Use Case · Support

RAG knowledge assistant on the Software Lifecycle Cloud

A RAG knowledge assistant on KERNSTACK is a project that stores your source documents in buckets, declares a vector search index over them, embeds questions with the catalogue model text-embed-v1, and answers through a function that returns the retrieved passages with the reply. The corpus, the index, the model calls, and the API are resources of one stack.json — not a vector vendor, a model vendor, and a backend glued together.

Vector SearchModel InferenceObject BucketsFunctions (WASM)
Talk to us Read Technical Documentation → Software Lifecycle Cloud →

The engineering bottleneck with assembled stacks

Support leads answer the same product questions every week because the knowledge sits in help-center articles and release notes nobody can query as one corpus. The usual assembly — a hosted vector database here, a model API key there, a small backend nobody owns — drifts: articles update, embeddings do not, and the assistant confidently cites last quarter’s behaviour.

Architecture on the Software Lifecycle Cloud

Documents land in a bucket; a vector index is declared over that bucket and re-embeds from source when articles change. A Go function behind a gateway route embeds the question, retrieves the closest passages, and calls a catalogue or custom inference endpoint with those passages as context. Query logs go to a table, so unanswered questions become an editorial list.

Keeping the corpus honest

The index is only as true as its source, so the source lives in the project. When a help article is replaced in the bucket, the index rebuilds from the new file; when an article is deleted, its passages leave the index with it. There is no separate upsert pipeline to fall behind, which removes the classic RAG failure — correct model, stale evidence.

  • Chunking is a property of the indexing declaration, reviewed in the same diff as the route that serves answers
  • Answers can cite the passages retrieved, because retrieval happens inside your function, not inside a black-box hosted chat widget
  • Low-confidence questions are logged to a table instead of being answered creatively

What it costs to run

Embedding a typical help-center corpus with text-embed-v1 costs 3 credits per 1,000 tokens, paid once per document change rather than per question. Question traffic pays for function compute at 0.01 credits per GiB-second and for the model call that drafts the answer. An idle assistant — nights, weekends — spends nothing on compute.

Start from the official work-python-minimal starter blueprint or the full-stack platform demo.

FAQ

Frequently asked questions

By default the catalogue model text-embed-v1 at 3 credits per 1,000 tokens. If you already run a model you prefer, declare it as a custom inference endpoint and point the index and the function at it instead.

Structurally: the function retrieves passages first and instructs the model to answer from them, and questions whose retrieval is weak are logged to a table for a human instead of answered. The corpus being project-owned also means a wrong source document is one bucket upload away from fixed.

Yes. Resolved tickets export into the same bucket or a table listed as an index source. Most teams start with published articles — the corpus they are willing to stand behind — and add ticket text once redaction rules are clear.

Connected in the Project Graph

Related capabilities, architectures, and comparisons

Category PillarThe Software Lifecycle CloudWhy people and agents build, deploy, and operate on a single vertically integrated stack.Service · RetrievalVector searchVector and hybrid retrievalService · ModelsInferenceCatalog and custom model callsService · DataBucketsObject storage with presigned uploadService · ComputeFunctionsGo WebAssembly HTTP handlersIndustry SolutionFinTech Cloud SolutionsFinTech products on KERNSTACK keep the records money moves through — accounts, payment events, approvals, invoice files — in one declared project in a European zone. The pattern fintech teams actually need is rarely exotic: an event trail that cannot be quietly edited, approvals with names on them, retention that is enforced by a job, and a bill that does not require a finance engineer to interpret.Use Case · ProductExperiment assignmentAssignment is a pure function of the subject key and the experiment’s salt — stable across requests, no coordination needed. The function records the assignment row once and serves the variant; outcome handlers were already writing events, which now simply carry the experiment id the assignment returned. Analysis reads assignments and outcomes from neighbouring tables with the same keys.Use Case · AgenciesAgency client deliveryThe studio’s delivery shape is a template stack.json: gateway, functions or workers in the client’s stack languages, tables, buckets, and apps under one target per client, on European infrastructure the studio can name in the pitch. Previews run the same declaration; the client’s approval happens in the bound conversation with a live link. Every client’s meter is separate by construction — the studio sees delivery cost per client without allocation spreadsheets.

Ship your stack on the Software Lifecycle Cloud.

Closed preview on dedicated European infrastructure. Bring a repository or start from a blueprint.

Talk to us See pricing & credits →