Client → API → operation
Browser or extension opens an SSE stream. The API authenticates, meters credits, routes intent, then fans out to agent tools or long-running Content Wizard jobs.
Stack map
Operations wired to real technologies
Every product surface (agent chat, Report Canvas, Content Wizard, drag-and-drop flows) routes through a single Flask API. Packets on the lines show live request paths.
Operations
API core
Integrations
Animated traffic
Web & extension
React shell · SSE client
Flask API
GCP Cloud Run · SSE hub
Intent router
Fast-path vs agent
LangGraph agent
Plan · execute · synthesize
90+ tool nodes
Atomic · orchestrator · meta
Content Wizard
Sheet jobs · Word export
Flow builder
DnD pipelines · skills
Skills registry
Versioned · Code Agent
Metered billing
Org wallet · api_call_log
OpenRouter
Model gateway · economy roster
LLM providers
Claude · Gemini · GPT
DataForSEO
SERP · keywords · links
Google APIs
Search Console · GA4 OAuth
Postgres
Skills · chats · billing
Observability
PostHog · trace IDs
How to read the map
Parallel product surfaces
Flows, skills, and the capability kit share the same backend. Saved flow definitions persist inputs and wire order, not one-off prompts.
Every call logged
DataForSEO, OpenRouter, and Google calls write to api_call_log with actual or computed cost before the UI shows completion.
FAQ
Stack map questions
How to read the live diagram and what the node types mean.
What does the stack map show?
How clients hit the Flask API, how intents route into tools and jobs, and which providers settle cost, not a marketing collage.
What are atomic vs agent vs job executions?
Atomic = one tool call. Agent = LangGraph tool loop with SSE phases. Job = long-running Content Wizard / audit work on a daemon thread.
Where do costs get recorded?
OpenRouter, DataForSEO, and Google calls write to api_call_log before the UI marks completion, wallet balance follows settlement.
Want provider logos and explainers?
See the Technologies page for chip-level detail on each integration.