How to configure AI capabilities and context in rConfig V8

After reading this page, you can explain how rConfig grounds AI responses in real context instead of generic model knowledge, and turn individual AI capabilities on or off.

When to use this

Use this page when deciding which AI capabilities your team should have access to, or when you need to turn a specific capability off, for example while evaluating a new provider, or in response to a support or compliance concern.

Prerequisites

• Admin role: the AI Configuration tab is hidden from non-admin users
• An active AI provider (see Set up an AI provider): capabilities can be toggled without one, but won’t produce a response until a provider is active

How rConfig grounds AI responses

A general-purpose model doesn’t know your naming conventions, your golden configs, or how rConfig itself works. rConfig closes that gap with two separate grounding mechanisms, used by different capabilities:

The context harness for config-work capabilities

Config Analysis, Drift Detection Narration, and Policy & Compliance Narration are grounded in the context harness: a layer of curated networking domain knowledge, composed with per-capability context assembly before the request reaches the model. This is what turns a general model into one that reasons like a network engineer familiar with your environment, rather than one guessing from generic training data.

The knowledge layer for product-knowledge capabilities

The Optimization Assistant is grounded differently. It doesn’t need networking domain knowledge. It needs to know how rConfig itself works. Its answers are retrieved per turn from an indexed copy of docs.rconfig.com content, pulling only the fragments relevant to your question rather than the whole site. This keeps it accurate even against a small local model with a limited context window.

Context budget, not a token budget

Each capability caps how much context it injects into a single request (roughly 8,000–14,000 characters, depending on the capability). This exists to keep requests fast and affordable and to keep every capability usable against a modest local model. It isn’t a spend or usage limit. See Governance, audit & privacy for what rConfig does and doesn’t track about cost.

Customer knowledge entries

The context harness supports a customer-owned layer on top of rConfig’s base knowledge: your own conventions, golden configs, naming standards, and design principles, added without overwriting the base layer. rConfig’s curated base knowledge is viewable but read-only (you can’t edit it), while your customer entries support full create, update, and delete. Both are managed through the AI Center API today (see API Docs) rather than a dedicated in-app form.

The four capabilities

Capability	What it does	Grounded by
Config Analysis & Explain	Plain-English summaries of device configurations	Context harness
Drift Detection & Narration	Detects and narrates configuration drift between versions, read-only	Context harness
Policy & Compliance Narration	AI-assisted compliance rule creation backed by CIS/NIST benchmark parsing; writes to the policy store only	Context harness
Optimization Assistant	Doc-grounded chat on the System Optimization settings page	Knowledge layer

Step-by-step walkthrough

1. Open AI Configuration

1. Click AI Center in the left sidebar.
2. Select the AI Configuration tab.

2. Turn a capability on or off

Each capability is shown as a card. Alongside the enable/disable toggle, the card lists the capability’s description, its Current model (shown as “Provider default”), and Usage this month. Click the toggle to enable or disable the capability.

Model overrides come later

Every enabled capability uses the active provider’s default model today, which is why Current model reads “Provider default” on every card. The Override model button on each card is intentionally disabled: per-capability model overrides are reserved for a future release. The Usage this month field is a placeholder for the same release and currently shows a dash.

Takes effect immediately, everywhere

Disabling a capability here removes its entry point from every UI surface immediately: the toolbar button, the diff panel, the compliance results panel, and the settings page all stop showing it. There’s no separate confirmation step, so double-check you’ve selected the right capability before toggling it off.

3. Confirm the change

The toggle updates immediately; there’s no separate save step for capability state.

Common gotchas

Common configuration mistakes

• Turning off a capability doesn’t stop billing on the provider side: if you’re trying to control spend, do that in your provider’s own dashboard. The toggle only controls whether rConfig users can reach the capability.
• All capabilities share one provider: you can’t ground Config Analysis in one model and the Optimization Assistant in another. Switching the active provider on AI Settings changes what every enabled capability uses.
• A capability with no active provider still shows as “enabled”: the toggle controls visibility, not whether a provider is available to answer. Check AI Settings if an enabled capability isn’t producing responses.

What’s next

• Set up an AI provider
• Use Config Analysis & Explain in the config viewer
• Review what data each capability sends to your provider