Patterns¶
Patterns are documented, reusable solutions to problems that come up repeatedly when building conversational agents on CX Agent Studio. Each pattern is grounded in production experience: a clearly defined problem, a concrete solution, and the trade-offs you should know before adopting it.
Unlike guides — which walk you through SCRAPI's mechanics — patterns focus on agent architecture. They answer the question "how should my agent be structured to handle X?" rather than "which API call do I make to do Y?"
Available patterns¶
| Pattern | Problem | When to use |
|---|---|---|
| Slot Filling | Collecting multiple structured inputs from a user in a natural conversation — without the LLM forgetting state or deciding on its own when to fire backend tasks. | Any agent that gathers information before taking an action: bookings, orders, registrations, claims, intake forms. |
| Dynamic Prompting | Adapting the agent's instruction mid-conversation based on what the user has done. | Agents with distinct phases (pre-auth vs. post-auth, intake vs. resolution) where the LLM's role meaningfully changes. |
| Self-Healing Errors | Giving the LLM exact recovery instructions when a tool call fails, instead of relying on the model to improvise. | Any agent with tool calls that can fail in predictable ways — API timeouts, validation errors, missing permissions. |
How to read these pages¶
Each pattern page is structured the same way:
- Problem — what breaks if you don't use this pattern.
- Solution — the approach, with an architecture diagram.
- Implementation — the code, config, and wiring you need.
- Gotchas — the non-obvious things that will bite you in production.
- Evaluation — how to write tests that verify the pattern is working.
When to use patterns¶
Patterns are not mandatory — you can build effective agents without them. But they become valuable when:
- You've hit a specific failure mode (the LLM forgot a slot, called a task too early, ignored an error).
- You're starting a new agent that resembles a pattern's problem statement.
- You're reviewing an existing agent and want to understand why it behaves unpredictably.
If you're new to SCRAPI, the best starting point is Slot Filling — it's the most comprehensive pattern and serves as the foundation for the Restaurant Reservation Tutorial.
Patterns vs. guides vs. tutorials¶
| Purpose | Audience | |
|---|---|---|
| Patterns | Architectural solutions to recurring problems | Developers designing agent behavior |
| Guides | How to use SCRAPI's features step by step | Developers learning the tooling |
| Tutorials | Build a complete agent from scratch | Developers learning end-to-end |
The Bella Notte restaurant reservation tutorial is the concrete implementation of the Slot Filling pattern. Reading the pattern first and then working through the tutorial is the recommended path.