Progressive Tool Disclosure
A context optimization strategy in AI Agent design that dynamically exposes tools to the agent on demand rather than loading all available tools at once. By selectively presenting only the tools relevant to the current task, the approach reduces redundant information in the context window, lowers token consumption, and mitigates the risk of hallucinations.
Timeline (last 90 days)
Progressive Tool Disclosure is used to alleviate tool bloat, while Skills targets bloat at the knowledge and instruction level, making the two complementary solutions.
Fabric Gateway adopts a Progressive Tool Disclosure approach, where clients start by seeing only three tools: search, get, and execute.
In the Progressive Tool Disclosure approach, search is used for semantic search to locate tools, get returns the Schema of the selected tool, and execute initiates the actual call.
Progressive disclosure is essentially an on-demand loading strategy, sharing the same principle as Lazy Loading in front-end development, injecting specific Schemas into the context only when the model needs them.
All Facts (4)
Progressive Tool Disclosure is used to alleviate tool bloat, while Skills targets bloat at the knowledge and instruction level, making the two complementary solutions.
50%UnverifiedFabric Gateway adopts a Progressive Tool Disclosure approach, where clients start by seeing only three tools: search, get, and execute.
50%UnverifiedIn the Progressive Tool Disclosure approach, search is used for semantic search to locate tools, get returns the Schema of the selected tool, and execute initiates the actual call.
50%UnverifiedProgressive disclosure is essentially an on-demand loading strategy, sharing the same principle as Lazy Loading in front-end development, injecting specific Schemas into the context only when the model needs them.
50%