Claude Code

NuRoute speaks Anthropic's Messages API at POST /v1/messages, so Claude Code works with it unchanged. Its requests go through the same pipeline as /v1/chat/completions: your routing, policies, PII redaction, budgets, credits and plan apply.

Set up

export ANTHROPIC_BASE_URL=https://nuroute.ai
export ANTHROPIC_API_KEY=aicp-...        # your NuRoute API key (or ANTHROPIC_AUTH_TOKEN)
export ANTHROPIC_MODEL=auto              # Artemis Task picks the model for each request
export ANTHROPIC_DEFAULT_HAIKU_MODEL=auto # Claude Code's quick background calls (older versions: ANTHROPIC_SMALL_FAST_MODEL)
claude

ANTHROPIC_MODEL can be any model your key can route to, for example claude-sonnet-4-6 to always use it. auto is Artemis Task, billed per token from credits.

What works

Feature	Supported
Chat and streaming	✓
Tools (Claude Code's file, shell and search tools)	✓
Prompt caching	✓ where the serving provider caches; cached tokens show as cache_read_input_tokens
Model stickiness per conversation	✓ (Claude Code's metadata.user_id keeps each person's conversations apart)
Images and PDFs in prompts	✗ refused with a 400
Extended thinking	Ignored: the serving model decides
Anthropic server tools (web search, code execution)	✗ refused with a 400

A project-scoped key keeps Claude Code's requests in its project, so the project's routing, budget and policy apply.

Errors you may see

Status	Meaning
401 authentication_error	The key is wrong or revoked
402 billing_error	Your credit balance is used up; top up on the Billing page
429 rate_limit_error	Too many requests at once, or your plan's request limit is reached