Context Token Optimization
Stop burning tokens on terminal noise your AI doesn't need to see.
Questions this answers
- How much does Claude Code cost per session?
- How to reduce LLM API token usage in terminal?
- Warp terminal AI burning through requests
- Monitor Claude Code token consumption in real time
- How to optimize context window usage for AI coding agents?
How it works
Context Token Optimization (CTO) intercepts and analyzes the API traffic between your AI agent and its LLM provider. It counts input and output tokens using provider-specific tokenization, then surfaces those numbers in real time on a per-tab and global basis. You see exactly how many tokens each prompt and response consumes as they happen.
CTO also identifies patterns that waste tokens: repeated large context inclusions, redundant file reads, and bloated system prompts. By making these patterns visible, it gives you the information needed to restructure prompts or adjust agent settings to reduce unnecessary token consumption.
The feature works transparently with Claude Code, Codex, Gemini CLI, and any agent that calls standard LLM APIs. No agent-side configuration is required. CTO can be toggled on or off globally, or overridden per tab for fine-grained control.
Why it matters
Every character in your terminal scrollback costs tokens when an AI agent reads it. ANSI escape codes, prompt decorations, progress bars, blank lines: none of it helps the model reason about your code, but all of it costs money. CTO mode rewrites terminal output to strip the noise before the AI reads it. Developers report 30-50% context reduction on typical sessions. That is real money across a team.
Frequently asked questions
Does CTO modify my API requests?
No. CTO is a passive observer. It reads and analyzes API traffic but never modifies, delays, or interferes with requests or responses between your agent and the LLM provider.
Which LLM providers does CTO support?
CTO supports all major providers including OpenAI, Anthropic, Google, and any provider using standard REST or streaming API formats. Custom endpoints are also supported as long as they follow common LLM API conventions.
Can I disable CTO for specific tabs?
Yes. CTO supports per-tab overrides so you can enable it globally but disable it for specific tabs, or vice versa. See the CTO Per-Tab Override feature for details.