Newton’sPrompt Cache Pass-Through
Terminal agents repeatedly send recent shell history and codebase trees. Newton’s passes prompt-cache pricing straight through: on Claude Sonnet 5.5, cached input costs $0.01/MTok versus $0.20/MTok fresh — up to 95% less.
TERMINAL AGENT GUIDE
Run Claude Code CLI with
Newton’s Developer Gateway.
Claude Code is Anthropic’s high-power terminal agent, but scanning full repositories consumes millions of context tokens. Route Claude Code through Newton’s to maximize prompt cache hits, track exact terminal expenditure, and avoid surprise billing cliffs.
Built for Context-Heavy Terminal Agents
Why terminal workflows save the most when routed through Newton’s.
No Accidental Overages
Claude Code running automated test loops can consume $50+ in minutes. Newton’s hard prepaid balance ensures you never get billed more than you funded.
Verified Route Catalog
Access live, verified routes in the Newton’s model catalog — all 9 Claude models (Opus, Sonnet, and Haiku), live today — under one prepaid key.
Launch Claude Code in 2 Commands
Configure your terminal session or shell profile (~/.zshrc / ~/.bashrc). The 9 Claude models in /models are live now.
# 1. Export Newton's gateway variables:
export ANTHROPIC_BASE_URL="https://newtons.dev"
export ANTHROPIC_AUTH_TOKEN="newt_your_api_key_here"
export ANTHROPIC_MODEL="claude-sonnet-5-5"
export ANTHROPIC_SMALL_FAST_MODEL="claude-haiku-4-5-20251001"
# 2. Launch Claude Code in any repository:
claude
# Or run one-shot tasks with direct execution:
claude "Analyze src/routes and optimize bundle size"
Crashing into Claude Code 429 Rate Limits?
Learn why multi-turn repository indexing blows through Anthropic ITPM tiers and how Newton’s pooled multi-supplier routing and 85% prompt caching resolves it.
Read Claude Code 429 Troubleshooting Guide →Generate Optimized CLAUDE.md
Give Claude Code precise project instructions, test commands, and architectural constraints to cut wasted roundtrips.