Abstract
Systems and methods provide machine-facing recommendations for autonomous agents selecting tools and services from large catalogs. A task specification including a task representation, required input and output types, and explicit constraints (cost, latency, reliability, chain length, and context-token budget) is received. Tools are represented with structured descriptors including learned capability vectors, schemas, type tags, and empirically observed performance. A directed composability graph encodes tool-to-tool compatibility using semantic type compatibility, schema compatibility, and historical co-execution success learned from outcomes. Candidate ordered tool chains are generated by traversing the graph under constraints and scored with a multi-objective function including predicted task success, cost fit, latency fit, and reliability, producing ranked chain recommendations. Execution outcomes across agents update capability vectors, composability weights, and collaborative filtering models. A context budget optimizer allocates a language-model token budget across ranked tool descriptions by assigning detail levels subject to a token constraint.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Anonymous, "Recommendation Systems for Autonomous Agent Tool and Service Selection", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/10743