optimize-integration

Improve LLM routing accuracy by enriching tool descriptions, refining field compaction, and tuning response behavior. This optimization process applies iteratively to existing adapters to reduce payload noise and guide routing decisions effectively.

Field compaction refinement tailors output specifications to reduce noise while preserving necessary routing information. This process minimizes irrelevant payload data sent to the LLM, directly improving response efficiency and routing accuracy across the adapter.

Calibrate response size and client behavior for raw HTTP adapters through response tuning. This optimization phase adjusts payload shape and client behavior to ensure efficient data delivery and prevent oversized responses from degrading LLM usability.

Optimize integration adapters non-disruptively by applying phase-driven changes across description enrichment, compaction refinement, and response tuning. Commit changes iteratively with unit tests to ensure frontmatter metadata consistency and maintain existing functionality.

Enrich tool descriptions when an LLM struggles to correctly route requests to the target adapter. Tiering and enriching descriptions guides the LLM's routing decisions, ensuring it selects the correct tool based on the user's intent.

Adapter optimization requires no external dependencies to modify payload shape and field lists. The process operates directly on existing adapter configurations to tune response behavior and refine compaction without additional packages.