Tool Recommendation for Multi-Step Agents
Finding every tool a task needs. A multi-step agent task
fails when a single tool it needs is missing, yet the tools
an agent may need are too many to list. This research paper
shows how Arcade gives an agent all of them through a fixed
interface of three tools, and what it does to accuracy and
cost.
What's in the paper
- How a fixed interface of three tools, SelectTools, UseTool and ListApps, reaches a catalog of roughly 8,200 tools without fine-tuning on customer tools.
- On ToolRet, 57% more complete tool chains than BM25, and a first recommendation that covered every required tool in 93% of Codex sessions.
- Against a 100-tool static list, 17% to 40% lower input tokens and a total bill that fell in all 14 runs, with no change in task completion.
- When a static list is still cheaper: small catalogs, short lookups and search-heavy sessions.