Product Vision & Evolving Map

Build the AI operating system that turns goals into outcomes — across every domain of work and life.

neww.ai is one workspace, one agent layer, and one data spine spanning 30 verticals. We do not build a smarter chat; we build the operating system that closes the loop between a user's goal and the artifact, transaction, or decision that satisfies it.

FinanceSearchMarketingCRMAgentsCodeDataCommerceLegalHealthcareRecruitingOperationsRevOpsProductivityGrowthSecurityCloudDevOpsSupportKnowledgeAnalyticsCompliancePersonasSkills

24+ domains. One identity, one data model, one agent layer.

The Thesis

Why neww.ai exists

The previous era of software was a fight over interfaces. Every category picked its winner — a CRM, an inbox, a spreadsheet, an editor — and a knowledge worker ran 10 of them in parallel. Each tool owned a sliver of the user's context and none of them could act for them.

The AI era is a fight over outcomes. A user does not want a better chat with their CRM; they want the deal closed. They do not want a faster query in their finance app; they want the books closed and the taxes filed. The companies that win the next decade will be the ones whose agents take the goal and return the artifact — with the receipts.

That product cannot be a chat layer on top of legacy SaaS. It must be an operating system: shared identity, shared data, shared memory, shared tools, shared billing. Every domain reachable from one prompt. Every action verifiable. Every workflow re-callable. We are building that operating system. We started with the wedge — finance, search, marketing, CRM, agents — and we are extending it across every category of work.

Six Pillars of the Domain OS

Each pillar is shipped or actively in flight in the codebase today — not a roadmap promise.

Domain OS

Not 30 SaaS apps glued together — one operating system with 30 verticals (15 consumer, 15 enterprise) sharing one identity, one data model, one billing surface, one agent layer. Switching domains is a context shift, not a re-login.

Master Dispatcher

A single goal-routing brain. The user states an outcome; the dispatcher decomposes it, picks specialist agents and tools, runs a verified plan, and reports artifacts. Streams a typed DAG plan with budget caps before any tool fires.

OSS Runtime

53 production-band engines wrapping best-in-class open source — Meilisearch, Qdrant, Crawl4AI, Firecrawl, Twenty CRM, Medusa, Lago, MLflow, Ollama, vLLM. We integrate, not rebuild. Cost per call collapses; capability per call compounds.

Agent Memory

Cross-domain memory is the moat. Lessons from a finance run inform the next legal task. Procedural memory mines repeated workflows into reusable skills. Tier-graded retrieval keeps the right context in the right window.

Tool-Use Everywhere

Every domain agent runs with real tools — read_context, search_vector, run_read_only_sql, browser-use, OSS connectors. Tools execute against real Postgres, real Qdrant, real APIs. Hard SQL safety on read-only paths. No simulated outputs.

Single Data Spine

Cloud SQL is the source of truth across every domain. Workspace-scoped reads, audit-logged writes, durable agent runs. The same row that powers a CRM card powers the dispatcher's recall and the analytics chart.

Living roadmap

The Evolving Product Map

The map evolves every cycle. What ships moves left; what slips moves right. Counts trace to docs/launch/COMPETITIVE_ANALYSIS_LAUNCH_DOCUMENT.md and the master repair plan.

Shipped
  • 53 OSS engines wrapped behind unified registry (production band)
  • 30 verticals declared, 5 wedge workflows verified end-to-end
  • Personal finance: accounts, transactions, budgets, taxes, invoices (22/35)
  • Unified search: Meilisearch + Crawl4AI + Firecrawl + AI overview (15/24)
  • Marketing: campaigns, segments, journeys, automations (16/25)
  • CRM: Twenty backend, contacts, deals, dunning (8/13)
  • Agents: builder, browser-use, scheduler scaffolding (13/21)
  • Stripe + Lago metered billing, 4 tiers, trial, dunning, portal
  • Master Dispatcher: real DAG plan, budget caps, lesson persistence
  • Multi-provider AI router: Anthropic, Google, OpenAI, Groq with fallback
  • Tool-use foundation: read_context, search_vector, read-only SQL with DDL block
  • Cross-domain memory tier (AgentMemoryItem) and procedural memory miner
In Flight
  • Master Dispatcher → 24-domain adapters (finance/legal/crm/data live; 21 to wire)
  • Agent skills marketplace (28 ECC + 5 neww-original skills loaded)
  • Composio-bridge for ~700 third-party connectors
  • Persona committee + RAG-eval (ragas-svc) gates
  • License-guard and IaC-runner with BUSL/GPL refusal at runtime
  • OmniGraph Postgres-backed knowledge graph (validated end-to-end)
  • Trivy + Wazuh + TruffleHog wired into security engine
  • MLflow bridge for model lineage + drift dashboards
  • Native Gemini tool-use provider in ai-router fallback chain
Next 90 days
  • Tranche D: GNN, time-series foundation models, Reflexion, DSPy, drift
  • Tranche E: cross-domain handoffs, tool-use ratchet, verification ensemble
  • Tranche E: 5-tier memory architecture, 240 evals, policy DSL
  • Tranche F: workspaceId migration, Feast feature store, streaming, CI gate
  • Domain depth parity for Healthcare, Real-Estate, Travel, Automotive
  • Email delivery (Resend/SendGrid), Sentry, S3 persistence, email verification
  • Plaid live link, receipt OCR, Metabase-parity dashboard builder
  • Public marketplace surface for share-links + referral tiers
Horizon
  • Self-evolving Domain OS: dispatcher proposes new domain specs from usage
  • Multi-tenant agent skill exchange across orgs (sandboxed, audited)
  • Real-time multimodal canvas: voice, screen, document, code in one loop
  • Edge agent runtime: local inference, on-device tools, offline-first
  • Verifiable execution: every agent action signed, replayable, attestable
  • Marketplace economy: third-party skills, personas, evals, datasets
Competitive thesis

How We Defeat the Revenue Leaders

We do not win by adding features to incumbents' categories. We win by collapsing categories into outcomes — and by being the only place where the outcome of one domain feeds the agent in the next.

DomainRevenue leaderOur wedgeHow we win
Personal FinanceMint, Copilot.moneyAn agent that closes books, files quarterly taxes, and chases invoices — not a passive dashboard.Mint reads. neww.ai acts. Same data, but the agent reconciles, drafts, and books.
Search & ResearchPerplexity, GoogleCross-source reasoning over the user's own data + the open web, with citations and a follow-up agent.Perplexity answers. neww.ai answers, then files the report, books the meeting, and updates the CRM.
Marketing & GrowthHubSpot, Klaviyo, Customer.ioCampaigns generated, sent, measured, and re-optimized by an agent loop, not configured by a human in 12 tabs.HubSpot sells seats for marketers. neww.ai replaces the marketer's 80% repeatable work.
CRM & Revenue OpsSalesforce, HubSpot, AttioA CRM where the AI updates the records, not the rep — every email, call, and meeting flows in.Salesforce charges for fields. neww.ai charges for outcomes — closed deals, recovered dunning, expanded ARR.
Code & BuildCursor, v0, ReplitAgent that ships full apps end-to-end with real preview, real deploy, and the rest of the OS attached.Cursor edits files. neww.ai edits the file, runs the test, deploys, and updates the changelog in one loop.
Data & AnalyticsMetabase, WrenAI, HexConversational analytics over the user's full domain spine — not a separate warehouse to wire up.Metabase needs an analyst. neww.ai is the analyst — schema, query, chart, narrative.
CommerceShopifyStorefront + ops + finance + CRM in one workspace, agent-managed across the whole funnel.Shopify is a storefront. neww.ai is the storefront, the merchandiser, the support rep, and the CFO.
Agents PlatformManus, AutoGPT, OpenAI AssistantsAgents with real tool-use, real memory, real budget caps, real verification — running on real domain data.General agents wander. neww.ai's agents run inside a domain OS with the data and tools they need built in.

Honest position per domain available in /vs for live comparison.

Seven Hero Loops

An OS is not a feature list — it is a set of recurring outcomes a user can rely on. These are the seven loops every domain in neww.ai is being built to deliver.

H1

Goal → Plan → Run → Artifact

User states an outcome; dispatcher returns a verified deliverable.

H2

Inbox → Triage → Action

Email, CRM, finance, and support inboxes triaged and resolved by domain agents.

H3

Question → Evidence → Decision

Cross-domain research that ends in a written brief, not a chat transcript.

H4

Workflow → Skill → Reuse

Repeated multi-step work auto-mined into a callable skill.

H5

Spec → Build → Ship

From product idea to running app with deploy, tests, and changelog wired.

H6

Signal → Insight → Campaign

Behavior signals in, segmented campaigns out, with attribution closed.

H7

Risk → Audit → Remediation

Security, license, and policy violations caught and fixed before they ship.

Why an Incumbent Cannot Catch Us

Each of the four moats below compounds with usage. The longer a user is in neww.ai, the more expensive it becomes for any single-category incumbent to replicate the surface area they get for one bill.

Cross-domain memory

What you do in finance teaches the legal agent. What you sell in commerce informs the marketing campaign. Incumbents are siloed; we are not.

Tools, not chats

Every agent ships with executable tools that touch real systems. We measure success in artifacts produced, not tokens emitted.

OSS leverage

53 engines wrapping best-in-class open source means our marginal capability cost approaches zero while incumbents pay to build each capability twice.

One bill, one identity, one workspace

The buyer replaces 5–10 SaaS subscriptions, not adds an 11th. Pricing power follows because we are the consolidator, not the add-on.

How the Map Evolves

  1. 1.Tranche-driven cycles. Every cycle ships a tranche (T0–T7) of capability — vertical slices that move at least one domain from beta to launch-ready.
  2. 2.Evidence-based promotion. Items move from In Flight to Shipped only when verified by a smoke test, a captured screenshot, and zero console errors in the live app.
  3. 3.Revenue-leader benchmarking. Each domain is reverse-engineered against its top 2–4 revenue leaders. Gaps are tracked in COMPETITIVE_ANALYSIS_LAUNCH_DOCUMENT.md — and closed before we declare parity.
  4. 4.Honest hiding. If a surface is not real yet, we either ship it as a typed 503 with a clear next action, or we hide it from navigation. We never ship mock UI presented as production.
  5. 5.Memory feeds the next cycle. Lessons captured by the dispatcher and gaps surfaced by usage become the prioritized backlog for the following tranche.

Use the OS that's being built to win every domain.

Start with the wedge — finance, search, marketing, CRM, agents — and grow into every workflow your team runs. One identity. One agent. One bill.