Feature Deep Dive

Jardine AI works best when each feature has a clear job. Most operational confusion comes from using the right tool for the wrong reason. The sections below explain what each major area is really for, when to use it, and when to avoid over-configuring it.

The Product in Five Functional Layers

A practical way to understand the product is to see it as five layers that cooperate.

The first layer is ingress. This is where conversations arrive through Email Inbox, Intercom, or Zendesk. If ingress is unreliable, nothing downstream can look stable.

The second layer is knowledge. This is the factual base for non-account-specific answers. It includes library ingestion, validation behavior, and analysis health signals.

The third layer is policy. This is routing behavior: tags, destinations, and rules when manual controls are enabled. Policy decides how conversations move when conditions differ.

The fourth layer is runtime data. This is connectors and templates for account-aware answers that documentation alone cannot provide.

The fifth layer is operations visibility. This is where you inspect conversation outcomes, ownership, escalations, and day-to-day quality signals.

When teams mentally separate these layers, troubleshooting and scaling become much cleaner.

Knowledge Surfaces: Library, Validate, and Analysis

Knowledge Library is where your source material lives. Collections and document states tell you whether content is active, processing, failed, or archived. Use this area to keep your support facts current and customer-safe.

Use Library when you need to improve what Jardine knows. Do not use routing as a substitute for poor source content. If answers are weak, stale, or contradictory, the fix usually starts in the library.

Knowledge Validate is where you test behavior before customers do. In advanced validation mode, you can run playground chats, inspect evidence, rate answers correct or wrong, save feedback, simulate outcomes, and run suites with saved runs. This is your fastest path to behavior confidence.

Use Validate when you are changing knowledge, introducing connector-backed behavior, or adjusting policy expectations. Skip broad rollout until your validation signals look healthy.

Knowledge Analysis provides health diagnostics over time. You can inspect indexed ratio, failed ratio, collection-level health, issue lists, and recommendations. This is where hidden quality drift shows up first.

Use Analysis for ongoing maintenance. If you only look at conversations after users complain, you are already reacting too late.

Routing and Conversations: Policy and Control in Production

Routing has two useful modes. In simple mode, Jardine handles most routing automatically. In manual mode, you can define explicit tags, destinations, and rules.

Use simple mode when you are early, when your support flows are still forming, or when default behavior is already meeting expectations. Use manual mode when policy must be explicit, auditable, or tightly constrained.

Manual routing gives you powerful condition controls such as confidence ranges, message content checks, and human-required paths. Use those controls carefully. A small, precise rule set usually performs better than a large, overlapping one.

Conversations is where policy outcomes become real. You can filter by status and channel, inspect message history, review confidence and intent context, and hand ownership between AI and human.

Use conversation detail as your operational truth. If a teammate asks “why did this escalate?” this is where your investigation should start.

Channels, Connectors, and Settings

Settings is where you configure channel-specific integration behavior.

Email setup includes forwarding address readiness and domain verification via DNS TXT records. Intercom setup includes connection state and optional advanced routing profile controls. Zendesk setup includes OAuth, webhook URL, and webhook secret management.

Use Settings to make ingress and delivery reliable. A channel that looks connected but is missing webhook or secret configuration is not production-ready.

Data Connectors add runtime context for account-aware support. Supported connector providers in the current app include Postgres, MongoDB, Supabase mode, and Stripe. You can run in automatic or manual connector modes, manage template libraries, test templates with parameters, and review schema snapshots for auto mode.

Use connectors only for clear support questions that require live account facts. Keep connector scope tight and template behavior explicit. Treat connector design like policy design: small, intentional, and testable.

How These Features Work Together in Practice

A strong production pattern usually looks like this.

A message enters through one connected channel. Jardine uses knowledge to ground an answer path. If account context is required, connector templates provide runtime facts. Routing policy determines whether to continue automated handling or escalate. Conversation views expose what happened. Validation and analysis close the loop so tomorrow’s behavior is better than today’s.

The mistake to avoid is optimizing one feature in isolation. You can build perfect routing and still fail with weak knowledge. You can build perfect knowledge and still fail with unstable channel setup. You can add connectors and still fail if template tests are not part of your release rhythm.

The teams that get durable results do one thing consistently: they treat Jardine as an operating system, not a collection of toggles. They define clear jobs for each feature, test changes before broad rollout, and keep control surfaces simple until complexity is truly needed.

If you want to sharpen operational execution after this overview, continue with Best Practices.