Claude

Claude (Anthropic)

Claude is Ptah’s default reasoning provider. It supports the full feature set: thinking, prompt caching, vision, and sub-agent spawning.

What you need

• An Anthropic API key from the Anthropic Console.
• Claude model access enabled on your account.

Available models

Model	Good for	Context	Thinking
Claude Opus 4.7	The hardest reasoning tasks, long multi-file refactors.	200k	Yes
Claude Sonnet 4.6	Daily development work — best speed/quality balance.	200k	Yes
Claude Haiku 4.5	Quick edits, autocompletion, low-latency chat.	200k	Limited

Configuration

Open Settings → Providers → Claude in the app and paste your API key. The key is stored in encrypted safeStorage.

Non-secret Claude settings live in ~/.ptah/settings.json:

{
  "anthropicProviderId": "anthropic",
  "llm.defaultProvider": "claude"
}

Set anthropicProviderId to:

• "anthropic" — direct Anthropic API (requires your API key).
• "openrouter" — route Claude traffic through OpenRouter instead (requires an OpenRouter key — see OpenRouter).
• "openai-codex" / "github-copilot" — use Codex or Copilot’s upstream Claude access.

Verifying it works

1. Open the chat.
2. Select Claude Sonnet 4.6 in the model selector.
3. Send a simple prompt: What is 2 + 2?
4. You should see a streaming response and the cost bar should show input/output tokens and a non-zero USD amount.

Troubleshooting

• invalid x-api-key — the key is wrong or revoked. Regenerate it in the Anthropic Console and re-paste it in Settings.
• rate_limit_error — you’ve hit your tier’s RPM or TPM cap. Wait or upgrade your Anthropic plan.
• model_not_found — your account may not have access to the selected model. Check model permissions in the Anthropic Console.
• High cost but no output — check the Execution Tree for thinking-block token usage. Switch to a lower effort level to cap thinking budget.

Tip

Claude’s prompt caching can drop costs by ~80% on long conversations. Ptah enables it automatically — watch the cache read number in the cost bar rise across turns.