AI Agent Overview

The SEO Agent is the heart of SEO Dash. It is a chat-based AI assistant with access to 80+ specialized SEO tools, the persistent crawl of your site, the change feed, and your GSC / Analytics / Ahrefs data. You can talk to it through the web UI, WhatsApp, Telegram, Slack, or Discord.

How It Works

Under the hood, the agent is a tool-calling loop. When you send a message:

1. Your message goes to the LLM (Claude, GPT, etc.)
2. The LLM decides which tools to call based on your request
3. Tools execute against real APIs (Search Console, Ahrefs, PageSpeed, etc.) and your local data (crawl, fact store, topical maps)
4. Results are fed back to the LLM
5. The LLM synthesizes everything into a human-readable response
6. The loop repeats if more tools are needed

The agent does not guess or make things up. It runs actual checks against your real data.

Key Capabilities

• Multi-step planning - For complex requests requiring multiple tools, the agent outlines its plan before executing. You see what it is going to do before it does it.
• Specialized sub-agents - The agent can hand off to focused specialists: an SEO Audit agent, a Content Strategy agent, or an Analytics agent. Each has a focused system prompt and a filtered set of tools.
• Change history - Every meaningful change to your site (title rewrites, ranking shifts, audit collapses, cross-source disagreements) is recorded with a timestamp. The agent can answer "what changed on /pricing in the last month?" without re-fetching live data, and surface material changes via the What Changed feed.
• Learns what you tell it - Use record_fact (the agent calls this on its own when you state a durable fact) to persist things like "this page targets keyword X", "we focus on local SEO", "don't audit /admin", "our main competitor is example.com". These survive across conversations.
• Site-aware out of the box - When you crawl your site, the agent gets full-text search, semantic search, link-graph traversal, and topical-map context. It answers questions about your own pages without re-fetching.
• Guardrails - Rate limits prevent the agent from running expensive tools too many times. Approval is required before creating monitors, schedules, or making lasting changes.
• Approval workflow - Before taking actions with lasting effects (setting up monitors, creating scheduled reports), the agent asks for your explicit confirmation.

Supported LLM Providers

• OpenAI - GPT-5 family, GPT-4o, etc.
• Anthropic - Claude Sonnet 4.x, Haiku 4.x, Opus 4.x, etc.
• Any OpenAI-compatible API - Ollama, LM Studio, vLLM, or any endpoint that speaks the OpenAI chat completions format

Configure your provider in Settings > AI / LLM. On the hosted SaaS, the managed AI defaults to Haiku 4.5 for cost-efficient runs; bring your own key on any tier for unlimited usage.