Skip to content

Use Cases

JanYork edited this page Aug 14, 2026 · 1 revision

Use cases

Language: English · 简体中文

LWC is most valuable when knowledge must survive a session, remain traceable to evidence, or improve as an Agent revisits the same domain. It is unnecessary for every prompt.

Choose LWC when

Use LWC when at least one of these is true:

  • a later session should reuse the result;
  • the task depends on several documents or earlier decisions;
  • facts must remain traceable to immutable evidence;
  • the project is large enough that rediscovery is expensive;
  • rules or runbooks must load reliably without search;
  • document relationships or code structure materially affect the answer;
  • new evidence should revise existing knowledge rather than create another standalone summary.

Common scenarios

Long-running software development

An Agent can preserve architectural decisions, module responsibilities, operational constraints, and verified debugging conclusions. A later session recalls the maintained Wiki, uses CodeGraph for current structure, and reads the relevant code before changing it.

Useful LWC capabilities: project scope, search, CodeGraph, changesets, source lineage.

Source-backed research

As reports, papers, and interviews accumulate, the Agent stores immutable snapshots and updates shared concept, entity, comparison, and synthesis pages. Contradictions remain visible instead of being silently flattened into one answer.

Useful LWC capabilities: source ingest, citations, provenance, document graph, retrieval acceptance.

Operational rules and runbooks

Core rules often need to behave like dynamic system context: the Agent must receive the complete current page at session start or after context compaction. Strong tags provide deterministic, bounded full-page loading.

Useful LWC capabilities: tags, autoload policies, lifecycle Hooks, checkpoints.

Understanding an unfamiliar repository

The Wiki explains prior intent and project-specific language. CodeGraph maps current symbols and dependencies. The Agent combines both with direct source reads instead of treating a stale architecture note as current code truth.

Useful LWC capabilities: bounded context, CodeGraph, document graph, verified write-back.

Reviewing a changing source

When a tracked file changes, LWC can compare the current bytes with its immutable snapshot and list directly citing pages. The Agent reviews whether the meaning changed, adds a new path revision when necessary, and updates only affected claims.

Useful LWC capabilities: source status, source diff, source refs, ingest, citations.

Building a durable answer library

A useful investigation, comparison, or decision can become a query, comparison, or synthesis Page. Later questions retrieve the maintained answer and its evidence instead of relying on chat history.

Useful LWC capabilities: search, Page write-back, provenance, retrieval weights and feedback.

Working with Office, PDF, or web-exported documents

When an optional adapter is installed and selected, LWC can invoke MarkItDown or AnyDoc to convert one authorized local file into a new Markdown file. The output is reviewed before it is added as a Source.

Useful LWC capabilities: document conversion, source ingest, immutable snapshots.

Exploring relationships visually

The read-only Viewer can show Wiki pages, Sources, document relationships, code relationships, and a bounded Word Graph without allowing browser-side mutation.

Useful LWC capabilities: Viewer, document graph, CodeGraph, Word Graph.

Use something simpler when

Skip LWC for:

  • a trivial, self-contained transformation with no future value;
  • formatting-only edits;
  • exact literal search that rg can answer immediately;
  • temporary build output or routine progress logs;
  • secrets, tokens, cookies, private keys, or sensitive raw transcripts;
  • unsupported guesses that are not useful enough to preserve as labelled hypotheses;
  • a hosted collaborative knowledge portal with real-time multi-user editing.

LWC can accompany a task without becoming the task. If memory maintenance would delay the user's immediate outcome, finish the primary work first and preserve only the verified result that is worth reusing.

Decision table

Question If yes If no
Should another session reuse this? Preserve verified knowledge in LWC. Keep the result in the current response or deliverable.
Does the claim depend on evidence? Store or inspect a Source and cite it. Use explicit user or Agent provenance when durable.
Do relationships matter? Use the document graph or CodeGraph after checking readiness. Use bounded lexical recall and direct reads.
Must complete pages load without search? Use a small strong tag with a limit. Use normal search.
Does the update span several dependent writes? Use a changeset. Use one transactional command.
Is the input non-Markdown? Configure one optional conversion adapter. Add the reviewed UTF-8 source directly.

Next: Installation and upgrades

LWC Wiki

English · 简体中文


Start here · 开始使用

Core capabilities · 核心能力

Practical guides · 实战指南

Capability configuration · 能力配置

Technical design · 技术设计

Operations · 运行与维护

Reference · 参考资料

Contributing · 参与贡献


Repository · Releases

Clone this wiki locally