Skip to content

LlmWikis knowledge page

LLM Wiki Setup Wizard

Browser-only setup planner

Choose a setup path, answer one step at a time, and review the generated packet before any repository or wiki change. The wizard supports both new LLM Wikis and repair work for flat, duplicated, malformed, poorly named, or confusing existing wikis.

  1. 1SelectNew, repair, conversion, evidence archive, handoff, catalog, or multisite.
  2. 2ConfigureScope, authority, taxonomy, URLs, metadata, sources, and review gates.
  3. 3ReviewSetup packet, restructure packet, files, JSON, and stop conditions.
Human visual entry for the LLM Wiki Setup Wizard.Human wizardBuild the wiki setup planUse visible controls for raw sources, compiled pages, review gates, and durable paths.Visitor AI digest entry for the LLM Wiki Setup Wizard at https://llmwikis.org/tools/llm-wiki-setup-wizard/.Visitor AI digestRead the same routeAI agents use the embedded digest on the public wizard URL, then defer writes to the current human.

Current boundary

This page generates a local planning packet in your browser. It does not import files, write to a repository, sync a wiki, automatically publish pages, install a package, expose public MCP or a public write API, open public editing, certify a system, or make LlmWikis the canonical source for UAI-1. Use UAIX.org for canonical AI Memory and Project Handoff definitions, and use the Canonical AI Memory guide to keep source memory, hot memory, transfer memory, and runtime work separate.

When you point an existing project at this wizard, use additive update mode: preserve current AGENTS.md, readme.human, .uai files, generated UAI files, package models, coding standards, active intake rules, and wiki index/log paths unless a human explicitly asks for replacement. When UAIX AI Memory is selected, treat UAI_MEMORY_SYSTEM_PROFILE.md, UAI_MEMORY_RECEIVER_BRIEF.md, UAI_MEMORY_STARTUP_PACKET.md, and optional LLM_WIKI_MEMORY_PLAN.md as local planning outputs, not automatic write or sync permission.

For AI agents

Use this page to plan, not to silently modify. Read the human-visible controls and generated packet; if improving an existing wiki, inventory first and stage a migration for human review.

  • Keep raw sources immutable and preserve provenance.
  • Use one canonical URL per important concept, entity, workflow, standard, source type, or task.
  • Use maps or canvases for discovery, then promote reviewed insights into canonical pages with source traces.
  • Do not count Agent File Handoff as complete until every safe relevant file has a proof-of-use outcome.
  • Add breadcrumbs, children, typed links, metadata, source boundaries, and redirects.
  • Stop on missing permission, secrets, private data, unsupported claims, destructive moves, or ambiguous authority.

Repair existing wiki workflow

Inventory all pages, routes, files, indexes, sitemaps, source records, and internal links. Classify each page as keep, move, merge, split, redirect, archive-noindex, or retire. Pick canonical winners by intent, use maps or canvases only to discover structure until reviewed pages become canonical, build hubs and child pages, replace vague labels such as Part 1 or Page 1 with idea-revealing names, add metadata and typed links, generate redirects, run lint, and stage the packet for approval before publishing.

Draft is saved only in this browser.

Step 1 of 7: Setup path

1. Choose setup path

Setup outcome
Advanced intake paths
Advanced setup outcome

2. Scope and audience

Repair or conversion details
Advanced workspace and context controls

3. Architecture and navigation

4. Optional Agent File Handoff intake

Select this only when loose files, reports, PDFs, screenshots, drafts, or local notes must be visible, reviewed, dispositioned, used for real project work when safe, and archived only after the outcome is recorded.

File intake mode

Intake rule

Agent File Handoff is chat-start review behavior, not CI pickup, a watcher, a queue daemon, automatic execution, or automatic publication. Create or verify Content, Improvement, and Archive buckets before accepting dropped files. Files sitting directly under agent-file-handoff/ are misplaced and must be classified into Content or Improvement before intake refresh. A complete outcome names disposition, actual source-site work, hot-memory outcome, long-memory/archive outcome, checks, blockers, and archive movement after those records exist. Use INTAKE_OUTCOME_LEDGER.md or an equivalent proof-of-use ledger so safe relevant files cannot be routed to memory and forgotten.

Skipped for now. The setup packet will not assume dropped-file intake unless you select it or local evidence requires it.

5. Optional Project Handoff alignment

Choose Project Handoff only when the wiki also needs portable hot context, transfer packets, or UAIX AI Memory handoff files. LlmWikis can remain a standalone durable wiki.

Project Handoff alignment
Project Handoff surfaces

UAIX file alignment

Use the UAIX AI Memory Package Wizard when the project needs a canonical AI Memory package. If the package uses a UAI plus Wiki architecture, include an optional LLM_WIKI_MEMORY_PLAN.md that names the wiki root, index, log or evidence path, steward, source boundary, promotion rules, and update timing. Keep wiki memory as background until reviewed and promoted into accepted project truth.

Mark the setup packet as unresolved and ask the owner before assuming any Project Handoff files, routes, or hot-memory behavior.

Skipped for now. The setup packet will keep the LLM Wiki standalone and note that Project Handoff can be added later.

6. Features and capability boundaries

Feature options

Capability rule

Document skill folders and capability modules in the wiki with owner, risk, dependencies, approvals, tests, and evidence. Keep executable artifacts outside the durable truth tree unless they are reviewed and deliberately promoted.

Knowledge graph rule

Build graph exports from reviewed pages, stable IDs, claim nodes, source traces, contradiction records, and deterministic checks. GraphRAG is retrieval assistance over governed evidence, not a replacement for source review.

AI Dreaming Memory rule

Use dreaming as review-only consolidation over visible source-backed memory. It can suggest summaries, duplicate merges, stale-claim cleanup, contradiction records, and hot/cold memory deltas, but it is not hidden model memory, model-weight training, an autonomous writer, automatic sync, automatic publication, hosted service, certification, or conformance evidence.

7. Safety and review

Visitor-AI digest

AI agents reading this shared URL should use the JSON below as page orientation, then inspect the human-visible controls before proposing setup work. The current human still decides what gets written.

{
    "name": "LLM Wiki Setup Wizard",
    "route": "/tools/llm-wiki-setup-wizard/",
    "purpose": "Plan a new LLM Wiki or repair an existing wiki by defining scope, authority, taxonomy, canonical URLs, metadata, source policy, map/canvas boundaries, review gates, redirects, proof-of-use intake outcomes, and safe staged migration work.",
    "schema_version": "2026-05-24.setup-wizard-digest.v4",
    "audience": [
        "humans configuring an LLM Wiki",
        "AI agents helping a human plan setup"
    ],
    "audiences": [
        "humans configuring an LLM Wiki",
        "AI agents helping a human plan setup",
        "AI agents repairing a malformed wiki with human permission"
    ],
    "supported_modes": [
        "New LLM Wiki",
        "Improve or repair an existing wiki",
        "Convert existing docs into an LLM Wiki",
        "Source/provenance archive or evidence wiki",
        "LLM Wiki plus optional AI Memory / Project Handoff alignment",
        "Capability or skill catalog",
        "Advanced multisite workspace"
    ],
    "use_when": [
        "new wiki setup",
        "improve or repair an existing wiki",
        "existing documentation migration",
        "source/provenance archive setup",
        "Agent File Handoff intake planning",
        "LLM Wiki plus UAIX AI Memory or Project Handoff planning",
        "AI Dreaming Memory review-queue planning",
        "combined Agent File Handoff plus LLM Wiki planning",
        "skill folder or capability catalog boundary planning"
    ],
    "current_support": [
        "browser-only planning form",
        "copyable Markdown setup packet",
        "structured setup model JSON",
        "step-by-step question set for setup mode, fresh setup versus existing-system update, single-site versus multisite workspace setup, multi-repository Git preflight, mutable runtime artifact policy, context budget controls, duplicate-file policy, generated-history policy, Canonical AI Memory layer routing, optional AI Dreaming Memory review-only consolidation, root index topology, map/canvas/outline navigation layer, LLM Wiki multisite interaction strategy, project stage, collaboration model, raw sources, compiled wiki pages, index, log, entity pages, evidence logs, archive target, trust labels, review gates, linting, optional knowledge graph strategy/storage/claim/review/export planning, selectable Project Handoff alignment, optional Agent File Handoff proof-of-use ledger, support escalation, and capability boundaries",
        "UAIX AI Memory alignment guidance for generated UAI_MEMORY_SYSTEM_PROFILE.md, UAI_MEMORY_RECEIVER_BRIEF.md, UAI_MEMORY_STARTUP_PACKET.md, optional LLM_WIKI_MEMORY_PLAN.md, and the rule that wiki memory remains background until reviewed and promoted",
        "public guidance with canonical UAIX links"
    ],
    "not_supported": [
        "automatic repository writes",
        "automatic LLM Wiki sync",
        "automatic publication",
        "hosted AI Dreaming Memory service",
        "hidden model memory or model-weight training",
        "hosted import validation",
        "public MCP server",
        "public write API",
        "open public editing",
        "live benchmark integration",
        "official UAIX generator, SDK, CLI, certification, endorsement, or conformance claim"
    ],
    "canonical_authority_boundaries": [
        "LLMWikis.org is canonical for practical LLM Wiki handbook guidance, setup patterns, repair guidance, metadata patterns, navigation patterns, templates, examples, comparisons, and public planning tools.",
        "AIWikis.org, when referenced, is evidence, provenance, archive, and dogfooding context rather than the canonical instructional source.",
        "UAIX.org remains canonical for UAI-1, AI Memory, Project Handoff, schemas, validator behavior, and UAIX specifications.",
        "Canonical pages are primary. Graph views, maps, JSON, APIs, and retrieval caches are secondary traversal surfaces derived from reviewed pages."
    ],
    "recommended_first_run_steps": [
        "Choose the setup outcome.",
        "Define scope, audience, sensitivity, wiki type, and canonical root.",
        "Classify the authority role: handbook, evidence archive, specification, operations wiki, or mixed.",
        "Generate taxonomy, page types, canonical URL rules, breadcrumb policy, redirect policy, and date-use rules.",
        "Choose whether a mind map, concept map, canvas, or outline is only an exploration layer or a reviewed navigation artifact.",
        "Generate page schema, index/log patterns, typed relations, related-pages blocks, review fields, and agent-use rules.",
        "Set raw/source, ingest, review, privacy, trust, contradiction, and human approval policies.",
        "When Agent File Handoff is selected, require an intake outcome ledger before archive movement.",
        "Copy the Markdown packet and JSON model into a staged review issue or repository note before writing."
    ],
    "repair_existing_wiki_workflow": [
        "inventory all pages, routes, files, indexes, sitemaps, llms.txt entries, internal links, and source records",
        "classify every page as keep, move, merge, split, redirect, archive-noindex, or retire",
        "identify duplicate concepts, orphan pages, flat mega-indexes, repeated global navigation, date-coded concept URLs, inconsistent titles, vague slugs, missing descriptions, missing canonicals, and untyped links",
        "flag SEO anti-patterns such as Part 1, Page 1, Untitled, Misc, raw JSON descriptions, hash-first names, and date-only public titles",
        "choose one canonical URL per important concept by intent",
        "build semantic hubs and child pages",
        "use maps or canvases only to discover structure until reviewed pages become canonical",
        "add breadcrumbs, children, typed related links, metadata, review state, and source boundaries",
        "generate redirects, aliases, page registry, lint checklist, and human review checklist",
        "stage changes for human approval; do not publish or destructively move automatically"
    ],
    "generated_outputs": [
        "Setup Packet",
        "Restructure Packet",
        "AGENTS.md instructions",
        "WIKI_STRUCTURE_PLAN.md",
        "WIKI_MIGRATION_PLAN.md",
        "SCHEMA.md",
        "frontmatter.schema.json",
        "wiki/index.md starter",
        "wiki/log.md starter",
        "wiki/knowledge-map.md starter",
        "raw/README.md starter",
        "INTAKE_OUTCOME_LEDGER.md for Agent File Handoff proof-of-use",
        "UAI_MEMORY_SYSTEM_PROFILE.md when Project Handoff alignment is selected",
        "UAI_MEMORY_RECEIVER_BRIEF.md when Project Handoff alignment is selected",
        "UAI_MEMORY_STARTUP_PACKET.md when Project Handoff alignment is selected",
        "LLM_WIKI_MEMORY_PLAN.md when UAIX AI Memory uses a UAI plus Wiki architecture",
        "redirects.csv",
        "aliases.yaml",
        "llms.txt recommendations",
        "page registry JSON",
        "lint checklist",
        "human review checklist"
    ],
    "stop_conditions": [
        "missing permission to edit",
        "secrets, credentials, private data, regulated data, or unclear consent",
        "unsupported claims without sources",
        "ambiguous canonical authority",
        "destructive move, merge, archive, or retire action without human approval",
        "contradicted source material without a contradiction relation and reviewer path",
        "security, privacy, legal, financial, medical, production, or policy impact without the correct reviewer"
    ],
    "review_gates": [
        "inventory complete",
        "raw sources preserved",
        "canonical URL winners selected",
        "redirect map reviewed",
        "metadata and typed relations added",
        "privacy and secrets preflight completed",
        "internal links updated",
        "active File Handoff files have proof-of-use outcomes before archive movement",
        "sitemap, robots.txt, llms.txt, and canonical tags verified",
        "lint and human review checklist passed"
    ],
    "canonical_links": {
        "start": "/guide/start/",
        "what_is_an_llm_wiki": "/guide/what-is-an-llm-wiki/",
        "build": "/guide/build/",
        "improve_existing_wiki": "/guide/improve-existing-wiki/",
        "structure": "/reference/structure/",
        "metadata": "/reference/metadata/",
        "navigation": "/reference/navigation/",
        "agent_rules": "/reference/agent-rules/",
        "knowledge_graph": "/reference/knowledge-graph/",
        "starter_bundle": "/tools/starter-bundle/",
        "checklist": "/tools/checklist/"
    },
    "project_handoff": {
        "question": "Do you want optional Project Handoff alignment?",
        "effect": "If selected, later wizard questions reveal hot-memory and transfer-packet fields. If not selected, the LLM Wiki stays a standalone durable knowledge layer and Project Handoff can be added later.",
        "source": "https://uaix.org/en-us/specification/project-handoff/",
        "canonical_ai_memory": "https://uaix.org/en-us/ai-memory/canonical-ai-memory/"
    },
    "workspace_setup": {
        "question": "Is this a single-site .uai handoff or a multisite workspace with workspace.uai?",
        "effect": "For a multisite workspace, read workspace.uai before site-local .uai files, resolve the target site from the human request, load only that site hot memory unless cross-site work is explicit, use wiki/global/ for company-wide memory, use wiki/{Site}/ for site-owned memory, and check every listed repository separately before Git sync or merge work."
    },
    "workspace_git_preflight": {
        "question": "What repo-health preflight is required before syncing a multi-project workspace?",
        "effect": "Before Visual Studio Sync, pull, merge, commit, or push, inspect each repository for branch tracking, MERGE_HEAD, unmerged index entries, ahead/behind state, and tracked generated artifacts instead of trusting the current shell repo only."
    },
    "runtime_artifact_policy": {
        "question": "Which local runtime artifacts must stay out of Git?",
        "effect": "Mutable WordPress Studio SQLite databases and similar generated runtime outputs should be ignored and removed from Git tracking with cached-only removal while local files stay on disk. Keep required drop-ins, must-use plugins, .htaccess, and index.php files tracked."
    },
    "context_budget_policy": {
        "question": "Which files and folders should stay out of routine agent context?",
        "effect": "Keep hot memory and indexes small. Generated history, stale generated pages, raw API dumps, package mirrors, and multi-megabyte source files should be represented by summaries, manifests, hashes, and bounded previews unless the task explicitly targets the full source."
    },
    "update_mode": {
        "question": "Is this a fresh setup or an update to an existing handoff and LLM Wiki system?",
        "effect": "When AGENTS.md, readme.human, .uai, generated UAI files, package-model.json, or wiki index/log files already exist, treat the wizard result as an additive update. Preserve preferences, coding standards, source evidence, and active intake rules unless the human explicitly asks for replacement."
    },
    "llm_wiki_multisite_strategy": {
        "question": "If multiple sites use this LLM Wiki, how do source sites, shared archive memory, and publication interact?",
        "effect": "Each source site processes active intake and hot memory first; a shared LLM Wiki or AIWikis-style archive preserves source site, source path, destination path, disposition, checksums, review state, trust label, and promotion status before public use. Site-specific memory goes in wiki/{Site}/; company-wide standards, org info, styling, and coding standards go in wiki/global/."
    },
    "mapping_layer": {
        "question": "Should visual maps, canvases, concept maps, or outlines participate in the wiki?",
        "effect": "Use maps for discovery and navigation, but keep canonical reviewed wiki pages as durable truth. Capture visually, canonicalize textually, retrieve conversationally, and revise versionedly."
    },
    "agent_file_handoff_completion": {
        "question": "When is Agent File Handoff complete?",
        "effect": "A handoff pass is complete only after every safe relevant file has a named disposition, actual source-site work outcome, hot-memory update, long-memory/archive evidence, checks, blockers, and archive movement or an explicit keep-active reason."
    },
    "ai_dreaming_memory": {
        "question": "Should the LLM Wiki use an AI Dreaming Memory consolidation pass?",
        "effect": "If selected, dreaming output remains proposal memory in review queues. It may suggest duplicate merges, stale-claim cleanup, contradiction records, summaries, and hot/cold memory deltas, but it must not write reviewed wiki pages, hot handoff files, production code, release notes, or archive deletions directly."
    },
    "root_index_topology": {
        "question": "Should the root wiki index list every file, or should it route to sub-wiki indexes?",
        "effect": "For one codebase, the root index can be the all-files catalog. For multisite systems, the root index should list sub-wiki directories such as wiki/global/ and wiki/{Site}/, links to each sub-wiki index, and global-only files such as coding standards, organization policy, governance, source maps, and workspace.uai."
    },
    "required_outputs": [
        "LLM Wiki root URL or repository path",
        "multi-repository Git preflight rule and runtime artifact ignore policy",
        "context budget, large-file, duplicate-file, and generated-history policies",
        "raw source path",
        "compiled wiki path",
        "wiki/index.md and wiki/log.md paths plus root index topology",
        "map/canvas/outline artifact path and promotion policy when a visual layer is used",
        "entity page pattern, episodic log pattern, transfer evidence log, archive target, source collection, and update policy",
        "Agent File Handoff proof-of-use ledger path and completion slots when intake is selected",
        "AI Dreaming Memory mode, input scope, raw run path, candidate queue, contradiction path, review policy, protected surfaces, and metrics path when selected",
        "knowledge graph storage model, stable IDs, claim/source-span policy, review-state policy, validation rules, export boundary, and retrieval/abstention policy when graph planning is selected",
        "owner, review cadence, trust labels, sensitivity policy, citation rules, lint checks, support escalation, and update boundaries"
    ],
    "direct_setup_paths": {
        "new_wiki": "/tools/llm-wiki-setup-wizard/#new-wiki",
        "repair_existing_wiki": "/tools/llm-wiki-setup-wizard/#repair-existing-wiki",
        "existing_docs": "/tools/llm-wiki-setup-wizard/#existing-docs",
        "convert_docs": "/tools/llm-wiki-setup-wizard/#convert-docs",
        "evidence_archive": "/tools/llm-wiki-setup-wizard/#evidence-archive",
        "file_handoff": "/tools/llm-wiki-setup-wizard/#file-handoff",
        "project_handoff": "/tools/llm-wiki-setup-wizard/#project-handoff",
        "combined_handoff": "/tools/llm-wiki-setup-wizard/#combined-handoff",
        "capabilities": "/tools/llm-wiki-setup-wizard/#capabilities",
        "advanced_multisite": "/tools/llm-wiki-setup-wizard/#advanced-multisite"
    },
    "query_setup_paths": {
        "new_wiki": "/tools/llm-wiki-setup-wizard/?mode=new",
        "repair_existing_wiki": "/tools/llm-wiki-setup-wizard/?mode=existing",
        "existing_docs": "/tools/llm-wiki-setup-wizard/?mode=existing",
        "convert_docs": "/tools/llm-wiki-setup-wizard/?mode=convert-docs",
        "evidence_archive": "/tools/llm-wiki-setup-wizard/?mode=evidence-archive",
        "file_handoff": "/tools/llm-wiki-setup-wizard/?mode=file-handoff",
        "project_handoff": "/tools/llm-wiki-setup-wizard/?mode=project-handoff",
        "combined_handoff": "/tools/llm-wiki-setup-wizard/?mode=combined",
        "capabilities": "/tools/llm-wiki-setup-wizard/?mode=capabilities",
        "advanced_multisite": "/tools/llm-wiki-setup-wizard/?mode=advanced-multisite"
    },
    "generated_guidance": [
        "setup_readiness_checklist",
        "first_files_to_create",
        "first_actions_before_ingest",
        "setup_model_json",
        "existing_system_update_guidance",
        "agent_file_handoff_plan_guidance",
        "agent_file_handoff_proof_of_use_ledger_guidance",
        "mapping_layer_canonicalization_guidance",
        "ai_dreaming_memory_review_queue_guidance",
        "support_escalation_guidance",
        "browser_only_draft_restore"
    ],
    "skill_boundary": "Document skill folders and modular capabilities in the wiki, but keep executable artifacts outside durable wiki truth unless reviewed and deliberately promoted.",
    "last_reviewed": "May 11, 2026"
}