How-To: Add Anthropic as an LLM Provider in Smarter

Goal

This tutorial shows how to add Anthropic as an LLM provider in Smarter using a provider manifest.

In this example, Anthropic is used because Claude Code is the intended downstream use case.

Prerequisites

This tutorial assumes you already have:

Administrative access to the Smarter environment
The Smarter CLI installed and available
Permission to apply manifests
An Anthropic API key
Basic familiarity with YAML and command-line tools

Setup

Before creating the provider manifest, make sure the Anthropic API key is available to Smarter as a secret or environment-level configuration.

See Anthropic’s documentation at Get started with Claude for current information.

Concept Overview

In Smarter, a Provider resource defines an upstream LLM provider that other Smarter resources can use.

For this tutorial:

Smarter acts as the internal platform
Anthropic is the upstream provider
Claude is the model family
The provider manifest defines how Smarter connects to Anthropic

Step-by-Step

Step 1: Generate a starting template

Run:

smarter manifest provider

Step 2: Create the provider manifest

Create a file named anthropic-provider.yaml:

apiVersion: smarter.sh/v1
kind: Provider
metadata:
  name: anthropic_provider
  description: Anthropic provider for Claude models
  version: 1.0.0
spec:
  provider:
    name: Anthropic
    description: Anthropic API provider
    base_url: https://api.anthropic.com/v1
    api_key: anthropic_api_key
    connectivity_test_path: /models

Step 3: Apply the manifest

Run:

smarter apply -f anthropic-provider.yaml

Step 4: Verify the provider

Run:

smarter describe provider anthropic_provider

Proof of Concept

A successful result is a provider resource that is active and available for later llm_client or agent configuration.

Troubleshooting

Provider does not apply

Check the YAML formatting and required fields.

Provider does not verify

Check that the Anthropic API key is available to Smarter and that the base URL and connectivity test path are correct.

Optional Next Step

After the provider is active, the next logical step would be to create an llm_client that uses it.

The following example shows the intended shape of an Anthropic llm_client manifest, based on the same style as the smarter-deploy OpenAI llm_client example. This is shown as a target-state example only.

Note

In the current Smarter repo, llm_client providers appear to be limited to OpenAI, Google AI, and Meta AI. Additional implementation would likely be required before an Anthropic llm_client could work end-to-end.

apiVersion: smarter.sh/v1
kind: LLMClient
metadata:
  description: "An Anthropic Claude pass-through llm_client."
  name: anthropic_claude
  version: 0.1.0
spec:
  apiKey: null
  config:
    appAssistant: Claude
    appBackgroundImageUrl: null
    appExamplePrompts:
      - Write a Python function that generates the Fibonacci sequence.
      - Summarize a technical document in 100 words or less.
      - Provide a technical definition for the term "machine learning."
    appFileAttachment: false
    appInfoUrl: https://www.anthropic.com/
    appLogoUrl: https://platform.smarter.sh/static/images/logo/smarter-crop.png
    appName: Anthropic Claude
    appPlaceholder: Ask me anything...
    appWelcomeMessage: Welcome to Claude!
    customDomain: null
    defaultMaxTokens: 4096
    defaultModel: claude-3-5-sonnet
    defaultSystemRole: You are a helpful llm_client.
    defaultTemperature: 0.5
    deployed: true
    provider: anthropic
    subdomain: null
  functions: []
  plugins: []

This step is outside the main scope of this tutorial. The main objective here is adding the provider itself.