A signed-in org gets the full Full-stack SEO workspace: eight workspace modules, 114 integrated agent tools (registry below), and durable assets (wallet, OAuth, chats, flows, boards, CE jobs), all metered and observable, not a disposable chat thread.
This map is generated from the same capability registry as production: 114 tools the agent can actually run. Expand a node to see what sits under each module — human names only, not internal API IDs.
/app)Signed-in org shell, eight modules, one LangGraph runtime, shared wallet, OAuth, KB, MCP, and observability. Regenerate this map: python3 scripts/export_tools_hierarchy_data.py.
Mirrors docs/ARCHITECTURE.md, the same pipeline serves Capability kit, Browser mode, and the Chrome extension side panel.
fast_path.identify_fast_pathSingle-tool prompts with no conditions (sitemap, ranked keywords, SEO tags) execute synchronously, often zero LLM routing tokens.
orchestration.prompt_asks_for_briefContent-brief requests skip generic chat routing and enter brief clarifications → generate_content_brief (deterministic/async).
intent_classifier.classify_intentSemantic intent, action_id, profile, confidence ≥ 0.72 before trusting the label.
intent_router.route_intentPattern fallback when classifier is unavailable or low confidence, brief, audit, export, analytics, conversation.
orchestration.run_orchestrationDeterministic handler (no agent) · scoped LangGraph agent (INTENT_TOOLSETS) · composite fallback (full registry).
FastAPI backend + web workspace + extension, one org API key, shared SSE execution graphs.
Chat/agent SSE, Content Wizard jobs, flow CRUD, billing, labs, metrics snapshots, MCP preflight.
langgraph_agent.py builds scoped tool lists via agent_factory.py, Strategic Decision Matrix in agent.py.
sse.py emits planned steps, tool start/end, and cost hints to the kit UI and extension.
web/app.html shell · Chrome side panel · URL from current tab (never overrides user prompt).
Atomic tools wrap these integrations, OAuth and API keys are org-scoped, not per chat.
OAuth per org; get_gsc_performance_data, get_ga4_traffic_data, property intel.
SERP, keywords, backlinks, AI Overview, multi-LLM content intel, LLM mentions APIs.
model_router.py + llm_presets, fast vs pro tiers by intent; economy picker is filtered to OpenRouter models with tool-calling support only.
peec_client.py, visibility and citation intelligence orchestrators.
Playwright batches (JS dependency), SEO Spider MCP (desktop), logfile upload analysis.
GCP NLP / structure forensics, html_structure_forensics, entity graph labs.
Chroma vector KB, url_grounding, user MCP servers (mcp_client.py, mcp_integrations.py).
Postgres (db.py), PostHog traces, cloud_job_dispatch for long jobs, metric_snapshots weekly store.
backend/orchestration.py, intent, profile, and tool allowlists before any vendor API fires.
INTENT_TOOLSETS in toolsets.py, composite registry has 114 UI features; allowlists trim what the agent sees per intent (meta-tools replace atomic siblings in analytics/audit).
Scoped allowlist from backend/toolsets.py, agent may only call these names when router selects intent analytics.
Scoped allowlist from backend/toolsets.py, agent may only call these names when router selects intent analytics_complex.
Scoped allowlist from backend/toolsets.py, agent may only call these names when router selects intent audit.
Scoped allowlist from backend/toolsets.py, agent may only call these names when router selects intent brief.
Scoped allowlist from backend/toolsets.py, agent may only call these names when router selects intent composite.
Scoped allowlist from backend/toolsets.py, agent may only call these names when router selects intent conversation.
Scoped allowlist from backend/toolsets.py, agent may only call these names when router selects intent export.
Scoped allowlist from backend/toolsets.py, agent may only call these names when router selects intent file.
Invoked directly by deterministic audit routes, never chosen from a 90-item list.
Injected into every allowlist, planned step map for SSE UI.
action_registry + intent_router, brief, export, file, analytics_sitemap, audit_lp_optimization, etc.
intent_contract.EXECUTION_PROFILES, fast (3 iter / 4 tools) · standard · full (10 iter / 15 tools).
content_engine.py + quality_gate + provenance + citations, batch jobs, not kit ReAct.
Ordered tool nodes from flow builder, same tool IDs as the kit.
Tabs in /app, expand for subsystems; Capability kit holds the full registry below.
Org home, wallet balance, recent agent chats, and shortcuts into modules. No separate tool registry here; it orients operators before they open the kit or Browser mode.
Credits, plan seats, and recent api_call_log visibility.
Resume LangGraph threads without losing tool context.
Jump into named flows from the flow builder.
Primary agent chat plus the full tool catalog (90+ capabilities). Most GSC, GA4, DataForSEO, audit, and export tools are invoked from here unless a flow or Content Wizard job owns the run.
Page-level forensics, schema, UX, and NLP. · ANL · 23 registered
Deep site reviews, competitive intelligence, and GEO readiness. · AUD · 12 registered
Briefs, articles, wireframes, and content quality. · CNT · 6 registered
Charts, tables, and document exports. · EXP · 10 registered
Files, URLs, skills, and retrieval. · KB · 11 registered
Your connected Model Context Protocol servers and their tools. · MCP · 3 registered
GSC, GA4, keywords, SERP, and backlinks. · MET · 20 registered
Screaming Frog MCP, local desktop crawl integration. · SF · 12 registered · Full atomic list (composite intent also exposes meta seo_spider).
Sitemaps, health checks, search, and helpers. · UTL · 15 registered
Batch article generation with deterministic pipeline stages, research, draft, citation gate, quality gate, and provenance, rather than ad-hoc ReAct tool picking.
Queued batches with per-article status and retries.
Grounding checks before drafts leave the org.
Scoped assistant for revising a single draft in CE context.
Drag-grid boards for QBR-style narratives, charts and tables fed by tool outputs, not a second analytics warehouse.
Resize, reorder, and persist layouts per org.
Bind agent or flow exports into visual modules.
Experimental surfaces (entity graph, RAG simulation, narrative consistency, keyword clustering), same backend wallet, flagged for early adopters.
Explore on-site entity relationships.
Test retrieval behavior against your KB.
Cross-page story alignment checks.
Visual graphs of repeatable work, each node maps to an atomic tool or orchestrator, executed deterministically when you run the flow.
Same IDs as the kit; wired as ordered steps.
Named flows persist per org for operators and Chrome extension.
Org skill registry plus coded skills the agent can invoke, extends the catalog without forking the core repo.
Prompt + tool bundles registered per org.
Author or patch coded skills with guardrails.
URL-grounded side agent, page context from the active tab (extension or workspace) drives analyze_page and related meta-tools.
Live DOM / URL context passed into the agent.
Chrome extension uses the same backend and wallet.
Multi-tenant Postgres, wallets, chats, flows, boards, CE jobs, OAuth tokens, API keys.
api_call_log metering, checkout, seats, see billing/api.py.
Agent chats, flows, Report Canvas boards, CE drafts, Labs snapshots, skill registry.
PostHog + OpenRouter cost capture + SSE mirrors for operators.
Generated from web/js/features.js (UI catalog), each row is an invokable backend function in TOOL_CATEGORIES / main.py tool registry.
Page-level forensics, schema, UX, and NLP. · ANL · 23 registered
Deep site reviews, competitive intelligence, and GEO readiness. · AUD · 12 registered
Briefs, articles, wireframes, and content quality. · CNT · 6 registered
Charts, tables, and document exports. · EXP · 10 registered
Files, URLs, skills, and retrieval. · KB · 11 registered
Your connected Model Context Protocol servers and their tools. · MCP · 3 registered
GSC, GA4, keywords, SERP, and backlinks. · MET · 20 registered
Screaming Frog MCP, local desktop crawl integration. · SF · 12 registered · Full atomic list (composite intent also exposes meta seo_spider).
Sitemaps, health checks, search, and helpers. · UTL · 15 registered
SVC runtime / platform ORCH multi-step recipe META cluster router ATOM single capability MOD workspace tab INT external integration
Sign-in unlocks a single org-scoped workspace at /app, not a demo chat. Every module below shares the same agent runtime, wallet, OAuth connections, and observability. You do not re-authenticate per tool; you do not copy API keys into one-off scripts.
The full capability map is in the hierarchy diagram above, Google Search Console and GA4 (OAuth), DataForSEO, crawl and NLP, exports, tech audits, content research, and GEO performance checks, invoked through LangGraph with live SSE execution graphs.
Dashboard (wallet, chats, flows), Capability kit (catalog + chat), Content Wizard (batch articles + quality gates), Report Canvas (boards), Labs (experiments), Flow builder (repeatable graphs), Skills (registry + code agent), Browser mode (URL-grounded side agent). Chrome extension side panel targets the same backend.
Usage wallet and billing visibility, saved agent chats, named flows (local + server sync), Report Canvas boards, Content Wizard jobs and drafts, Labs snapshots, model defaults per seat, GSC/GA4 connections, API workspace key, and production tracing (PostHog + SSE). Data is org-isolated, not shared across clients unless you invite seats.
Deterministic flows for weekly audits, exploratory kit chat for one-offs, Browser mode for on-page reviews, Content Wizard for content programs, Report Canvas for QBRs. Ambient explain overlays and tutorials document the UI; these tool pages document capability for buyers and team leads. Metering is transparent on pricing; stack detail on stack.
See org wallet (credits), API health, last chats, and saved flows in one place. Quick-route into Content Wizard, Browser, Report Canvas, Labs, or Flow builder. Default landing tab for operators and team leads.
Read overview →Browse 90+ tools by category, run kit chat with any OpenRouter preset, and watch LangGraph fan out GSC, GA4, and DataForSEO in one thread. Every call is metered and logged, production SEO ops, not a plugin.
Read overview →Sheet-style topic batches, SERP-aware research, brand-voice drafting, citation panels, quality gates, and edit chat for surgical rewrites. Outputs stay in your workspace with export paths to CMS workflows.
Read overview →Pin GSC/GA4 slices, tool snapshots, and tables to a grid board. Reuse layouts across clients by swapping properties and date ranges. Built for QBRs and weekly monitoring, not one-off answers.
Read overview →Cluster keywords into intent groups, map entity graphs with crawl + NLP + Wikidata, simulate RAG context, and compare brand narrative drift. API-backed steps meter through the org wallet.
Read overview →Visual tool graph with validation, auto-save, server sync when signed in, and run-from-chat via flow ID. Mirror weekly audits as playbooks instead of re-prompting the agent each time.
Read overview →Browse enabled skills, version metadata, and triggers. Code Agent authors Python skills; planner invokes them as single contracts. Complements flows (user graphs) and MCP connectors.
Read overview →Split view: live page plus contextual agent. Each message carries the active URL; saved flows run without copy-paste. Ideal for staging QA, competitive reviews, and on-page SEO.
Read overview →They are public marketing depth, signed-in users get live tooltips and ambient explain overlays in the product.
Dashboard → Capability kit → Browser mode matches most SEO workflows; content teams lead with Content Wizard.
Module access follows plan seats and beta flags; metered tools always respect org wallet balance.
Plans and credit math live on pricing; tools pages describe capability, not contract limits.