Tailoring Within for non-standard use cases

This is an advanced guide. Once you’re comfortable with the basics, Within gives you two powerful levers to shape AI behavior for non-standard needs: the Context Store (to change how features behave) and custom templates (to define repeatable input → analysis → output workflows). Used well, they let you tailor Within to your business without custom development.

​

Lever 1: Shape behavior with the Context Store

The Context Store is the business-knowledge layer you give the AI. Instead of repeating context in every prompt, you define rules once and they’re applied automatically across features. It’s organized along two dimensions:

• Scope — who it affects: User (just you), Workspace (everyone in the workspace), Organization (everyone, everywhere). Rules cascade down, and all applicable rules combine.
• Feature — where it applies: All Features, or a specific one — Advisor, Companion, Signals, Operations. This is the key to non-standard tuning: you can change one feature’s behavior without touching the others.

​

Tuning behavior per feature

Because rules can target a single feature, you can fine-tune each one for your situation. Examples drawn from the configuration playbook:

• Advisor (report behavior): “Structure reports with Executive Summary, Analysis, Recommended Actions.” · “Flag any recommendation requiring capex over $500K.” · “Always include regulatory compliance implications.”
• Companion (capture awareness): “Our AP team works primarily in SAP VIM.” · “SOPs live in Confluence.”
• Signals (personal relevance, at User scope): “I’m a Senior AP Manager focused on processing efficiency and error reduction.”
• All Features / Organization (guardrails): “Never include PII in generated output.” · “All recommendations must comply with SOX.”

​

How to set it

1. Open Settings → Context Store.
2. Pick the scope (User or Workspace) and the feature target.
3. Write your rules as clear, specific plain text, and save.
4. Hard-refresh and run a quick test interaction to confirm the behavior changed.

​

Best practices (and pitfalls)

• Be specific, not vague. “Limit Advisor to 5 recommendations, each one paragraph” beats “keep it simple.”
• Distill, don’t paste. Don’t dump whole policy docs in — extract the few rules the AI needs every time. Long, unstructured context degrades output across features.
• Start narrow. Prefer feature-specific rules; reserve “All Features” for truly universal context. A bad “All Features” rule has a wide blast radius — it can silently degrade Advisor, Companion, and Signals at once.
• Check for conflicts. Contradictory rules across scopes (e.g., “be comprehensive” vs. “under 500 words”) force the AI into arbitrary trade-offs. Audit before adding.

​

Lever 2: Build custom templates

Templates define how Within turns process understanding (and other inputs) into a specific output — SOPs, BRDs, SOX narratives, PDDs, and your own custom formats. A custom template is, in effect, a small chained workflow: it takes defined inputs, applies a specific kind of analysis or structure, and produces a consistent output every time — much like an agent with a built-in skill. This is the lever for non-standard outputs: when the standard document types don’t match what your team needs, a custom template lets you encode your own logic — what to look for in the input, how to analyze it, and exactly how the result should be shaped — so anyone can run it and get the same structured result. You can also add input files to a template run to analyze them on top of your process understanding, extending what a single template can reason over.

​

Putting them together

The two levers compound. Context Store sets the behavioral defaults every feature follows (tone, constraints, terminology, guardrails); custom templates define repeatable workflows that turn inputs into specific outputs inside those defaults. Together they let you adapt Within to genuinely non-standard use cases without engineering work.

​

Where to go next