Anthropic AI Provider Plugin

The Anthropic plugin connects Ever Works to Anthropic's Claude API. Claude is recognized for producing well-structured, nuanced content and adhering closely to instructions, making it particularly suited for work descriptions and detailed content generation.

Source: packages/plugins/anthropic/src/anthropic.plugin.ts

Overview

Property	Value
Plugin ID	anthropic
Category	ai-provider
Capabilities	ai-provider
Version	1.0.0
Configuration Mode	user-required
Provider Type	anthropic
Auto-enable	No
Visibility	public

Like the OpenAI plugin, the Anthropic plugin extends BaseAiProvider and uses AiOperations for all AI interactions. The AiOperations layer abstracts away the differences between providers using LangChain.

Architecture

Configuration

Settings Schema

Setting	Type	Required	Default	Scope	Widget	Description
apiKey	string	Yes	--	user	--	Anthropic API key. Secret.
defaultModel	string	Yes	claude-sonnet-4-5-20250514	global	model-select	Default model for all AI tasks.
simpleModel	string	No	claude-haiku-4-5-20251001	global	model-select	Model for tags, descriptions, classifications.
mediumModel	string	No	claude-sonnet-4-5-20250929	global	model-select	Model for listings, summaries, reformatting.
complexModel	string	No	claude-sonnet-4-5-20250514	global	model-select	Model for full page generation, multi-step analysis.
baseUrl	string	No	https://api.anthropic.com/v1/	--	--	Custom API endpoint for proxies. Hidden.
temperature	number	No	0.7	--	--	Sampling temperature (0--2). Hidden.
maxTokens	number	No	4096	--	--	Max tokens per response. Hidden.

Model Tiers

The Anthropic plugin maps Claude model families to task complexity tiers:

Tier	Default Model	Use Case
Simple	Claude 4.5 Haiku	Tags, short descriptions, quick classifications
Standard	Claude 4.5 Sonnet	Listings, summaries, content reformatting
Complex	Claude 4.5 Sonnet	Full page generation, multi-step analysis

Capabilities

getCapabilities(): AiModelCapabilities {
  return {
    supportsStructuredOutput: true,
    supportsStreaming: true,
    supportsToolCalling: true,
    supportsVision: true,
    maxContextLength: 200000
  };
}

Capability	Supported	Description
Structured Output	Yes	JSON mode and tool use for typed responses
Streaming	Yes	Server-sent events for real-time token delivery
Tool Calling	Yes	Tool use for agent-based workflows
Vision	Yes	Image understanding with Claude's vision capabilities
Max Context	200,000	Maximum tokens in a single context window

The 200K context window is the largest among the platform's AI providers, allowing Claude to process significantly more source material per request.

Core Methods

Chat Completion

async createChatCompletion(options: ChatCompletionOptions): Promise<ChatCompletionResponse>

Routes through AiOperations with resolved configuration. The LangChain abstraction translates the standard ChatCompletionOptions format to Anthropic's Messages API format.

Streaming Chat Completion

async *createStreamingChatCompletion(options: ChatCompletionOptions): AsyncIterable<ChatCompletionChunk>

Yields ChatCompletionChunk objects as Claude generates tokens. Used by the conversational AI assistant.

Embeddings

async createEmbedding(_options: EmbeddingOptions): Promise<EmbeddingResponse> {
  throw new Error('Embeddings not supported by Anthropic');
}

Anthropic does not offer an embedding API. If embeddings are needed, use the OpenAI plugin or another provider that supports them alongside the Anthropic plugin for generation.

Model Listing

async listModels(settings?: PluginSettings): Promise<readonly AiModel[]>

Fetches the available Claude models from the API. Used to populate model selection in the CLI and web dashboard.

Availability Check

async isAvailable(settings?: PluginSettings): Promise<boolean>

Tests the connection by calling aiOps.testConnection() with the resolved configuration.

Lifecycle

Method	Behavior
onLoad(context)	Calls super.onLoad(), creates AiOperations with Anthropic defaults.
onUnload()	Sets aiOps to null, calls super.onUnload().
healthCheck()	Returns healthy.

Initial Configuration

this.aiOps = new AiOperations({
	apiKey: '',
	model: 'claude-sonnet-4-5-20250514',
	baseURL: 'https://api.anthropic.com/v1/',
	temperature: 0.7,
	maxTokens: 4096,
	providerType: 'anthropic'
});

The providerType: 'anthropic' tells AiOperations (and the underlying LangChain layer) to use the Anthropic message format and API conventions.

Error Handling

Scenario	Behavior
Plugin not loaded	Throws Error('Anthropic plugin not loaded')
Embedding request	Throws Error('Embeddings not supported by Anthropic')
API errors	Propagated from AiOperations / LangChain
Connection test failure	isAvailable() returns false

Differences from OpenAI Plugin

Aspect	Anthropic	OpenAI
Context window	200K tokens	128K tokens
Embeddings	Not supported	Supported
Default model	Claude Sonnet 4.5	GPT-5.1
Base URL	api.anthropic.com/v1/	api.openai.com/v1
Provider type	anthropic	openai
API format	Messages API	Chat Completions API

Both plugins share the same BaseAiProvider base class and AiOperations abstraction, making them interchangeable from the platform's perspective. The AiFacadeService can switch between them based on user configuration.

Usage in the Platform

When Anthropic is selected as the AI provider:

1. Content generation -- Claude generates work item descriptions, summaries, and metadata. Its instruction-following precision is useful for maintaining consistent formatting.
2. Conversational AI -- Powers the chat assistant in the web dashboard.
3. Structured extraction -- Uses tool calling to extract structured data from web content.

Choosing Between AI Providers

Consider Anthropic when...	Consider OpenAI when...
You need a large context window (200K)	You need embedding support
You value precise instruction following	You want the latest GPT models
You prefer Claude's writing style	You need multi-modal capabilities beyond vision
You want consistent formatting output	You want a wider selection of model sizes