Appearance
Know
Know is Liddr's grounded question-and-answer surface. You ask a question in plain language; Liddr retrieves the most relevant entities from your organization's knowledge graph, synthesizes an answer, and links every claim back to the source it came from.
Unlike a general-purpose chatbot, Know answers only from your connected sources. If the corpus doesn't contain the answer, Liddr says so rather than guessing — and, where you've connected a personal MCP connector, it can reach out to live or personal data to fill the gap.
How grounding and citations work
When you ask a question, Liddr runs a retrieval pass over the knowledge graph (Jira issues, Confluence pages, GitHub PRs, issues and code, Figma designs, the engineering handbook, and any other connected sources), hydrates the top matches with surrounding context, and passes them to the model as the evidence it must reason from.
The answer comes back with citation chips — inline references such as [PLAT-1990] or [ADR-018] that map to the exact entity used. Each chip is clickable and resolves to the source ticket, page, PR, or design.
Citations are verified after the fact. Liddr checks every reference the model emitted against the sources that were actually retrieved. Any citation that doesn't map to a real source is flagged as unverified rather than silently shown, so you can tell a grounded reference from a hallucinated one at a glance.
Read the evidence, not just the answer
The answer is a synthesis; the citations are the truth. When a result matters, open the cited sources and confirm the detail for yourself.
Answer styles
Know can shape its response to the task:
- Thorough (default) — a long-form, essay-style answer that connects ideas across several sources, names specific artifacts, and explains the practical implications.
- Concise — a 2-4 sentence quick reply.
- Email — a ready-to-send draft (a Slack message, email, or announcement) grounded in the same sources but written as a natural message rather than a report.
Multi-turn conversations
Know is conversational. A thread keeps its history, so you can refine and follow up without restating context:
- Ask your first question.
- Read the grounded answer and its citations.
- Ask a follow-up — for example, "Who owns the riskiest of those?" — and Liddr answers with the earlier turns in mind.
- Start a fresh thread with New conversation when you switch topics.
Conversations are saved to your history so you can reopen an earlier thread and continue it. Each turn records the question, the answer, and the citations behind it.
Reaching live and personal data with MCP
The knowledge graph holds org-shared structure, work, and knowledge — it deliberately excludes per-user, record-level, and personal data. When a question needs that kind of data, Know can extend itself through your own MCP connectors.
If you've enabled personal MCP connectors (for example Gmail, Google Drive, Calendar, Slack, or Linear), Know exposes their tools to the model. When the corpus alone can't answer — or the question is clearly about a connected live service — Liddr calls the relevant connector, reads what it returns, and folds that into the answer. This is the "connector unlock": the corpus answers what it can, and your live connectors answer the rest.
Personal connectors stay personal
MCP connectors are scoped to you and use your own credentials. Data fetched through them is used to answer your question in the moment — it is never embedded into the shared corpus and never feeds the learning loop. Answers that drew on a personal connector are marked with a badge so it's clear when live, personal data was involved.
Example questions
The empty state offers starter questions generated from your actual corpus — Liddr looks at which sources have data and suggests prompts that fit. Depending on what you've connected, you might see:
text
What are the biggest open risks or blockers right now?
Which issues are highest priority right now, and who owns them?
Summarize the most significant code changes merged recently.
What do our most recently updated docs cover?If the knowledge base is still empty, the starters point you toward connecting a source instead.
Tips for good questions
- Be specific about scope. "Why was the auth migration delayed?" beats "What's going on with auth?"
- Name the artifact when you have it. Mentioning a ticket key, repo, or doc title steers retrieval straight to it.
- Ask one thing at a time. Follow-ups within a conversation work better than a single compound question.
- Use the right style. Switch to Email mode when you want a draft you can send, not a report.
When Liddr can't answer
If the answer isn't in your sources, Know will tell you plainly rather than speculate. When that happens:
- Connect more sources. The most common cause is that the relevant system isn't ingested yet. Add the connector that holds the answer — see Ingest connectors.
- Enable an MCP connector for live or personal data the shared corpus intentionally doesn't hold — see MCP connectors.
- Give it a moment. If a source is connected but freshly added, its content may still be syncing before it shows up in answers.