AI Agent

An agent that does the work.
By your rules.

"Report this week's orders, email the supplier, put the meeting on my calendar" — your agent runs it with 50+ built-in tools. Every risky step queues for your approval; nothing critical ever runs silently.

Cancel anytime · You choose the autonomy level

Automation ≠ loss of control

The agent passes through the policy engine on every tool call. Not "can do everything but is allowed nothing" — fully authorized but supervised.

Copy-paste work ends

Report compiling, email traffic, calendar wrangling, data transfer — the repetitive 30% of your day turns into scheduled tasks.

Not from scratch every time

The agent distills a playbook from successful tasks and learns rules from your corrections. The second time is faster; by the third it's habit.

Toolset

Two tiers, 50+ tools

Read/prepare tools (T1) are side-effect-free and run freely; write/send tools (T2) require approval according to your autonomy level.

T1 · Preparation tier

Reads, researches, drafts; generates Word/PowerPoint/Excel and images — never writes to your systems.

kb_searchweb_researchdrive_readdata_aggregatedoc_generatexlsx_generateimage_generateschedule_tasksocial_readrecall_history

T2 · Action tier

Sends, writes, updates — queued for approval based on risk.

email_sendgmail_read/searchcalendar_create_eventdrive_uploadhttp_requestdb_queryshopify_mutationwhatsapp_sendsocial_postdelegate_task

Connectors

Google (Gmail, Calendar, Drive), Microsoft, Notion, Postgres/MySQL/MSSQL — credentials live in an encrypted vault, never reach the LLM.

Read the attachment → archive it in Drive

Just say "put the invoices from [email protected] into Drive under Invoices > Company > Month": the agent searches the mail, pulls attachments through a safe quarantine (raw file bytes never enter the model) and uploads them into foldered Drive paths — creating folders as needed. It also reads the text inside the attachment: PDF, Word, Excel, CSV — and because OCR kicks in on scanned PDFs, it can pull the amount off an invoice that is nothing but an image. All file operations in one turn are collected into a single approval card. Turn it into an autonomous task to run every month by itself.

Save the spreadsheet it generated to Drive

Just say "work out what I paid in invoices in 2026 and save it as an Excel sheet in the xyz folder on Drive": the agent reads the mail, does the maths, generates the sheet and puts the file it generated straight into Drive (the upload comes to you for approval). Same for decks: it reads a Drive folder, builds the deck and drops it back in the same folder. The limit: the one-click Drive connection lets the agent see only the files and folders it created itself — so let it create the target folder, or put a fully-scoped Google identity in the Vault.

MCP: bring your own tools

Add your MCP server; the agent auto-discovers your tools and routes them through the same policy engine.

Social media: read + publish

Read Instagram/X/TikTok/Reddit; publish to connected accounts with approval (social_post). A 6,000+ document expertise template teaches platform rules.

Trust model

4 autonomy levels, per-tool risk

You decide how much to trust. Every tool call is evaluated before it runs: capability gate → allow-list → risk class → approval threshold. This is code-level enforcement, not a prompt rule — prompt injection cannot override it.

  • Supervised: only read/draft runs automatically
  • Low / High trust: medium risk auto, critical approved
  • Autonomous: everything auto — except never-silent actions
  • Allow-lists: email and HTTP targets stay within your list
policy engine · decision flow
email_send(to: [email protected])
 ① capability  ✓ T2 enabled
 ② allow-list  ✓ example.com listed
 ③ risk        medium (external send)
 ④ autonomy    supervised → ceiling: safe
 → DECISION: ASK APPROVAL
Memory

An agent that knows you and learns

Three memory layers: it knows who you are (profile), remembers what happened (history summaries) and derives lasting rules from your corrections. It distills a playbook (skill) from successful multi-step tasks — the same job gets sharper every time.

  • User profile: your preferences, your working style
  • Episodic memory: remembers "last week's campaign work"
  • Procedural rules: what you correct twice becomes a rule
  • Fully consent-based; view / delete from the panel (GDPR/KVKK)
PANEL · LEARNED RULES
Always send reports as PDF, not Word
learned from 2 corrections · active
Use a formal tone in supplier emails
profile preference · active
Summarize price lists converted to €
candidate rule · 1 observation
Proactivity

It notices before you ask

You schedule tasks from chat ("every morning at 8..."), and the agent suggests things on its own when needed: the morning briefing sums up your day, what you're following and your commitments — delivered in your timezone, with your approval.

  • Scheduling from chat: hourly → weekly, one-off
  • Event triggers: Shopify webhooks are handled instantly (not polled)
  • Personal morning briefing (delivered to email / WhatsApp)
  • The notice → suggest → approve → act loop
  • Commitment tracking: reminds you of the "I'll get back tomorrow" promises you made
  • Delegation: splits big jobs across sub-agents (delegate_task)
morning briefing · 08:00
Good morning Taner, 3 things need you today:

1. 2 PM supplier call — I summarized
   yesterday's price list, attached.
2. Campaign orders up 18%
   (47 orders yesterday). Stock: Trail-X critical.
3. You told Ali "I'll reply today" —
   draft ready, just approve.
Channels

In the panel, on WhatsApp, by voice

Your agent doesn't live in one place: chat from the panel, hand it tasks over WhatsApp, have results delivered to email/SMS/WhatsApp. Say "Hey SemAgent" — work hands-free with live voice conversation.

  • Two-way WhatsApp: send a task, get the result
  • Live voice (STT + TTS) + wake word
  • Result delivery: email, WhatsApp, SMS
  • Progress messages on long jobs: the agent never "goes dark"
WHATSAPP · 12:41
Summarize today's orders, and flag any critical stock
31 orders today (₺48.2K). Critical stock: Trail-X no. 42 (3 left). Shall I draft a supplier order?
"Hey SemAgent" — voice mode active
Expertise templates

Your agent is an expert from day one

Platform-curated knowledge packs load ready-made domain expertise into your agent. The Social Media Management template ships with 6,000+ official source documents (Meta, X, LinkedIn, TikTok and Pinterest help and policy content) — your agent knows the platform rules better than you do.

Social Media Management6,000+ documents
CoverageMeta · X · LinkedIn · TikTok · Pinterest
UpdatesContinuous (curated crawl)
Security
Code-level policy engine Secrets in an encrypted vault, never to the LLM Prompt-injection quarantine
Trust Center →
FAQ

About the AI Agent

Can the agent accidentally email my customer?
Not with the default settings. Email sending is in the T2 (action) tier, and in supervised mode every send queues for your approval. And if you define a domain allow-list, the agent can never send outside it under any circumstance — that's a code-level block, not a prompt.
What guards against prompt injection? (What if a page the agent reads gives it instructions?)
Content from the web or any external source is marked "untrusted" and passes through a quarantine layer before it ever reaches the planner. For critical actions the policy engine is already in force: even if external content fools the agent, it cannot get past the approval threshold and the allow-list.
What happens if the agent makes a mistake or a tool returns nothing?
The agent doesn't invent reasons: if a tool returns empty or errors, it reports the likely causes (e.g. a provider limit) with diagnostic detail rather than guessing. Multi-step tasks can pick up where they left off (resume) and won't accidentally perform the same action twice (idempotency).
Where are my credentials (API keys, passwords) stored?
In an encrypted vault (Fernet). The tool layer uses secrets at runtime; secret values are never sent to the LLM and are masked in logs.
Does the memory feature collect personal data? GDPR/KVKK?
Memory is consent-based and, by default, has to be enabled by the tenant admin. The learned profile, history summaries and rules can be viewed in the panel and deleted with one click. Details in the Trust Center.
How do I control task cost?
A usage meter in the panel shows real-time consumption; plan limits are a hard ceiling — when the limit is reached no new task starts, so there are no surprise bills. Scheduled tasks draw from the same budget.

Give the first task today.

Type 'summarize this week's orders' - the agent handles the rest and asks you before anything risky.

Start free
Cancel anytime · 5-minute setup