What could Tool Recommendation save your team?

Try our interactive dashboard below for an estimate, then reach out for a custom analysis of your business.

Your setup

Use the dropdowns and sliders below to customize as you'd like.

Not sure? The options below are what we typically see for companies of that size.

Max is the most tools an agent can load up front before they fill the model's context window.

Estimated savings per year

$119K

Claude Opus 5 · 1,000 people · 100 tools

How is this estimated?

Let's talk about how Tool Recommendation could help your team

Video

Arcade's Tool Recommendation in Action

See Tool Recommendation at work inside Arcade.dev.

Introducing Tool Recommendation: Save Up to 37% on Token Cost While Giving Your Agent the Whole Tool Catalog
PRODUCT RELEASE

Introducing Tool Recommendation

Arcade.dev's Tool Recommendation gives your agent the whole tool catalog at a fraction of the token cost: up to 91% fewer input tokens per session, up to 37% lower token costs, and industry-leading odds of finding the right tool.

How I Built & Shipped Tool Recommendation
THOUGHT LEADERSHIP

How I Built & Shipped Tool Recommendation

I was hired to ship Tool Recommendation. Nine months later, here's what it actually took: version numbers that broke agents, a missing newline that cost us points of recall, and air gaps that needed punching.

How this is estimated

  • Compares an agent using Tool Recommendation with the same agent given a fixed list of tools up front, at each vendor's published prices as of October 6, 2026.
  • Assumes 20 conversations per person each month. A conversation is 3 requests with about 8 tool calls each. Input cost only; output tokens are excluded.
  • Assumes the AI app keeps large tool results out of the conversation, as most modern clients do. With results left inline, the percentage saved is lower, but the dollars saved are the same.
  • A fixed list's cost grows in step with its size. Below about 25 to 28 tools, the fixed list costs less.
  • Real sessions vary. Long conversations, a search before every tool call, or resuming onto a grown tool list reduce the saving. On real Claude Code and Codex runs against 100 tools, the total bill fell 7.6% (95% interval 2.5% to 12.6%).

Read the full breakdown in our published paper.