Anmelden

How Our Agents Act

MagicAI SEO is not a chat box. Each of our 88 agents plans the work in steps, researches the live web, checks its own report, remembers what it learned, and can act inside your WordPress, Trello, Gmail and Slack — with you in control of every external action. Individuals, teams and agencies can use it: an Agency workspace shares one memory between up to 5 team members.

Search Console: clicks drop Agent starts by itself Plan + live research Auditor review (up to 4 rounds) WordPress draft + Telegram alert
1

Planning & task decomposition

Agents turn a goal into a short plan before doing anything.

  • A planner splits each task into 3–5 steps (search, read a page, analyse, write) and shows the plan live while it runs.
  • Each step receives the verified output of the steps before it.
  • Optional “approve the plan first” mode lets you edit the plan before execution.
  • The AI router reads a plain-language problem and picks the best agent for it.
2

Autonomy & event triggers

Agents start without a prompt when a schedule or a live signal says so.

  • Search Console autopilot: once a day the last 7 days of every page are compared with the 28 days before; a click drop above your threshold (default 15%), a 3+ place position loss or a page leaving the results starts the Content Refresh agent automatically.
  • Scheduled runs: agents and multi-agent workflows (up to 6 agents) run daily or weekly on their own.
  • Runs continue in the background on the server, even when your phone is locked.

Account → Search Console autopilot · Workflows

3

Real-world actions

Results go where the work happens, not only into a report.

  • WordPress: the agent builds the whole post — clean HTML, SEO title and meta description (Yoast / Rank Math), slug, tags, internal links chosen from your own posts and an AI featured image — and saves it as a draft or publishes it.
  • Trello: task agents create a new list with cards (title, details, due date) on your board.
  • Gmail: email agents save ready drafts in your Drafts folder (compose-only permission: the agent can never send mail).
  • Slack: finished reports are posted to your team channel (title, summary and a link), by button or automatically.
  • Telegram: start agents from chat and receive finished reports and approval links.

Account → Apps / WordPress / Telegram

4

Interoperability & API

Agents plug into the tools teams already use.

  • REST API with API keys to start any agent and fetch the result as JSON.
  • Signed webhooks after every run (HMAC-SHA256), with automatic retries on rate limits or errors.
  • Works with n8n, Zapier and Make; inputs from text, web page URLs, sitemaps and files (PDF, DOCX, CSV, TXT, JSON, code).

API documentation

5

Adaptation & self-correction

Reports are checked and fixed before you see them.

  • An auditor agent (writer + critic + fact-checker) scores each draft; failing drafts go back to the writer, up to 4 rounds.
  • When the auditor finds missing, unsupported or outdated facts, the agent runs a new live web search and rewrites with the new evidence.
  • If an AI provider is slow or fails, the run switches automatically to the next provider instead of stopping.
  • Money maths (for example India’s FY 2026-27 income-tax slabs) is calculated by code, not guessed by the model.
6

State & shared memory

Your workspace remembers, so agents do not start from zero.

  • Shared memory: key facts from finished research (prices, competitors, numbers, dates) are saved automatically and offered to your other agents later — for example a cold-email agent using prices a research agent found.
  • Brand memory: your brand, audience, tone and words to use or avoid are applied to every run.
  • Site memory: titles of your recent WordPress posts are remembered to avoid repeat topics and to add internal links.
  • Your own documents in the Knowledge Vault are searched before agents answer.

Account → Brand Memory · Knowledge Vault

7

Reliability & live grounding

Answers are based on current sources you can check.

  • Research agents search the web and read pages in full, with today’s date given to every run.
  • Every research report ends with its source links.
  • Reports include key-fact cards, tables, charts, diagrams and maps that can be exported to PDF, Word, Slides, Google Docs and Sheets.
8

Operator control (human in the loop)

Nothing leaves the platform without the setting you chose.

  • WordPress posts are saved as drafts by default; direct publishing is an opt-in setting.
  • Gmail access is draft-only; hands-free posting, cards and drafts must be switched on per user.
  • The plan, every step and the live progress of each run are visible, and plans can be approved before they run.
9

Data safety

Connections are stored securely and can be removed any time.

  • WordPress passwords, Trello keys and Google tokens are encrypted (AES-256-GCM) before storage.
  • Google permissions are minimal: read-only Search Console and compose-only Gmail.
  • Every connection has a one-click disconnect; history entries can be cleared by the user.
10

Teams & agencies (multi-user)

One workspace for the whole team, not individual accounts only.

  • Agency / Team plan: the owner invites up to 5 members by email or by a private, single-use invite link (valid 7 days).
  • Shared workspace: every member uses the same Brand Memory, Knowledge Vault, shared agent memory and report history; each history entry shows who ran it.
  • Role-based access: members run agents, create reports and add documents; only the owner manages billing, API keys and app connections (WordPress, Search Console, Slack, Trello, Gmail, webhooks).
  • Members send reports through the owner's connected apps, so the agency's WordPress, Slack and Trello stay in one place; the owner can remove a member at any time.

Account → Team / Agency · Pricing

Recent agentic releases

  • v19 · Nov 2026 — Team / Agency workspaces: owner + members, shared memory and history, owner-only billing and app connections.
  • v18.2 · Nov 2026 — Slack integration, real user ratings, /tools directory.
  • v18 · Nov 2026 — Trello cards and Gmail drafts from finished reports.
  • v17 · Nov 2026 — Shared memory between agents; webhook auto-retry.
  • v16 · Nov 2026 — Google Search Console autopilot (event-triggered refresh runs).
  • v15 · Nov 2026 — WordPress action agent (draft / publish with SEO meta, tags, internal links, image); 4-round self-correction with re-research.
  • v14 · Nov 2026 — What-If calculators, exact tax maths, Slides view, public report showcase.
See an agent workPick any agent, type your task and watch the plan, research and review live.
Try an agent
MagicAI SEO  - Featured on AI Agents Directory