Anthropic API (direct)

The simplest way to use Claude — direct API key from console.anthropic.com. Use this when you don't have AWS Bedrock available and want to call Claude directly.

For enterprise / regulated workloads, prefer Bedrock (data stays in your AWS account) or Vertex AI Anthropic if you're already on GCP.

Configuration

~/.insightworker/.env:

LLM_PROVIDER=anthropic
ANTHROPIC_API_KEY=sk-ant-...
AGENT_MODEL=claude-sonnet-4-5

Get an API key

Visit console.anthropic.com/settings/keys
Sign in / create an account
Create Key — name it insightworker
Copy the key (starts with sk-ant-)

Recommended models

Use case	Model
General apps, tool use	`claude-sonnet-4-5`
Heavy reasoning, complex translations	`claude-sonnet-4-5` (1M context variant)
Cheap classification	`claude-haiku-4-5`

Where data goes

Anthropic's standard API: prompts go to Anthropic's cloud, no training on customer data by default for API customers, retention policy at anthropic.com/legal/aup.

If your security review requires prompts stay in a specific cloud tenant, use Bedrock (AWS) or Vertex AI (GCP) — same models, your tenant.

Context windows

Model	Context
Sonnet 4.5 / 4.6	200k
Opus 4.1	200k
Opus 4.7 (1M variant)	1M
Haiku 4.5	200k

Common gotchas

Symptom	Cause	Fix
401 Unauthorized	Bad key	Regenerate at console.anthropic.com
429 rate-limited	Hit your usage tier ceiling	Wait, or upgrade tier in the Anthropic console
400 input too long	Hit the context window	Use a longer-context model or trim history