> For the complete documentation index, see [llms.txt](https://docs.meganova.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.meganova.ai/libra-os/use-cases.md).

# Use Cases

Practical, end-to-end guides for building common products on Libra OS. Each follows the same arc — **decide → define → set success criteria → configure → evaluate → improve → integrate** — the difference from a raw-model guide is that in Libra OS each one is a **deployable digital employee**: a configured persona with its own knowledge, structured output, tools, and guardrails, multi-tenant out of the box.

All examples use `https://demo.meganova.app` as the instance URL and a bearer token; swap in your own host and key.

## Guides

| Guide                                                                             | Build                                                                                                               | Libra OS pieces                                                          |
| --------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------ |
| [**Customer Support Agent**](/libra-os/use-cases/customer-support-agent.md) ⭐     | A context-aware support chatbot that answers product questions, stays on topic, and takes actions (quotes, lookups) | Persona + `knowledge_bindings` + `custom_tools` + guardrails + streaming |
| [**Ticket Routing**](/libra-os/use-cases/ticket-routing.md)                       | Classify and route support tickets at scale by intent, urgency, and more                                            | `output_type` classifier + `repair`                                      |
| [**Content Moderation**](/libra-os/use-cases/content-moderation.md)               | Flag and risk-score user-generated content against your policy                                                      | `output_type` classifier + the built-in AI Firewall                      |
| [**Document Q\&A Copilot**](/libra-os/use-cases/document-qa-copilot.md)           | Grounded, cited answers over your uploaded documents                                                                | Knowledge collection + `knowledge_gate`                                  |
| [**Research Report Generator**](/libra-os/use-cases/research-report-generator.md) | Multi-section, evidence-backed reports over knowledge + the live web                                                | `generate_report` + model tiers                                          |
| [Legal Research Assistant](/libra-os/use-cases/legal-research-assistant.md)       | Vertical worked example — matter copilot with case-law research + citations                                         | Persona + knowledge + `deep_research` + `pii_redactor`                   |

## The through-line: three settings decide most of the quality

Every guide comes back to the same levers:

1. **Pick the right model per tier.** The **brain** (which plans multi-step turns) must be a *fast, non-thinking* model — thinking models spend their budget reasoning and stall. Reserve a *premium* model for the **answer** tier (user-facing prose) and a *fast, cheap* model for the **skill** tier (retrieval, classification, report sections). Classification-only tasks (ticket routing, moderation) need no brain at all — just a fast model.
2. **Ground and structure the output.** Bind personas to a knowledge collection so answers cite sources; use `output_type` (a JSON Schema) when you need a machine-readable result (a route label, a moderation verdict) instead of prose.
3. **Set the guardrails for your domain.** The AI Firewall + PII redactor are on by default; tune the threshold, and disable the redactor only where a workflow legitimately needs raw names.

Start with the [**Customer Support Agent**](/libra-os/use-cases/customer-support-agent.md) — it exercises knowledge, tools, guardrails, and streaming together.