Best iTerm2 Setup for Lower-Cost Claude Code
Claude Code can become expensive when every task uses a large model, a long conversation, and an oversized project context. A better setup combines sensible iTerm2 profiles, explicit model selection, short-lived sessions, and a low-cost API relay. This guide shows a practical configuration using 59API, whose API base URL is https://api.59api.com and which supports Claude Code with official-quality Claude models.
1. Create a dedicated iTerm2 profile
Open iTerm2, go to Settings, Profiles, and create a profile named “Claude Code”. Give it a distinct tab color so you can immediately recognize an AI-assisted terminal. Set the working directory to your main development folder, or leave it dynamic if you work across many repositories.
Use a readable font at 13–14 points and increase the scrollback buffer to at least 10,000 lines. This makes it easier to inspect tool output without rerunning commands. Avoid enabling automatic session restoration for sensitive projects if your machine is shared.
2. Route Claude Code through 59API
In the Claude Code profile, add environment variables under Profiles, General, Environment. Use the credentials shown in your 59API dashboard. A standard Claude Code configuration is:
ANTHROPIC_BASE_URL=https://api.59api.com
ANTHROPIC_AUTH_TOKEN=YOUR_59API_KEY
If your 59API account specifies ANTHROPIC_API_KEY instead, use that variable exactly as documented by the service. Keeping the key in the iTerm2 profile prevents it from being committed to a project’s .env file. For additional protection, restrict the key in your 59API account and rotate it if it appears in shell history or logs.
3. Use the smallest capable Claude model
Model choice is the largest controllable cost. Use Sonnet for normal implementation, debugging, refactoring, and code review. Use Haiku for quick explanations, small edits, test generation, and repetitive file changes. Reserve Opus for difficult architecture decisions, unfamiliar legacy systems, or problems where several failed attempts would cost more than a premium request. If your 59API account exposes Fable, test it on suitable workloads and compare output quality before making it your default.
Start a session with an explicit model:
claude --model sonnet
For a short utility task, use:
claude --model haiku
You can also switch models inside Claude Code with its model selection command. Make the model visible in your shell prompt or tab title when running multiple sessions, because accidentally using Opus for routine edits is an easy source of unnecessary spend.
4. Control context before it controls your bill
Token usage grows when a session repeatedly carries large files, command output, and old discussion. Keep each session focused on one change. Before asking for implementation, name the target files, acceptance criteria, and test command. Do not paste an entire log when the last 50 relevant lines are enough.
- Start a new session after completing a feature or fixing a separate issue.
- Ask Claude Code to inspect only the relevant directory or files.
- Use compacting when the conversation becomes long, ideally before it reaches roughly 60–70% of the available context.
- Exclude generated folders such as node_modules, dist, build, coverage, and large fixture data from routine investigation.
A useful target is to keep routine requests below 20,000 input tokens and reserve 50,000-token investigations for genuinely complex work. Reducing a repeated task from 40,000 input tokens to 20,000 cuts that portion of usage by 50%, regardless of the provider’s exact rate.
5. Add simple cost guardrails
Set a monthly 59API budget that matches your workload. For example, begin with $20, review usage after one week, and set a personal alert at $15. A developer making 10 tasks per day for 20 working days should track roughly 200 tasks monthly. If the average task consumes 30,000 input tokens and 5,000 output tokens, that is approximately 6 million input tokens and 1 million output tokens, which gives you a concrete baseline for comparing model rates.
59API is useful here because it provides pay-as-you-go access to Claude Opus, Sonnet, Haiku, and other supported models without forcing a subscription. Its low relay pricing, native model quality, and referral rebate can reduce the cost of regular Claude Code usage while preserving the same workflow. Sign up through 59API, create a restricted API key, and start with Sonnet plus a small monthly limit before expanding your setup.