emcie-co/parlant

Parlant is a Python framework for building customer-facing AI chat agents that reliably follow business rules. The core problem it addresses is that AI language models, when given a long list of instructions in a system prompt, often ignore some of them or behave inconsistently across conversations. Parlant takes a different approach: instead of hoping the model reads the prompt carefully, it introduces a structured layer that matches relevant rules to each incoming message and enforces them before a response is generated. You define agent behavior through guidelines, each guideline pairs a condition (for example, "customer asks about refunds") with an action ("check order status first"). You can also attach external tools such as APIs or database lookups to specific interaction events, define scripted journeys for multi-step flows, teach domain-specific vocabulary, and use pre-written response templates for situations where you need exact wording rather than generated text. Parlant includes a built-in testing framework for validating that your agent behaves as intended, and provides explainability features that show which guidelines were matched and why. The framework is installable via pip and comes with a local playground for interactive testing.