RAG - Basic Architecture
A reviewed baseline for retrieval-augmented generation systems.
Find reusable engineering knowledge, add your own requirements, and export one clean, versioned context bundle for any coding agent.
Transparent files. Pinned versions. No hidden runtime.
Editorially selected starting points.
Product, platform, and delivery directions in the catalog.
Published Blocks in date order.
Operations · Bounds version retention, indexes current pointers, and monitors storage growth so history stays operable.
Data · Decides when rows leave the hot schema for cold storage, what stays queryable, and how restores re-enter the live model without breaking references.
Data · Chooses document/aggregate boundaries, embedding vs reference, and identity so document stores encode access patterns without accidental join hell.
Implementation · Rebuilds hierarchy encodings with ordered backfills and integrity checks so tree moves do not corrupt ancestry.
Data · Chooses how tenant-defined attributes attach to entities (typed columns vs EAV vs JSON) so products can extend schemas without forking the core model.
Data · Keeps immutable history of record versions so prior states remain reconstructable for audit, rollback, and comparison.
A curated collection of Blocks recommended for new projects. Inspect the set before adding individual Blocks to your Workspace.
A Block can be a feature specification, architecture pattern, integration guide, security baseline, MCP setup, schema, reference implementation, or a carefully selected set of official links.
Some Blocks contain a complete folder of requirements and examples. Others need only one excellent README. The format follows the knowledge — not the other way around.
Search by problem, direction, or pattern. Refine results with independent facets and tags.
Read the overview. Open every file. Check the version, maturity, dependencies, license, and source links.
Add exact Block versions to your Workspace and bring your own project requirements.
Download one transparent Solution ZIP and hand it to Cursor, Codex, Copilot, Claude Code, or another coding agent.
Authentication, RAG, uploads, retries, MCP, evaluation, observability — write the hard-won decisions once and reuse them.
Every Block is made of ordinary inspectable files. Nothing is hidden behind a generated prompt.
Latest helps new work start quickly. Pinned versions keep existing Solutions stable when documentation and APIs change.
Your export is a ZIP with Markdown, code, schemas, links, and a small JSON index. It works without a PureIntent runtime.
· · · ·
Keep code as reference. Point to official packages. Add verification steps. Use the artifact that explains the decision best.
PureIntent never runs Block files, installs packages, or asks you to trust code because an AI said it looked safe. Scripts remain readable reference material. Executable binaries are not accepted.
You decide what reaches your coding agent and what it is allowed to do.
Find something you already explain to your coding agent, add your project context, and export a cleaner starting point.