An AI sales-operations command center

Every message earns its send.

VANGUARD drafts outreach grounded in firm-specific research — then makes every AI-written word pass three automated moderation layers and a human gate before it ships. Nothing sends itself.

29,996 CONNECTIONS · 11,091 COMPANIES · ONE OPERATOR

“AI made outreach cheap. Cheap made it worthless. The scarce asset now is trust — and trust cannot be automated. It can only be protected.”

Autonomous senders flooded every inbox. Reply rates collapsed, and with them the reputations of the people who let machines speak unsupervised in their name.

The answer is not less AI. It is more discipline. Machines are extraordinary at research, drafting, and pattern-finding. They should do all of it — right up to the moment a message leaves.

That last step belongs to a person. VANGUARD is built around this line: the system proposes, the human disposes. It is slower by seconds and better by orders of magnitude.

The edge

Every rival sits in one box. This one holds the intersection.

The category is real and crowded — network engines, enrichment platforms, autonomous senders. Each does one thing well and falls into one trap. VANGUARD takes the opposite bet on the six axes that decide who a serious operator trusts.

AxisThe fieldVANGUARD

OwnershipRented SaaS. Your graph lives in their cloud, billed back per seat.You own it — self-hosted on your own database. The graph is yours, and portable.

CostCredits and seats that scale with every enrichment.Near-zero — free and open sources, roughly a dollar of embeddings for the life of the corpus.

TrustTrending autonomous — volume that risks the sender's reputation and the channel.Draft-only. Three automated layers, then a human approves every send. Zero ban risk.

ScopeA point solution — network, or enrichment, or signals, or outbound.The whole loop — intelligence, network, outreach, content, and revenue in one spine.

InteropA closed app you log into, one more tab.MCP-native — drive it from Claude, Claude Code, or ChatGPT; its agents borrow external tools in kind.

FitBuilt for a generic sales team.Tuned to one operator's motion — consulting, service-as-software, partnerships, OSS-led.

A private, agent-native GTM operating system — the relationship, intelligence, and outreach graph you own, not rent — drivable from the AI tools you already use.

Competitor radar · tracked daily by Scoutnetwork engines · enrichment platforms · autonomous senders · signal tools · community CRMs

The doctrine

Six stages. Two gates. One signature.

Intelligence

Signals, filings, and network warmth fuse into 524 firm dossiers. The system knows why it is writing before it writes.

Draft

A model writes from the dossier — every claim mapped to a numbered source. Unsourced sentences never survive.

03GATE

Moderate

A PII redaction pass, then three layers judge every word — claims, factuality, brand. Most drafts stop here, get corrected, and try again.

04GATE

Approve

A human reads what remains. Approve, edit, reject — one keystroke each. The approval is signed.

Send

Only now does the message leave. Tracked opens, clicks, and replies write back to the record.

Learn

Outcomes retrain routing, warmth, and phrasing. The next draft starts smarter than the last.

01 / 06The dossier opens. Halcyon Grid Partners — tier A, Series C two hours old.

The modules

One platform runs the whole lifecycle.

PIPELINE

Pipeline

Every target firm as a living dossier — tier, warmth, signal score, last touch, next step. A table you can trust because agents refresh it nightly and flag what they cannot verify.

524 live dossiers

app.vanguard — /pipeline

STUDIO

Studio

Drafting with the dossier open on one side and the moderation pre-check on the other. Claims map to numbered citations; absolutes get caught while you type.

every claim carries its source

app.vanguard — /studio

QUEUE

Queue

The approval room. One draft at a time, three verdicts then you, a keystroke to decide. Calm, spacious, and impossible to bypass.

A · E · R · S — the human gate

app.vanguard — /queue

AGENTS

A fleet, not a swarm

Named specialists — Scout, Strategist, Creator, Concierge, and the divisions behind them — each earning autonomy on an L0–L4 ladder through evaluated performance. A moderation flag demotes automatically; cold first-touch is capped at L1 forever; every one sits a keystroke from a kill switch.

cold first-touch: L1, forever · outbound never autonomous

app.vanguard — /agents

INTEL & NETWORK

Intel & Network

The relationship graph behind the outreach — 29,996 connections mapped, warm paths surfaced, signals fused to dossiers within hours of the news.

2-hop warm paths, computed live

app.vanguard — /network

NARAD

Narad

Ask the platform anything about itself. Answers cite their spans — hover a claim and see exactly which record produced it.

span-cited RAG over the system of record

app.vanguard — /narad

MODEL STUDIO

Model Studio

Eight providers behind one gateway with free-first routing, live health, and per-run cost written to the ledger. The cost brake engages itself.

8 providers · one gate

app.vanguard — /models

Architecture

A cutaway drawing of the machine.

The page you are reading just crossed into the ops room. This is what stands between a language model and your reputation.

STAGE 1

8-provider LLM gateway

free-first routing · per-run cost tracking

AnthropicOpenAIGoogleMistralAmazon BedrockOpenRouterNVIDIA NIMLocal GPU

STAGE 2 — THREE AUTOMATED LAYERS

Content moderation — a PII redaction pass, then three layers

Claims

Absolutes and unverifiable capability claims are flagged.

Factuality

Every stated fact must trace to a cited source.

Brand

Voice, register, banned vocabulary. Hype words never ship.

STAGE 3 — THE HUMAN GATE

Approval queue

draft-only outbound · no exceptions

No message reaches a recipient without clearing all three automated layers and a named human's explicit approval. The approval itself is signed.

STAGE 4

Audit log with replay

signed action tokens · kill switches on every agent

391 automated tests61 data models30 migrations23 MCP toolsevery action signed + auditable14 named agents · outbound authority: zero

Agents earn autonomy on a trust ladder — and lose it automatically.

connections mapped

companies classified

priority targets

live dossiers

AI providers, one gate

Origin

This platform was built end-to-end by a single operator directing fleets of AI coding agents — the same discipline it enforces on outreach, applied to its own construction. It is the proof of what the approach produces: enterprise AI agents and GenAI platforms that do serious work under real governance.

The work is deliberate.
The engine is ready.