> ## Documentation Index
> Fetch the complete documentation index at: https://klarity.ai/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Refining the context store (when, why, how)

> A practical guide to tuning an already-set-up Context Store safely — when to revisit it, why it matters, and how to avoid wide-blast-radius mistakes.

export const NeedHelp = () => <Note>
    <strong>Need help?</strong> Use the in-app chat — click the chat bubble in the bottom-right corner of Klarity Architect (staffed 24×5) — or email <a href="mailto:support@klarity.ai">support@klarity.ai</a>.
  </Note>;

Your Context Store is rarely "done" — as your business, tools, and how you use Klarity evolve, the rules that shape AI behavior need tuning. This guide covers **when** to refine, **why**, and **how** to do it safely.

<Note>
  New to the Context Store, or want the full picture of how it shapes each feature? Start with [Tailoring Klarity for non-standard use cases](/docs/user-docs/advanced/tailoring-for-non-standard-use-cases). This page is the practical refinement guide for a Context Store that's already set up.
</Note>

## Quick refresher

The Context Store is the business-knowledge layer the AI applies automatically. Every rule has a **scope** (User / Workspace / Organization — who it affects) and a **feature target** (All Features, or a specific one like Advisor, Companion, Signals, Operations — where it applies). Effective context for any interaction is all applicable rules combined.

## When to refine it

Revisit your Context Store when:

* **Outputs are consistently off** — Advisor reports miss a constraint you care about, Companion captures miss your terminology, Signals surface the wrong things. A recurring miss usually means a missing or vague rule.
* **Your business changes** — new ERP or systems, a reorg, a new compliance requirement, a new market or entity. The AI only knows what the Context Store tells it.
* **You've added a new use case** — rolling out a feature to a new team or function often needs context that wasn't relevant before.
* **A rule is doing too much (or too little)** — an "All Features" rule that's affecting features it shouldn't, or a rule too vague to change behavior.

## Why refining matters

Good context is the difference between generic AI and AI that's grounded in your business. But the Context Store is shared and always-on, so a careless rule can quietly degrade quality everywhere — refining is as much about *removing and tightening* as adding.

## How to refine — safely

1. **Find the right scope and feature.** Change behavior for one feature? Target that feature, not "All Features." Just for you? Use User scope. This limits the blast radius if the change has side effects.
2. **Write specific, actionable rules.** "Limit Advisor to 5 recommendations, each one paragraph" beats "keep it concise." Vague rules produce inconsistent results.
3. **Distill, don't paste.** Don't drop whole documents in — extract the few points the AI needs every time. Overstuffed context crowds out the AI's ability to reason about the actual request.
4. **Check for conflicts first.** Before adding, review existing rules at "All Features" and higher scopes. Contradictory rules (e.g., "be comprehensive" vs. "under 500 words") force arbitrary trade-offs.
5. **Save, then verify.** Hard-refresh, open the affected feature, run a quick test interaction, and confirm the behavior changed the way you intended.

## Common pitfalls to avoid

* **The blast radius problem.** A Workspace rule under "All Features" hits every feature for every user. A poorly worded one can silently degrade Advisor, Companion, and Signals at once. Use feature-specific targeting unless the rule is truly universal.
* **Contradictory rules across scopes.** Audit before adding so lower-scope rules complement, not fight, higher-scope ones.
* **Overstuffing.** Keep each scope + feature combination concise (a few hundred words). More context isn't better — it competes with the user's actual request.

## When it's not behaving

| Symptom                           | Likely cause                         | Fix                                                                              |
| --------------------------------- | ------------------------------------ | -------------------------------------------------------------------------------- |
| AI ignores a rule                 | Vague or conflicting rule            | Make it more specific; check for conflicts at All Features / Organization scope. |
| Changes don't take effect         | Browser cache                        | Hard-refresh (Cmd/Ctrl+Shift+R) and retry.                                       |
| Quality dropped after a change    | Overstuffed or contradictory context | Trim length; remove conflicting rules across scopes.                             |
| Works in one feature, not another | Wrong feature target                 | A Companion rule won't affect Advisor — check the target.                        |
| Edit controls disabled            | Not enough permissions               | You need Workspace Admin or Orchestrator to edit Workspace scope.                |

If a rule still isn't behaving as expected, reach out through the in-app chat or to your Klarity team.

## Where to go next

<CardGroup cols={2}>
  <Card title="Tailoring Klarity for non-standard use cases" icon="sliders" href="/docs/user-docs/advanced/tailoring-for-non-standard-use-cases">
    The full picture: Context Store + custom templates.
  </Card>

  <Card title="Running an Advisor analysis" icon="play" href="/docs/user-docs/improve/advisor-analyze-processes">
    Where many tuned behaviors show up.
  </Card>
</CardGroup>

<NeedHelp />