All Things Agentic · Track 2 — The Collaborative Partner

ConsultingDAG

Your autonomous co-founder agent. Give it a repo or an idea and a Gemini multi-agent executive team plans, collaborates with you, and builds a full business blueprint — CTO architecture, CFO model, CMO go-to-market.

Prompt → executive team

Hand it a GitHub repo or a startup idea. The Google ADK + Gemini orchestrator plans a DAG of specialist agents — CTO, CFO, CMO, and more — tailored to your asset.

You stay in the loop

This is a collaborative partner, not a black box. Agents pause to ask the real strategic calls — pricing, native vs web, raise vs bootstrap — you decide, they build.

Real work, not mockups

Each agent runs on Gemini via the ADK with tools (repo analysis, financial math) and produces a real deliverable — tables, diagrams, slide decks, even a generated ad poster.

Enterprise-grade

Real ADK Model Armor guardrails, a persistent Memory Bank, and a full reasoning trace — plus a no-key heuristic mode so it's fully demoable offline.

How it works

A collaborative executive DAG

The orchestrator plans a directed acyclic graph of specialist agents from your asset. It runs top-down — analyst first, then the executives build on each other — pausing whenever a decision is really yours to make.

Asset
repo · idea
Orchestrator
plans the team
Analyst
market / repo
CTO
architecture
CFO
pricing · model
CMO
go-to-market
Blueprint
report
Orchestrator — plans & routes Specialist agent — a Gemini executive Blueprint — the compiled report
Under the hood

Gemini, the ADK — with real guardrails & memory

Gemini via the Google ADK

One Gemini model powers both the orchestrator that plans the DAG and every specialist agent that executes — driven through the ADK's LlmAgent + runner, with FunctionTools for repo analysis and financial calculation.

Model Armor guardrails

Real ADK before/after-model and before-tool callbacks block prompt injection, redact secrets/PII from output, and veto unsafe tool calls — the platform's guardrail concept, running locally.

Persistent Memory Bank

A SQLite-backed ADK BaseMemoryService gives the team durable cross-session memory: each agent writes its deliverable and later agents recall relevant prior work — surviving restarts.

React Flow + Next.js + Zustand

A React Flow canvas with a dagre layout renders the live executive DAG and per-node status; a Zustand store drives the topological run; Next.js (App Router) serves the UI and API routes.

StackNext.jsReact FlowGoogle ADKGeminiZustandSQLiteTailwind CSS
Architecture

From a known asset to a business blueprint

Full detail in the docs — here's the shape end-to-end.

Known asset
a GitHub repo, or a startup idea in plain English
Orchestrator agent · ADK + Gemini
plans a DAG of specialist agents (JSON), hardened by Model Armor
Executive DAG on the canvas
React Flow · orchestrator → analyst → CTO / CFO / CMO …
Per-node execution · Gemini via ADK
Human-in-the-loop
agents pause for your strategic calls
Model Armor
ADK guardrail callbacks
Memory Bank
persistent cross-session recall
Business Blueprint
per-node artifacts (tables · diagrams · slides · ad poster) → one report

Meet your co-founder agent

Point it at a repo or describe your idea, weigh in on the calls that matter, and get a founder-ready blueprint — demoable even without an API key.