Skip to content

providers.md

Latest commit

 

History

History
186 lines (132 loc) · 4.23 KB

providers.md

File metadata and controls

186 lines (132 loc) · 4.23 KB

Providers & Authentication

Supported Providers

Provider Auth Method Notes
Anthropic API Key / OAuth Prompt caching, streaming, vision
OpenAI API Key Compatible with DeepSeek, Qwen, Ollama, vLLM
AWS Bedrock SigV4 Regional endpoints, AWS credential chain
Google Vertex AI GCP OAuth2 / Service Account Metadata server auto-detection

Custom Provider Alias

如果你的后端兼容某个内置 provider 的协议，可以给它定义一个自定义 alias，而不是把 provider 直接写成内置名字。

[default]
provider = "my-service"

[providers.my-service]
provider = "openai"
model = "custom-model-v1"
api_key = "sk-xxx"
base_url = "https://my-service.example.com/api/openai"

规则：

  • provider = "my-service" 是配置层 alias
  • [providers.my-service].provider 必须指向底层内置 provider
  • 底层 provider 目前只能是 anthropic、openai、bedrock、vertex
  • alias 条目的 model、api_key、base_url、compat 会覆盖底层 provider 的默认配置

这适合 DeepSeek 网关、内部 OpenAI-compatible 服务这类场景。

Profile Inheritance

Profiles support extends to inherit settings from another profile, avoiding duplication.

Configuration

# Base profile
[profiles.base-anthropic]
provider = "anthropic"
api_key = "sk-ant-xxx"

# Inherits base-anthropic, overrides model
[profiles.claude-fast]
extends = "base-anthropic"
model = "claude-haiku-4-5-20251001"
max_tokens = 4096

[profiles.claude-deep]
extends = "base-anthropic"
model = "claude-opus-4-20250514"
max_tokens = 16384

# Profile can specify which MCP servers to use
[profiles.dev]
extends = "base-anthropic"
model = "claude-sonnet-4-20250514"
mcp_servers = ["filesystem", "github"]

Usage

aionrs --profile claude-fast "Quick question"
aionrs --profile claude-deep "Deep security audit"
aionrs --profile dev "Create a GitHub issue"
  • Supports multi-level inheritance chains
  • Auto-detects circular inheritance
  • Child profile settings override parent

AWS Bedrock

Access Claude models via AWS Bedrock with SigV4 authentication.

Configuration

[default]
provider = "bedrock"

[bedrock]
region = "us-east-1"
# Option 1: Explicit credentials
access_key_id = "AKIA..."
secret_access_key = "..."
# session_token = "..."

# Option 2: AWS profile
# profile = "my-profile"

# Option 3: Environment variables (AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY)
# Used automatically when no credentials are configured

[profiles.bedrock-claude]
provider = "bedrock"
model = "anthropic.claude-sonnet-4-20250514-v1:0"

Credential Priority

  1. Explicit credentials in config file
  2. AWS profile
  3. Environment variables (AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY, AWS_SESSION_TOKEN)

Google Vertex AI

Access Claude models via Google Vertex AI with GCP OAuth2 authentication.

Configuration

[default]
provider = "vertex"

[vertex]
project_id = "my-gcp-project"
region = "us-central1"

# Option 1: Service Account key file
credentials_file = "/path/to/service-account.json"

# Option 2: Application Default Credentials
# Run: gcloud auth application-default login

# Option 3: Metadata Server (auto on GCE/GKE/Cloud Run)
# Used automatically when in GCP environments

[profiles.vertex-claude]
provider = "vertex"
model = "claude-sonnet-4@20250514"

Auth Methods

Method Use Case
Service Account Key CI/CD, server-side apps
Application Default Credentials Local development (requires gcloud CLI)
Metadata Server GCE/GKE/Cloud Run and other GCP environments

OAuth Login (Claude.ai)

Use your Claude.ai subscription (Pro/Team/Enterprise) directly — no API key needed.

Login

aionrs --login
  1. Displays an authorization URL and code
  2. Open the URL in your browser and enter the code
  3. Credentials are saved alongside the global config (run aionrs --config-path to find the directory)
  4. Subsequent runs auto-load saved credentials (with auto-refresh)

Logout

aionrs --logout

Configuring OAuth Endpoints

[auth]
auth_url = "https://claude.ai/oauth"
token_url = "https://claude.ai/oauth/token"
client_id = "aionrs"