Anthropic

Setting up Anthropic Claude as a provider.

Get a key

https://console.anthropic.com/settings/keys → Create key.

Add the provider

Setyenv → Agent → Providers → Add Provider → Anthropic.

Required:

• API key: starts with sk-ant-.
• Default model: claude-sonnet-4 recommended (good cost/quality balance). claude-opus-4 for the deepest reasoning, claude-haiku-4-5 for the cheapest fast turns.

Pricing

Claude Sonnet 4 is $3/M input + $15/M output (last published numbers; check https://www.anthropic.com/pricing). For the same 15K + 1K turn shape, that's about $0.06 per turn.

Tool use

Fully supported with Anthropic's tools parameter. Models also support extended thinking — the model can reason at length before deciding what tool to call. The agent uses this when available, showing the reasoning trace in the Action Inspector.

Prompt caching

Anthropic supports marking parts of the prompt as cacheable; subsequent requests pay a small fraction of the normal input price on the cached part. The plugin opts the agent into this automatically on the parts that stay constant within a session. Cache lifetime is short (a few minutes); the agent schedules turns to keep the cache warm where it can.

Long context

Claude supports 200K-token (Sonnet 4) and 1M-token (Opus 4 1M) contexts. When a conversation approaches the model's context limit, the agent automatically summarises the older middle of the conversation to keep recent turns intact. See Conversation length in Settings.