Credo: Declarative Control of LLM Pipelines via Beliefs and Policies

AI assistants are becoming commonplace in domains that require stateful decision-making in continuously evolving environments. As such, correctness depends on an assistant's ability to flexibly adapt by incorporating new evidence or revising prior conclusions. However, existing approaches rely on either static control flow (e.g., fixed pipelines) or unreliable agentic reasoning. We therefore introduce Credo, a database-backed semantic control plane for LLM pipelines that represents state as high-level beliefs and guides execution through declarative policies defined in terms of those beliefs. Through this design, Credo enables adaptivity based on runtime signals while also simplifying auditing and debugging. This demo will illustrate these concepts using an LLM pipeline built for a financial question answering benchmark.