ContextQMD
Back to Eliza

Anthropic Plugin

packages/docs/plugin-registry/llm/anthropic.md

2.0.13.1 KB
Original Source

The Anthropic plugin connects Eliza agents to Anthropic's Claude API and exposes the current Claude Opus 4.7, Claude Sonnet 4.6, and Claude Haiku 4.5 models.

Package: @elizaos/plugin-anthropic

Installation

bash
eliza plugins install @elizaos/plugin-anthropic

Auto-Enable

The plugin auto-enables when ANTHROPIC_API_KEY or CLAUDE_API_KEY is present:

bash
export ANTHROPIC_API_KEY=sk-ant-...

Configuration

Environment VariableRequiredDescription
ANTHROPIC_API_KEYYesAnthropic API key from console.anthropic.com
CLAUDE_API_KEYNoAlias that also triggers auto-enable (resolved to ANTHROPIC_API_KEY internally)
ANTHROPIC_SMALL_MODELNoOverride the small model identifier (default: claude-haiku-4-5-20251001-5-20251001)
ANTHROPIC_LARGE_MODELNoOverride the large model identifier (default: claude-sonnet-4-6)
ANTHROPIC_EXPERIMENTAL_TELEMETRYNoEnable experimental telemetry features for debugging and usage analytics (default: false)
ANTHROPIC_BROWSER_BASE_URLNoBrowser-only proxy endpoint base URL for Anthropic requests (no secrets in the client)

eliza.json Example

json
{
  "auth": {
    "profiles": {
      "default": {
        "provider": "anthropic",
        "model": "claude-sonnet-4-6"
      }
    }
  }
}

Supported Models

ModelContextBest For
claude-opus-4-7200kMost capable model for complex reasoning and long-running agents
claude-sonnet-4-6200kDefault large model for coding, analysis, and general use
claude-haiku-4-5-20251001200kFast, lightweight tasks

Model Type Mapping

elizaOS Model TypeAnthropic Model
TEXT_SMALLclaude-haiku-4-5-20251001
TEXT_LARGEclaude-sonnet-4-6

Features

  • Streaming responses
  • Tool use (function calling)
  • Vision (image input on all models)
  • Adaptive/extended thinking on claude-sonnet-4-6 and claude-opus-4-7
  • Structured JSON output via tool use
  • 200k token context window on all models
  • Prompt caching for cost reduction on repeated context

Extended Thinking

Claude Sonnet 4.6 and Claude Opus 4.7 support Anthropic's adaptive/extended thinking modes for complex reasoning and multi-step planning.

typescript
const response = await runtime.useModel("TEXT_LARGE", {
  prompt: "Design a database schema for a multi-tenant SaaS application.",
  thinking: { type: "enabled", budgetTokens: 10000 },
});

Rate Limits and Pricing

Rate limits depend on your Anthropic usage tier. See docs.anthropic.com/en/api/rate-limits for current limits.

Pricing: anthropic.com/pricing