omni-anthropic

README

OmniLLM Provider for Anthropic

Thick provider for OmniLLM using the official anthropic-sdk-go SDK.

Installation

go get github.com/plexusone/omnillm-anthropic

Quick Start

import (
    omnillm "github.com/plexusone/omnillm-core"
    _ "github.com/plexusone/omnillm-anthropic" // Auto-registers thick provider
)

client, _ := omnillm.NewClient(omnillm.ClientConfig{
    Provider: omnillm.ProviderNameAnthropic,
    APIKey:   os.Getenv("ANTHROPIC_API_KEY"),
})

Feature Support

Feature	Supported
Chat Completion	Yes
Streaming	Yes
Tool Calling	Yes
System Messages	Yes
Extended Thinking	Yes
JSON Mode	No

Extended Thinking

Anthropic Claude models support extended thinking for complex reasoning tasks:

import omnillm "github.com/plexusone/omnillm-core"

// Using Anthropic-style Thinking config (native)
budget := int64(8192)
response, err := client.CreateChatCompletion(ctx, &omnillm.ChatCompletionRequest{
    Model: "claude-sonnet-4-20250514",
    Thinking: &omnillm.ThinkingConfig{
        Type:         omnillm.ThinkingTypeEnabled,
        BudgetTokens: &budget,
    },
    Messages: messages,
})

// Or use ReasoningEffort for cross-provider compatibility
effort := omnillm.ReasoningEffortHigh
response, err := client.CreateChatCompletion(ctx, &omnillm.ChatCompletionRequest{
    Model:           "claude-sonnet-4-20250514",
    ReasoningEffort: &effort,
    Messages:        messages,
})

ThinkingType Values

Value	Description
omnillm.ThinkingTypeEnabled	Enable thinking with explicit budget
omnillm.ThinkingTypeDisabled	Disable thinking
omnillm.ThinkingTypeAdaptive	Let the model decide

ReasoningEffort Mapping

ReasoningEffort	Anthropic Behavior
"none"	Thinking disabled
"low", "medium"	Adaptive thinking
"high"	Enabled with budget from MaxTokens

See Reasoning Feature Guide for details.

Configuration

Field	Required	Description
APIKey	Yes	Anthropic API key
BaseURL	No	Custom endpoint

Documentation

See OmniLLM Core for full API documentation.

License

MIT