← Back to the catalog
C7KNOWLEDGE

Universal internal knowledge search

A single search interface that lets anyone in the company ask questions and get answers drawn from across the firm's knowledge sources — wikis, shared drives, past project archives, Slack threads, ticket history. Returns a synthesized answer with citations linking back to source documents. The pattern's value is less about finding documents and more about finding the answer across documents that don't always sit in the same system. Replaces the typical 'I know we wrote that down somewhere' expedition with something that takes ten seconds.

WHERE THIS FITS
BUSINESS SHAPES
B2B servicesProfessional servicesProduct company
VOLUME THRESHOLD
Below 500 documented knowledge assets a month, the payback rarely earns the build. Patterns this shape reliably pay back at 10,000+.
FITS BEST
Any firm >25 people. Pain scales superlinearly with headcount.
PAYBACK · 8-14 moBUILD · Medium-HighVALUE · $40k-$180kWHEN · >25 employees
FAILURE MODE TO DESIGN AROUND
Permissions sprawl → respecting source-system ACLs is the hard problem, not the AI.
REQUIREMENTS · 4 REQUIRED, 2 OPTIONAL

Requirements describe capabilities the pattern needs in your environment, not the vendors you must buy. Any system that fills a requirement satisfies it — that’s what makes the catalog portable across the long tail of SMB tooling.

  1. knowledge_corpus_set
    REQUIREDREADcorpus

    The collection of sources the pattern indexes and searches. Plural because most firms have several disjoint stores.

    DATA SHAPE
    Documents, pages, threads, tickets from multiple sources. Each indexed with title, source system, last-modified date, author, access permissions.
    COMMONLY FILLED BY
    • internal wiki
    • shared document drive
    • archived project folders
    • internal blog or announcements
    • historical support tickets
    • past chat threads in select channels
  2. user_query_surface
    REQUIREDREAD + WRITErequest

    Where people actually ask. Has to be where they already are, not a separate tool they have to remember to use.

    DATA SHAPE
    Natural language question with the asking user's identity (for permissions filtering) and any context about what they're working on.
    COMMONLY FILLED BY
    • chat slash command or DM to a bot
    • search bar embedded in the wiki
    • browser extension that activates on highlight
    • command palette in the firm's working tools
  3. permission_resolver
    REQUIREDREADrequest

    Determines which sources and documents the asking user is allowed to see. Critical: the pattern must never return content the user doesn't have access to in the source system.

    DATA SHAPE
    Per-user-per-document permission lookup, or per-user-per-source-tag.
    COMMONLY FILLED BY
    • SSO identity tied to source-system permissions
    • permission groups maintained per source
    • permission inheritance from the source systems themselves (preferred)
  4. answer_destination
    REQUIREDWRITErequest

    Where the synthesized answer with citations goes back to the user.

    DATA SHAPE
    Answer text with inline citations linking to source documents. Includes a confidence indicator and explicit 'I don't know' option.
    COMMONLY FILLED BY
    • reply in chat thread or DM
    • answer panel in the wiki
    • browser extension overlay
    • command palette result
  5. feedback_capture
    RECOMMENDEDWRITEevent

    Per-answer thumbs-up/down or correction. Without this, the pattern's accuracy is invisible.

    DATA SHAPE
    Per-query feedback: query, answer, sources cited, user rating, optional correction.
    IF MISSING
    Pattern works but quality drift is invisible. The team has no idea whether the answers are good. Strongly recommend at minimum a thumbs widget.
    COMMONLY FILLED BY
    • thumbs widget under each answer
    • follow-up chat reaction the user can use
    • weekly digest of queries asking for ratings
  6. user_role_context
    OPTIONALREADrequest

    What role the asking user has. Lets the answer adjust depth and framing — engineers get technical detail, salespeople get the business-facing version.

    DATA SHAPE
    Per-user role tag, team, seniority.
    IF MISSING
    Answers stay neutral in framing. Works fine for most queries; only noticeably worse when the same content needs different framings for different audiences.
    COMMONLY FILLED BY
    • HR system role lookup
    • directory or org chart
    • user profile in the firm's working tools
RUNTIME FLOW · 8 STEPS
  1. 01
    User asks a question through the query surface
    user_query_surface
  2. 02
    Resolve user permissions across the indexed corpus set
    permission_resolver
  3. 03
    Search the permissioned subset of the corpus via semantic search
    knowledge_corpus_set
  4. 04
    Look up user role context to tailor answer framing
    user_role_context
    DECISION Skip if user_role_context not filled.
  5. 05
    Synthesize an answer grounded on retrieved passages, with inline citations
  6. 06
    If confidence is low or no good sources found, return 'I don't have a good answer for that' with what was tried
    DECISION Honest don't-know is better than confident wrong.
  7. 07
    Deliver the answer to the user through the answer destination
    answer_destination
  8. 08
    Capture feedback if provided
    feedback_capture
EMISSIONS · 3

Structured outputs this pattern produces. Other patterns and client systems can subscribe to them, which is how the catalog composes over time.

  • query_pattern_signal

    What people actually ask about, aggregated. Tells you where knowledge gaps and documentation pain points are.

    CONSUMED BY
    • knowledge management workflows
    • documentation prioritization
    • training material development
  • stale_content_signal

    Sources that get cited but get flagged as outdated, candidates for refresh.

    CONSUMED BY
    • wiki maintenance workflows
    • content owner alerts
  • answer_quality_signal

    Per-query feedback aggregated, used to tune retrieval and synthesis.

    CONSUMED BY
    • pattern quality dashboards
    • prompt refinement