
default:"ProviderConfig()"
Google Gemini provider configuration.
​
moonshot
ProviderConfig
default:"ProviderConfig()"
Moonshot (Kimi) provider configuration.
​
minimax
ProviderConfig
default:"ProviderConfig()"
MiniMax provider configuration.
​
aihubmix
ProviderConfig
default:"ProviderConfig()"
AiHubMix gateway configuration.
​
siliconflow
ProviderConfig
default:"ProviderConfig()"
SiliconFlow (硅基流动) gateway configuration.
​
volcengine
ProviderConfig
default:"ProviderConfig()"
VolcEngine (火山引擎) gateway configuration.
​
openaiCodex
ProviderConfig
default:"ProviderConfig()"
OpenAI Codex (OAuth) provider configuration.
​
githubCopilot
ProviderConfig
default:"ProviderConfig()"
GitHub Copilot (OAuth) provider configuration.

​
ProviderConfig

Individual provider configuration.
​
apiKey
str
default:""
API key for the provider. Can be set via environment variables.
​
apiBase
str | null
default:"null"
Custom API base URL. If not specified, uses provider’s default endpoint.
​
extraHeaders
dict[str, str] | null
default:"null"
Custom HTTP headers to include in requests (e.g., APP-Code for AiHubMix).

​
Provider Registry

The provider registry (PROVIDERS) defines metadata for each supported provider:

​
Provider Types

Standard Providers: Directly supported by LiteLLM
  • Anthropic (Claude)
  • OpenAI (GPT)
  • DeepSeek
  • Gemini
  • And more
Gateways: Route any model through a unified API
  • OpenRouter
  • AiHubMix
  • SiliconFlow
  • VolcEngine
Local Deployments: Self-hosted models
  • vLLM
  • Ollama (via vLLM config)
OAuth Providers: Use OAuth instead of API keys
  • OpenAI Codex
  • GitHub Copilot

​
ProviderSpec

Each provider is defined by a ProviderSpec with the following attributes:
​
name
str
Configuration field name (e.g., dashscope, anthropic).
​
keywords
tuple[str, ...]
Model name keywords for auto-detection (lowercase).
​
envKey
str
Environment variable name for API key (e.g., ANTHROPIC_API_KEY).
​
displayName
str
Human-readable name shown in nanobot status.
​
litellmPrefix
str
Prefix for LiteLLM routing (e.g., deepseek/ → deepseek/deepseek-chat).
​
skipPrefixes
tuple[str, ...]
Don’t add prefix if model already starts with these.
​
envExtras
tuple[tuple[str, str], ...]
Additional environment variables to set. Supports placeholders: {api_key}, {api_base}.
​
isGateway
bool
If true, provider can route any model (OpenRouter, AiHubMix).
​
isLocal
bool
If true, provider is a local deployment (vLLM).
​
detectByKeyPrefix
str
Auto-detect by API key prefix (e.g., sk-or- for OpenRouter).
​
detectByBaseKeyword
str
Auto-detect by substring in api_base URL.
​
defaultApiBase
str
Default API base URL if not specified in config.
​
stripModelPrefix
bool
If true, strip provider prefix before re-prefixing (AiHubMix behavior).
​
modelOverrides
tuple[tuple[str, dict], ...]
Per-model parameter overrides (e.g., Kimi K2.5 requires temperature: 1.0).
​
isOauth
bool
If true, uses OAuth instead of API keys (OpenAI Codex, GitHub Copilot).
​
isDirect
bool
If true, bypasses LiteLLM entirely (custom provider).
​
supportsPromptCaching
bool
If true, provider supports prompt caching (Anthropic, OpenRouter).

​
Provider Matching

Nanobot automatically selects the correct provider using this priority:
  1. Explicit provider prefix: github-copilot/claude → GitHub Copilot
  2. Model name keywords: claude-opus → Anthropic
  3. API key prefix: sk-or-xxx → OpenRouter
  4. API base URL: https://openrouter.ai/... → OpenRouter
  5. Forced provider: provider: anthropic in config
  6. Fallback: First available gateway with API key

​
Configuration Examples

​
Multiple Providers

providers:
  anthropic:
    apiKey: ${ANTHROPIC_API_KEY}
  
  openai:
    apiKey: ${OPENAI_API_KEY}
  
  openrouter:
    apiKey: ${OPENROUTER_API_KEY}
  
  deepseek:
    apiKey: ${DEEPSEEK_API_KEY}

​
Custom API Base

providers:
  anthropic:
    apiKey: sk-ant-xxx
    apiBase: https://api.anthropic.com/v1  # Custom endpoint
  
  vllm:
    apiKey: dummy  # Optional for local
    apiBase: http://localhost:8000/v1

​
Gateway with Custom Headers

providers:
  aihubmix:
    apiKey: ${AIHUBMIX_KEY}
    apiBase: https://aihubmix.com/v1
    extraHeaders:
      APP-Code: your-app-code

​
OAuth Providers

providers:
  githubCopilot:
    # No apiKey needed - uses OAuth flow
    apiBase: ""  # Auto-detected
  
  openaiCodex:
    # No apiKey needed - uses OAuth flow
    apiBase: https://chatgpt.com/backend-api

​
Environment Variables

Providers can be configured via environment variables: 
# Standard format
NANOBOT_PROVIDERS__ANTHROPIC__API_KEY=sk-ant-xxx
NANOBOT_PROVIDERS__OPENAI__API_KEY=sk-xxx

# With custom base
NANOBOT_PROVIDERS__VLLM__API_BASE=http://localhost:8000/v1

# Direct LiteLLM env vars (also supported)
ANTHROPIC_API_KEY=sk-ant-xxx
OPENAI_API_KEY=sk-xxx
DEEPSEEK_API_KEY=sk-xxx

​
Provider-Specific Notes

​
Anthropic

  • Supports prompt caching via cache_control
  • No prefix needed (claude-opus-4-5 works directly)

​
OpenRouter

  • Auto-detected by sk-or- key prefix
  • Routes any model through unified API
  • Supports prompt caching

​
DeepSeek

  • Requires deepseek/ prefix for LiteLLM
  • Models: deepseek-chat, deepseek-reasoner

​
Moonshot (Kimi)

  • Requires moonshot/ prefix
  • Kimi K2.5 enforces temperature >= 1.0
  • Default base: https://api.moonshot.ai/v1 (international)
  • Use https://api.moonshot.cn/v1 for China region

​
Zhipu AI

  • Uses zai/ prefix for LiteLLM
  • Also sets ZHIPUAI_API_KEY env var for compatibility

​
AiHubMix

  • Gateway that strips provider prefixes
  • Example: anthropic/claude-3 → claude-3 → openai/claude-3
  • Requires APP-Code header for some configurations

​
vLLM (Local)

  • For self-hosted models
  • Uses hosted_vllm/ prefix
  • Requires api_base configuration

​
GitHub Copilot

  • OAuth-based (no API key)
  • Uses github_copilot/ prefix
  • Models must be explicitly selected

​
Programmatic Access

from nanobot.config.schema import Config
from nanobot.providers.registry import find_by_model, find_gateway, PROVIDERS

# Load config
config = Config()

# Get provider for model
provider = config.get_provider("claude-opus-4-5")
print(provider.api_key, provider.api_base)

# Get provider name
name = config.get_provider_name("deepseek-chat")
print(name)  # "deepseek"

# Find provider spec by model
spec = find_by_model("claude-opus-4-5")
print(spec.name, spec.litellm_prefix)

# Find gateway by API key
gateway = find_gateway(api_key="sk-or-xxx")
print(gateway.name)  # "openrouter"

# List all providers
for spec in PROVIDERS:
    print(f"{spec.label}: {spec.keywords}")

Build docs developers (and LLMs) love

Get started for freeTalk to us