Runtime visibility

Glass-box: workflow phases

Left column shows stack + live metrics. Center highlights the LangGraph execution order inferred from returned fields (not token-level trace).

Eagle-eye architecture

How the system is wired

Orchestration layer of the vpeetla-ai stack — what agents do, before AegisAI decides what they may do.

Pairs with AegisAI (governance), Enterprise RAG (knowledge), and AI Content Factory (application output).

Chat workbench

Operator experience

Intent routingThread memoryDelivery toggles

Orchestration

LangGraph workflows

ChiefPlannerSpecialistsCritic

Knowledge + notify

RAG + channels

7 RAG strategiesSlack/Telegram/WhatsAppAegisAI gateway

Ops

Production proof

workflow_runsObservability/api/v1/ops/metrics

Principal tradeoffs

Decisions with explicit costs

LangGraph over linear chains

Gain — Checkpoints, HITL, and multi-step enterprise workflows

Trade — Higher graph complexity than single-prompt UX

Postgres workflow_runs

Gain — Live ops metrics without Langfuse dependency

Trade — Render Postgres vs pure static demo

Gateway-wrapped notify channels

Gain — Side effects gated like production fleets

Trade — Extra hop through AegisAI for delivery

Mock LLM on free tier

Gain — Always-on public demo

Trade — Response depth ≠ production models

Architecture record

ADRs, case studies, and SLOs

Production metrics

Live operational proof

Loading metrics or API is waking from idle…

Orchestrator phase replay

inferred from response fields

Chief

Planner

Workers

Content

Insight

Critic

Compose

Notify

Run the workflow to see phase-by-phase breakdown from the response.

Run workflow

One message to the Chief; the graph fans out to specialists and composes a final executive report.

Message