An experimental AI coding agent and Claude Code integration that explores token-efficient workflows, reducing token usage by 20–40% (and beyond) while maintaining output quality across many use cases. Fork of the great OpenCode.

Install

Standalone agent
curl -fsSL https://chippery.ai/install.sh | sh click to copy
Claude Code integration
curl -fsSL https://chippery.ai/install.sh | sh --claude-code click to copy

Standalone agent requires an API key or active subscription* from providers such as Anthropic, OpenAI, Google, GitHub Copilot, or a local model. Chippery runs on your machine; optimization happens on-device.

* Subscriptions are always subject to the Terms of Service of the provider.

What this is

Chippery uses navigation tools, smart indexing, and pre-trained heuristics to reduce token usage during LLM-assisted coding. In our benchmarks, we've seen 20–50% reductions on average, but results vary significantly by task, model, and codebase.

This is experimental software. We're publishing it because it's useful to us and might be useful to others, but we make no guarantees about token savings at this time. While we do see overall improvements in most tasks, some may not see such benefit.

We try to be transparent about what works and what doesn't. See our benchmarks and methodology.

Modes

/balanced — Moderate optimization, low variance. A reasonable default.
/balanced-proPro — More aggressive heuristics, typically ~ 25-35% savings at no quality loss
/frugalPro — Minimal-token mode. Highest variance, occasional regressions.
/all-in — No optimization. Baseline behavior.

Tools

Standalone agent only. Not available in Claude Code integration.

/unstuck — Interrupts loops when the model stalls.
/map — Manage the codebase index.
/spend-controlPro — Set token budgets with approval thresholds.
/pinPro — Pin responses for longer context retention.
/vibecheckPro — Check for SQL injections, hardcoded API keys, etc.
/config — Configure Chippery to your preferences.

Pricing

Chippery Lite is free. Pro exists to fund ongoing development, benchmarking infrastructure, new optimization strategies and maintenance. Chippery is experimental.

Lite
Free forever
/balanced mode at ~20% token savings, /unstuck, /map, core workflow. Genuinely useful on its own.
Pro
$7/month or $99 once (lifetime)
Everything in Lite, plus /balanced-pro (can achieve ~ 25-40% token savings), /frugal (up to ~ 50%), /spend-control, /pin, /vibecheck. 14-day refund. Up to 3 devices.
Pay per month $99 lifetime
Claude Code integration supports /balanced, /balanced-pro, and /frugal modes only.

Students or anyone who can't afford Pro: reach out.

FAQ

What is this?
A local-first coding agent and Claude Code integration, forked from OpenCode, that experiments with token-optimization strategies.

Will it actually save tokens?
Quite often. We've measured 20–50% reductions on specific benchmarks, but your results will vary. Some tasks won't improve. The more aggressive modes have higher variance. The best experience overall is balanced-pro at ~ 25-40% savings at no quality loss. Balanced mode offers 15-20% savings in most use cases. Frugal saves the most (up to 50% in some cases) but can struggle with large codebases.

Why charge for Pro?
Running benchmarks, maintaining infrastructure, and exploring new optimizations is quite costly. Pro funds that work.

Is lifetime truly lifetime?
Yes, this is a limited offering for early supporters. It's truly lifetime and includes all future upgrades. However should future token saving strategies necessitate server-side inference then lifetime would not cover those running costs. Our main focus is on-device savings however.

Is it open source?
Chippery is a fork of MIT-licensed OpenCode. It is 100% open source. To support our work and continued development, consider subscribing to Pro or purchasing Pro lifetime. We are committed to contributing upstream wherever it is sensible.

Where does it run?
You can run it entirely locally. Pro does a brief license check (~weekly) via our payment provider Polar.sh, but no code or project data leaves your machine. If you run a local model, you can run it entirely offline. Your LLM provider's terms apply if you use an API or subscription.

Commercial use?
Yes.
We are also working on features for teams, contact us for information. If you need Enterprise features today, we recommend checking out OpenCode Enterprise.

Refunds?
Full refund within 14 days. Email [email protected].