One terminal, two trust levels — running Claude Code against a real subscription and a cheap proxy
Part of an ongoing series on model routing and trust tiering for agentic coding tools. This one's the boring, working half — no bug hunt, just a setup that's been running clean across two machines. The problem Claude Code does one thing well: careful, scoped edits with a real plan-then-execute loop behind them, backed by a subscription you're already paying for. Not every task needs that.…
This article describes a setup for running Claude Code, a coding tool, on two different trust levels using a real subscription and a cheaper proxy. The main challenge was that Claude Code only communicates with Anthropic's Messages API, and it doesn't have a built-in way to differentiate between the same tool but different models.
To address this, a self-hosted proxy was set up using LiteLLM, which translates Anthropic-format requests to DeepSeek V4. The proxy was configured to route requests to either the real or cheap agent based on the command used. The cheap agent runs a shell function that creates an isolated subshell, changes the terminal tab label and icon, and resets the settings when the function exits.
This ensures that the user can easily distinguish between the two sessions. The proxy configuration involves setting various environment variables to point Claude Code at LiteLLM's /v1/messages endpoint and strip unsupported parameters. This allows the tool to operate seamlessly with the cheaper model without the user noticing any difference.
The setup is designed to separate planning and review tasks from high-volume, low-stakes work, with the subscription session handling the more critical aspects of the job.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
