llm-provider-contract

Standardize LLM provider calls by routing every request through a unified adapter contract that enforces common shapes for requests, responses, and errors, while applying strict secret redaction to prevent API keys from leaking into business logic.

Streaming normalization converts provider streams into consistent events and requires local JSON Schema validation before accepting structured output, ensuring tool-use compatibility and preventing malformed data from breaking downstream execution flows.

Implement capability-aware fallback by defining timeout, retry, and rate-limit boundaries within the provider adapter contract, ensuring cancellation-aware behavior and normalized SafeLLMError handling across multi-provider LLM backends.

Yes, a provider and model registry is required to configure timeout and rate-limit boundaries, enabling the adapter contract to correctly apply cancellation-aware retry policies and route calls safely across different LLM providers.

Multi-provider LLM integrations fail without streaming normalization and local JSON Schema validation, which standardize provider-specific stream events and guardrail structured data before it enters your child-agent execution flows.

Yes, the provider adapter contract includes tracing requirements and error normalization, enabling observability across planning and child-agent execution flows by emitting normalized LLMStreamEvent objects for every tool-use interaction.