A working model of the real builder — click the steps, rename the agent, toggle sources and tools.
Knowledge and actions are separate on purpose
Most agent builders blur the two and you lose track of what your agent can touch. Klart keeps the line sharp.
Datasources — what it knows
Read-only. The agent answers from these and cites them; when the answer isn’t there it says so instead of inventing one.
Drive, Notion, Confluence, Jira, your website, uploaded files
Existing permissions are inherited, never widened
Re-indexed as content changes — no stale answers
Tools — what it can do
Write access, granted one action at a time. Every tool can require your approval before it runs.
Create tickets, update CRM records, send email, file issues
Any internal API: endpoint, auth header, JSON parameters
Every call logged with who asked, what ran, what came back
One agent, three surfaces
Build it once. Your team uses it in chat, your product calls it over the API, your customers meet it on your site.
Your team just picks it from the list
A published agent shows up in workspace chat next to Klarty and the rest. It answers with citations, and when it runs a tool you see exactly which one ran.
Cited answers, so anyone can check the source
Tool receipts in the thread, not hidden in a log
Also reachable from Slack and Microsoft Teams
What teams build first
Three custom agents that pay for themselves in a week.
Onboarding agent
Answers every new-hire question from the handbook and IT wiki, and files the access requests itself.
Tier-1 support agent
Sits on your site, resolves the repeat questions, and opens a ticket with context when it can’t.
RFP response agent
Drafts answers from past bids and security docs, cites each one, and logs the deal in the CRM.