Compose AI Context

This use case focuses on combining data from multiple APIs into one capability to deliver richer, task-ready context to AI clients. When agents need data from several sources to complete a task, forcing them to make multiple tool calls wastes context window space, increases latency, and introduces fragile multi-step orchestration logic inside the AI layer.

Teams need to register multiple consumed APIs with unique namespaces and orchestrate cross-source calls using ordered steps—so outputs from one source feed inputs of the next. A single composed output model via mapped output parameters gives AI clients one coherent result instead of multiple fragmented API responses.

Pain Points

• AI tasks require data from multiple APIs but agents make separate calls to each
• Multiple tool calls waste context window space and increase token costs
• Cross-source orchestration logic ends up inside the AI layer where it is fragile
• No standard way to combine and shape data from different API sources
• Fragmented responses require agents to reconcile and merge data themselves
• Reuse of composed patterns across teams is nonexistent

Expected Outcomes

• Richer, task-ready context delivered to AI from multiple sources in one call
• Cross-source orchestration defined declaratively in YAML
• Reusable source adapters shared across composed capabilities
• Fewer agent tool calls and less context pollution
• One coherent result model instead of fragmented API responses

Narrative

An organization’s AI agents need to complete tasks that require data from multiple internal and third-party APIs. A customer lookup might need data from the CRM, billing system, and support platform. Each agent makes three or four separate tool calls, filling the context window with raw responses that must be reconciled.

Teams begin composing capabilities by registering multiple consumed APIs with unique namespaces. Ordered steps orchestrate cross-source calls—step mappings and per-step input injection ensure outputs from one source feed inputs of the next. The orchestration logic lives in the capability spec, not inside the AI layer.

A single composed output model returns one coherent result via mapped output parameters. AI clients see a unified response instead of fragmented data from three different APIs. The agent makes one tool call, receives everything it needs, and proceeds with the task.

Source adapters are reusable across composed capabilities. The CRM adapter declared for one capability is available to any other capability that needs CRM data. Teams build on shared source definitions instead of redeclaring the same APIs in every composition.