Skip to main content
    PRODUCT · AI CHAT

    Ask across company knowledge. Verify every answer.

    Search approved mail, files, conversations, tickets, code, and internal systems from one surface. Linkence synthesizes the answer with citations and keeps governed execution behind a visible handoff.

    Read-first by design. Sources stay visible. Scope follows the configured workspace and connector model.
    Starts withA natural-language question
    Grounds inSelected, approved connected sources
    ProducesA synthesized answer with citations
    BoundaryRead-first; supported writes hand off to Actions
    THE PRIMARY JOB

    One question should not require five tabs and tribal memory.

    AI Chat retrieves the configured evidence, reconciles exact identifiers with natural-language context, and returns one answer a user can inspect. Existing systems remain the source of record, and a question never becomes an invisible write.

    1. 01

      The answer is split across inboxes, files, tickets, and conversations.

    2. 02

      People find different versions and cannot tell which source is current.

    3. 03

      A summary without citations is difficult to trust or hand to another team.

    SEE IT WORK

    A cited summary shows the Friday configuration patch, the Monday implementation note, and the missing owner on ENG-731.

    Fictional demo for Northstar Services, Atlas Retail, in Project Atlas. No external system is changed.

    LAI ChatProject AtlasCited answer
    Demo data · Northstar Services
    Question

    What did we promise Atlas Retail about the SSO fix, and what is still open?

    GmailSlackJiraConfluence
    Cited answer

    Northstar committed to a configuration patch by Friday and an implementation note before Monday's review. The patch is linked to ENG-731, while the implementation note still has no owner.

    Gmail · Source 1Atlas Retail escalation

    Configuration patch due Friday. Implementation note required before Monday review.

    HOW THE PRODUCT WORKS

    Three visible parts of AI Chat.

    01

    Follow every answer back to evidence.

    Citations identify the document, thread, ticket, or conversation behind each material point so the reader can inspect the source.

    UnderstandSentence-level citation controls and a compact source drawer
    02

    Choose the scope the question needs.

    A user can narrow the active source set instead of searching every connected system for every request.

    UnderstandVisible selected scope with source identities
    03

    Carry verified context into reviewed work.

    When the next step needs a supported write, the answer and its cited source bundle move into an Actions preview.

    UnderstandRead-first handoff with no automatic execution
    AI CHAT TO ACTIONS

    A cited answer can become reviewed work.

    Send the verified source bundle into an Actions preview. The exact target and payload stay visible before any supported write can run.

    Explore Actions
    Verified context
    A synthesized answer with citations
    Actions
    A user request or product handoff
    AI CHAT CONTROL BOUNDARY

    The product handles the work. Your team keeps the decision.

    Linkence handles

    • Retrieve from configured connected sources
    • Synthesize the answer and preserve citations
    • Keep exact source identities visible
    • Prepare a handoff when the user requests work

    Your team controls

    • Connected sources and active scope
    • Whether the evidence is sufficient
    • Interpretation, priority, and business judgment
    • Whether a supported write proceeds in Actions

    Not designed for

    • Invisible system changes
    • Replacing a source of record
    • Making legal, HR, security, or engineering judgment
    AVAILABILITY AND LIMITS

    What is available, and where the boundary sits.

    available

    Required setup

    • At least one supported source connected to the workspace
    • Source scope configured by an administrator
    • Search access granted to the person asking

    Known limitations

    • Available sources, freshness, and access behavior vary by connector.
    • A citation proves the retrieved source, not the correctness of every interpretation.
    • Supported writes are handled by Actions, not directly by the read-first chat surface.
    Answers display citations to retrieved source material.Verified product claimRetrieval uses the configured source scope available to the request.Verified product claim
    OPEN, REPRODUCIBLE RETRIEVAL EVALUATION

    96.6% strict section accuracy@10 on a 500-query Open RAG Bench sample.

    The exact gold section appeared in the top ten results for 96.6% of the stratified sample. On all 3,045 valid queries, strict section accuracy@10 was 94.6%. Raw per-query results, manifests, failures, and a verification script are public.

    Review methodology and results
    Stratified sample96.6%

    500 queries · Strict section accuracy@10

    Full valid set94.6%

    3,045 queries · Strict section accuracy@10

    Strict section accuracy@10The exact gold document and section appears within the top ten retrieved units.

    Retrieval evaluation only. It does not measure answer correctness, workflow execution, production latency, or every customer corpus.

    CONCRETE WORKFLOWS

    Three ways teams use AI Chat.

    Customer escalation

    A question about promises and open work

    GmailSlackJiraConfluence

    A cited timeline, owners, and unresolved items

    Prepare a reviewed follow-up in Actions

    Policy lookup

    A question about the current policy

    SharePointDriveConfluence

    An answer with the exact source section

    A person validates the exception path

    Engineering history

    A question about a technical decision

    JiraGitHubConfluenceSlack

    A cited decision narrative

    An engineer decides the formal work
    AI CHAT QUESTIONS

    Direct answers about AI Chat.

    Availability and connector-specific behavior are stated directly, without universal claims.

    Will AI Chat send email or change tickets on its own?+

    No. AI Chat is read-first. A supported write moves into Actions, where the target, payload, and required approval remain visible.

    What does the retrieval benchmark measure?+

    It measures whether the exact gold section appeared in the retrieved top ten. It does not measure answer correctness or every customer corpus.

    Does every connector use the same permission model?+

    No. Scope and access behavior differ by connector. Each integration page states the verified model for that system.

    SEE AI CHAT ON YOUR DATA

    Start with one real question or workflow.

    Connect the relevant sources and show Linkence the approval rules. See the path from evidence to follow-through on your own data.

    Representative data is enough for the first demo. No production access required.