Mock Demo — Multi-Agent Orchestration

Ask a Sports Question.
Watch the Agents Work.

This is a simulated pipeline. Where we lack API access, we explain exactly what WOULD happen with real credentials. Every step maps to a production agent in the stack below.

Giannis playoffs Warriors dynasty Jordan vs LeBron Best rookie debut
Live Pipeline

Agentic Processing — Step by Step

Each step fires a different agent in the orchestration stack. Click "Ask" above to watch the pipeline animate. Red notes show where real API access would change the behavior.

1
Hermes — Intent Router
Classifying query type and dispatching agents
Query analyzed across 136+ skill classifiers. Detected: player search, playoff context, narrative style. Routing to parallel agents. ⬤ MOCK: In production, Hermes routes through a WebSocket gateway with sub-50ms classification. 136 specialized skill definitions live in a PostgreSQL-backed skill registry updated via continuous deployment.
2
Paperclip — Parallel Agent Dispatch
Spawning 3 specialized sub-agents simultaneously
Agent A: Stats — queries structured game data.
Agent B: Narrative — searches story content for clutch moments.
Agent C: Media — locates highlight video metadata. ⬤ MOCK: Paperclip would spawn these agents as isolated Docker containers with scoped tool access. Each agent gets a time-boxed session. Results merge via a reducer pattern before the user sees anything. In this demo, we simulate the merge logic locally.
3
MCP Servers — Tool Bridges
Connecting agents to external systems
Storyteller SDK — queries story metadata index.
NBA Stats API — retrieves structured game/player data.
Content Library — pulls full story text for synthesis. ⬤ MOCK: In production, MCP servers run as persistent processes bridging Hermes to external APIs. The Storyteller SDK MCP server would expose story search, metadata retrieval, and content delivery as typed tools. The NBA Stats MCP server would map stat queries to the official NBA API. We describe the interaction as if connected.
4
LLM Synthesis — Grounded Answer
Merging agent results into a coherent response
Stats agent returned 12 relevant games. Narrative agent returned 8 stories. Media agent returned 5 highlight reels. Synthesizing with grounding: every claim must cite a source story. ⬤ MOCK: The synthesis prompt enforces source citation. No sentence appears without a reference. In production, Claude Sonnet or GPT-4o runs the merge with a structured output schema — every claim is a JSON object with text, source_story_id, and confidence score.
5
Audit Trail — Immutable Log
Recording every decision for verification
SHA-256 hash chain updated. Query logged. Agent decisions logged. Source citations recorded. Content partner can verify their stories are surfaced correctly. ⬤ MOCK: In production, every agent action writes to a hash-chained audit log. The chain is independently verifiable — no black-box algorithm. Content partners receive a verification endpoint showing exactly which queries surfaced their stories and how they ranked.
Results

Synthesized Stories — Ranked & Grounded

What the fan sees. Every result cites its source. Every ranking is traceable to the agent that selected it.

Production Architecture

The Stack This Demo Simulates

Intent Router
Hermes
→
Delegation
Paperclip
→
MCP Bridge
3 Servers
→
Synthesis
Claude / GPT
→
Audit
SHA-256
Integration Map

What Connects Where — Real vs Mock

IntegrationWhat It DoesStatus
Storyteller SDK Story metadata search, content retrieval, player/team tagging Mock — describes interaction as if connected
NBA Stats API Structured game data, player stats, season aggregations Mock — describes interaction as if connected
Hermes WebSocket Gateway Real-time intent routing, skill registry, tool execution Live — running on production VPS :8642
Paperclip Orchestrator Multi-agent delegation, parallel execution, result merging Live — running on production VPS :3100
Agent Zero Autonomous Loop Self-improving embeddings, content gap detection, FAISS memory Live — running on production VPS :50080
PostgreSQL Story index, agent state, skill registry, audit log storage Live — production database
SHA-256 Audit Chain Immutable log of every agent decision, independently verifiable Live — deployed on all production systems