Product · overview

Answers it won't invent.
No action without your OK.

SemAgent does two things: a widget you install on your site in 5 minutes answers only from your knowledge base; your AI agent runs work in email, calendar, Shopify and WhatsApp with your approval. We solved the two fears about AI with architecture.

Fear 01 — "What if it makes things up?" It speaks only from your knowledge

Strict-KB grounding: if the answer isn't in your knowledge base it won't invent one — it says "I don't know" and shows its source.

See the mechanism →
Fear 02 — "What if it causes trouble?" It does nothing without your approval

A code-level policy engine stops every risky action. The "never-silent" class asks for approval even in full-autonomy mode.

Trust architecture →
Start free Explore the product ↓
No credit card · 7-day free trial · 250 credits
Who are you?

Start from your scenario

Instead of one generic tour, go straight to your job — each path is for a real department or workload.

E-commerce

End "where is my order?"

The widget answers order status live from Shopify; the rest comes from your knowledge base.

E-commerce solution Customer Support

Hand off the routine

Answer repeat questions 24/7 with sources; let the team focus on the hard cases.

Support solution Enterprise / Department

An assistant per department

From HR to Legal, every team gets an assistant that speaks from its own data.

Enterprise solution Knowledge Base

Consolidate scattered knowledge

Turn docs, wikis and PDFs into one searchable, sourced answer layer.

Knowledge Base solution Autonomous Ops

Give the routine to the agent

Let the approval-gated agent run definable repetitive work on its own.

Autonomous Ops Social Media

Analyze, approve, publish

Content management with an agent that knows platform rules from 6,000+ official docs.

Social Media solution
Two products · one knowledge base

Need an answer, or an action? Both are here.

Product 01 — RAG Widget

An assistant that speaks with your knowledge, on your site

It crawls your website, forum and documents; answers visitors only from your knowledge base, showing sources.

  • 6 crawl engines: sitemap BFS + 5 forum engines
  • Strict-KB grounding — says "I don't know" when it doesn't
  • Live Shopify order/stock answers
  • Persona, color and look that match your brand
Explore RAG Widget →
Product 02 — AI Agent

A digital worker that runs tasks on your behalf

With 50+ built-in tools it sends email, manages the calendar, produces reports, updates Shopify, publishes on social — asking for your approval at every risky step.

  • Gmail, Calendar, Drive, Notion, WhatsApp, MCP…
  • 4 autonomy levels + per-tool risk class
  • Task scheduling, morning briefing, learning memory
  • Voice: "Hey SemAgent" (Enterprise)
Explore AI Agent →
How it works

Live in 3 steps

Build the knowledge base

Give your site address — SemAgent crawls your sitemap, forum and PDFs and builds a clean knowledge base with smart chunking and de-duplication.

Paste the widget

Add the single-line embed code; set persona, color and greeting from the panel to match your brand. Typical setup is 5 minutes.

Authorize the agent

Connect connectors, choose the autonomy level. The agent works and comes to you for approval on risky steps.

Under the hood

Benefit first. Detail on demand.

The title says what you get; expand to see how it works. For the curious, each line leads to a deep-dive page.

6Multi-engine crawlTurns every corner of your site, forum included, into a clean knowledge base.
  • Sitemap BFS + 5 forum engines: phpBB, inSided, Discourse, XenForo, vBulletin (auto-detected)
  • Stealth crawl for Cloudflare-protected sites; checkpoint/resume for resilience
  • A reason log for every crawled page — what was taken and why is traceable
Knowledge base & crawl →
✓Grounded answer qualityClosed to hallucination: only from the KB, with sources.
  • Strict-KB mode: if the answer isn't in the KB it says so and routes to a human
  • A source indicator under every answer
  • Anti-confabulation: says "none" is "none" even under pressure
RAG Widget →
50+Agent tool setTwo tiers: reading is free, actions need approval.
  • T1 · Prep (read/research/generate): kb_search, web_research, document/slide/sheet/image generation…
  • T2 · Action (write/send): email, calendar, Drive, DB, HTTP, social publishing…
  • Connect your own MCP server — your tools pass through the same policy engine
Tools & MCP →
3Learning memoryIt knows you; you don't explain from scratch daily. All consent-based.
  • Three layers: user profile + episodic summaries + persistent rules from repeated corrections
  • Visible in the KVKK/GDPR panel ("Learned Rules"), deletable with one click
  • Distills a playbook (skill) from successful tasks — faster the second time
Memory →   Skills →
08Scheduling & proactivityJust say "every morning at 8" — no cron.
  • Natural-language scheduling; event triggers (a new Shopify order is handled instantly)
  • A personal morning briefing sums up your day
  • Notice → suggest → approve → do flow
Scheduling & proactivity →
🛡Security & controlThe agent is never loose — a code boundary, not a prompt.
  • 4 autonomy levels + per-tool risk class + domain allow-lists
  • "Never-silent" actions ask for approval even in full-autonomy mode
  • External-content quarantine against prompt injection; tenant-isolated data
Trust & Security →
50+
built-in agent tools
6
crawl engines (sitemap + 5 forum)
24/7
health monitoring + auto-failover
5min
widget setup time
Case · Tonmode

~64% of support messages never reach a human anymore

Made-to-order clothing brand Tonmode answers order questions live from Shopify with RAG Widget + AI Agent; the founder starts the day with the agent's morning briefing.

"'When will my order arrive' was half the day. The widget answers that live from Shopify."

Read the Tonmode case →

Figures are approximate (~) and reflect a single customer's experience; the widget dialogue is representative.

~64%
messages resolved without a human
Instant
first response (before: hours)
~2 hrs
weekly reporting effort removed
1 day
from setup to live
Real screens

Not talk — from the real product

We prove the two fears with the product's own screens, not mockups. (Screens are from the Turkish-first product UI.)

Agent approval card: an action awaits your approval — Approve / Reject
No unapproved actionIt stops before a risky step. Even a "low"-risk task comes to you for approval; nothing runs until you click "Approve".
Agent works grounded from the knowledge base; with no connector it won't fake sending — it presents an honest draft
Won't make things upGrounded and honest. The agent works from your knowledge base (kb_search); with no email connector it won't claim it "sent" — it gives the draft and states the situation clearly.
Trust is not a badge; it's an architecture
Tenant-isolated data No model trained on your data GDPR/KVKK-aligned memory
Trust Center →

Platform architecture

Crawl → knowledge base → strict-KB retrieval → Widget + Agent → tools and channels. See how one architecture fits together. Explore the platform →

Honest comparison (TR)

Compare SemAgent with Chatbase and Intercom Fin line by line, at real prices — no spin. See the comparison →
FAQ

Questions on your mind

Can the widget give my customers wrong / made-up information?
No — in strict-KB grounding mode the widget answers only from the content in your knowledge base and shows its source. If the answer isn't in the knowledge base, it says so instead of making something up and routes to human support.
What can the agent do "on its own" on my behalf?
Whatever you allow. Every tool call passes through a code-level policy engine: you choose one of 4 autonomy levels, risky actions require your approval, and "never-silent" actions never run without approval at any level.
How long does setup take, do I need technical knowledge?
You give your site address and start the crawl; the widget is added with a single line of embed code. Typical setup is 5 minutes and requires no coding. Documentation is ready for teams that want deeper API integration.
What's in the free trial, is a credit card required?
The trial plan lasts 7 days: with 250 credits and 300 pages of crawling you can take the widget live. We don't ask for a credit card; when time or credits run out, new requests stop and your data isn't deleted.

Set it up today, notice it this week.

Your widget is live in 5 minutes and your agent is working on day one. Try it free for 7 days — 250 credits, no card.

Try it free
No credit card · sign up in 2 minutes · cancel anytime