Layer 1 · Foundation
Before a single agent runs, the truth has to live somewhere immutable: who the company is, who it sells to, what it says and in whose voice, and how it sells. We standardized that into a brand bible with two spines and a commercial layer — and a modular context system so every agent loads exactly the truth it needs.
The brand bible
The visual spine (how we look) and the content spine (what we say, in whose voice) sit beside a commercial layer most companies never write down. Every skill, agent, and asset resolves brand truth through these files — never by hardcoding.
Tokens, design system, logo usage, imagery, components, mascot. A compliance gate keeps every asset on-brand regardless of which tool generated it.
Positioning, offering, methodology, proof points, ICP segments, content pillars — plus a hybrid voice registry: company voice, founder voice, and per-rep voices.
GTM strategy, the channel engine, the sales funnel, account scoring, packages, and the org/fleet map. The half clients don’t usually carry.
Standardized & cloneable: the bible has a fixed shape (L1→L3 depth). BioCreative is the gold reference; every client brand is cloned from it and audited against it.
Modular context (Karpathy, applied)
An operating system doesn’t load every file into memory at once. Neither do we. Each project, agent, and harness gets a modular context system — scoped files, scoped rules, scoped memory — so the model reasons over what matters and nothing else. It works at large scale (a whole department) and small (a single skill).
A skill, an agent, a channel, a client — each carries its own context spine. Loading less, but loading the right less, is what keeps reasoning sharp and cost low.
Always-on rules and conditional rules enforce how the company behaves — naming, scoring, email formatting, cost discipline — so behavior is consistent across every agent.
The data engine
The market truth doesn’t arrive in one feed. We engineered ingestion from many sources — company registries, clinical-trial and grant databases, patents, conference and event data, news, web research, and enrichment providers — each with its own code path, all resolving into a single account graph. A hub-and-spoke architecture keeps one master intelligence base; every client gets a filtered, enriched spoke of it, with data flowing one way so the master stays clean.
Trials, grants, patents, events, news, web research, and enrichment all normalize to the same companies and people — so a signal anywhere becomes context everywhere.
A master intelligence base feeds each client a customized spoke. One-way sync means a client’s view is always a clean, enriched slice of the whole — never a fork that drifts.
Classification & enrichment
Raw records aren’t intelligence. We run multi-level classification and enrichment so the system knows not just who exists, but what they are and why they matter.
How an engagement starts
Before a kickoff call, we run a market-intelligence pass on a prospect and deliver a personalized proposal — the page you’re reading is an example, built entirely from our own research with zero input from the company. That research becomes the seed of the brand bible: the first draft of who they are, who they sell to, and how. Through Phase 1 we ingest their real material — their voice, assets, pipeline, and priorities — and align everything to that spine, building toward the full commercial operating system.