Skip to content
AskVault Docs
Try Free →

Deployment patterns

Written by , Principal Architect, Networks and Tech, AskVault · Reviewed by Allison Chan
Last updated: · 4 min read

Before you read this page

AskVault is early, and we have no published customer case studies and no customer testimonials. We would rather tell you that than show you numbers we invented.

So this page does the one useful thing we can honestly do: it describes five deployment shapes the product actually supports, and names the specific sources, channels and skills each one uses, so you can judge fit for yourself. Every capability referenced below is implemented in the product today. Nothing below is an outcome, a benchmark, or a promise about what you will achieve.

Pattern 1: B2B SaaS customer support

Shape. A public-facing support assistant over your product documentation. Growth+

How you would configure it.

  • Index product docs, FAQ and blog content as knowledge sources.
  • Deploy the web widget on your marketing site and customer portal.
  • Connect WhatsApp for customers who prefer it.
  • Enable the collect_lead and escalate_to_human skills, plus ticketing_router if you connect Zendesk — Zendesk is the only ticketing target today.
  • Subscribe to the lead.captured outbound webhook to route captured leads into your CRM or automation tool. AskVault sends the webhook; your own automation does the CRM write.

Pattern 2: E-commerce order tracking and cart recovery

Shape. Post-purchase "where is my order" plus abandoned-cart follow-up. Growth+

How you would configure it.

  • Sync your product catalog. There is a per-workspace ceiling on how many products are synced, so a very large catalog cannot exhaust your embedding budget.
  • Connect Shopify. The OAuth scope includes order read access, which is what lets the assistant look up a specific order rather than only answer from the catalog.
  • Deploy the web widget and WhatsApp.
  • Enable wismo_order_status and cart_recovery; both are off by default and need the Shopify connection before they will run.
  • discount_negotiator is separate: it needs a Stripe connection, is Business-tier, and asks the visitor to confirm before it applies a coupon. Business+

Pattern 3: Internal HR helpdesk

Shape. An employee-facing assistant over your HR handbook. Growth+

How you would configure it.

  • Connect Notion and index your employee handbook.
  • Install the Slack integration, then invite the bot into the channel you want it in with /invite @AskVault, such as #people-ops. It only responds in channels it has been invited to.
  • Mark HR sources as internal rather than public. Audience is a two-value setting - public or internal - applied per knowledge source, and it is enforced at retrieval, so internal content is filtered out before the model ever sees it.
  • Turn on identity verification so a claimed identity must be signed by your own backend before the assistant trusts it; a bad signature is rejected. This is an HMAC signature you compute server-side and pass to the widget. It authenticates the identity a visitor claims — it does not by itself require every visitor to be logged in. AskVault does not provide SSO - it verifies a signature from whatever identity system you already run.

Pattern 4: Engineering helpdesk

Shape. An internal assistant over your engineering documentation. Business+

How you would configure it.

  • Connect Confluence for design docs.
  • Connect GitHub. The connector reads a repository's README, the top-level Markdown and text files in its /docs directory (up to 50), and the wiki home page. It does not recurse into /docs subdirectories and does not read issues or arbitrary repository paths, so anything you want indexed needs to live in one of those places.
  • Invite the bot into an engineering channel in Slack.
  • Chunking is heading-aware: Markdown is split at H2/H3 boundaries and each chunk carries its heading path, so a section keeps its context instead of being cut mid-thought.

Pattern 5: API documentation assistant

Shape. A developer-facing assistant embedded in your API reference. Growth+

How you would configure it.

  • Index your API documentation.
  • Embed the web widget on your documentation site.
  • Enable escalate_to_human so questions the docs do not answer reach your engineering team instead of dead-ending.

What these patterns have in common

  • One or more knowledge sources, a channel to reach people on, and a small set of enabled skills.
  • Retrieval is scoped to your workspace, and audience-restricted sources are filtered before the model sees them.
  • Answers are grounded in your indexed content. Where the content does not cover a question, the assistant is designed to say so and hand off rather than guess.

What works less well

Honest exclusions:

  • Highly creative tasks. The assistant does not generate marketing copy or visual designs.
  • Emotional support. It handles factual questions; complex emotional cases need a human.
  • Brand-new domains. It answers from your content. No content means no useful answers.
  • Anything needing a system we do not connect to. If the answer lives in a tool with no AskVault connector, indexing will not reach it.

Limits of this page

  • These are configurations, not results. Nothing here is a measured outcome.
  • This page publishes no deflection, savings or ROI figures, because we have not measured any across real deployments.
  • Plan badges show the tier that unlocks each pattern's gated pieces, not the whole pattern. See the pricing breakdown for what each tier includes.

FAQ

Do you have customer case studies or testimonials?

No. AskVault has not published any customer case studies, results or testimonials, and there are none available on request. This page previously carried illustrative quotes presented as customer endorsements; they were not real and have been removed.

Can I talk to a current customer?

No. We do not run a customer reference programme and cannot arrange an introduction, so we are not going to offer you one. Email sales@askvault.co if you want to talk through whether AskVault fits your use case - that conversation is with us, not with a customer.

How long until I see results?

We do not have a measured answer, and we are not going to estimate one. What we can tell you is what the work involves: indexing your content, choosing channels, and enabling the skills your use case needs. The getting started guide walks through it.

What if my use case isn't here?

Contact us. These five patterns are illustrative, not a supported-use-case list.

Was this page helpful?