Named intelligence systems designed around specific business domains. Each runs on the same nine-layer architecture. Each taps the same substrate. Deploy the foundation and every system that follows is faster and cheaper because the infrastructure, the signal store, and the pattern library are already live. New domains are new framework clusters, not new architecture. The intelligence compounds across your business, not just within one domain.
19 agents. 10 reasoning frameworks. A live signal store and pattern library that encodes how your company specifically wins. Not generic sales methodology. Your patterns, from your deal history, confirmed by your people.
The output isn't a report or a dashboard. It's a substrate that any downstream workflow can reason against. Harness-agnostic, data source neutral.
Five readers run daily in parallel, each watching one analytical dimension: deal properties, engagement cadence, conversation signals, friction events, and use case positioning. They never see each other's output. One lens per agent.
The assembler aggregates into versioned deal state with full history. Pattern agents run weekly across the version history, finding what repeats across deals. The hypothesis engine promotes observations into testable patterns. Humans confirm. Confirmed patterns match against live deals automatically.
26 agents. 11 channels. A content operations system with a three-speed learning cycle driven by human editing patterns. It scores against your actual ICP, writes in voices learned from your edits, and earns autonomy through demonstrated quality. Runtime adjustments happen fast. Structural changes require human approval. The system never outpaces your ability to verify.
The pipeline runs from signal intake through ICP scoring, format shaping, evidence curation, ghost writing across four distinct voices, two-layer validation, and targeted distribution. The scorer never recommends formats. The shaper never assesses ICP fit. The ghost writer never touches the database. Perspective separation throughout.
Three learning speeds. Speed 1: runtime adjustments from observed editing patterns. 5+ observations required, decay automatically if not reinforced. Speed 2: proposed changes to the reasoning layer, surfaced for human review. 10+ observations over 4+ weeks. Speed 3: structural methodology changes. Human-gated, versioned, permanent.
The natural extension of the Metis signal store into the customer lifecycle. The same architecture, the same pattern library infrastructure, watching customer relationships instead of deals. A new framework cluster that encodes what expansion and churn look like in your data: the language shifts in QBRs, the stakeholder changes, the engagement drops. Every company running Deal Intelligence has the same next question: we understand how we win. Do you do the same thing for how we keep?
One substrate. Every system taps it.
The signal store and pattern library are the central intelligence. Praeco taps it to understand what messaging resonates with which buyer profiles. Customer intelligence taps it to recognise the signals that predict churn or expansion. Delivery intelligence taps it to align onboarding with how the deal was sold. None of them build their own data layer. They all reason against the same substrate.
Everything circles back. Confirmed patterns refine the frameworks. The frameworks sharpen the analysis. The analysis produces better patterns. Soft adjustments decay automatically if they're not reinforced. Permanent changes require human approval. The system forgets what's no longer true.
The workflows are scalable because the core is aligned. Once the foundation is calibrated, downstream workflows become straightforward. The routing, orchestration, lifecycle, and metadata layers are already built. The signal store is already populated. A new domain is a new framework cluster, not new architecture.
Anthropic Managed Agents, Cursor, LangChain, Claude Code, or custom. Any CRM. Any transcript provider. The five readers are the only integration point. Everything downstream operates on generic signal rows.
You describe the domain. We map out what framework clusters it needs, how the architecture connects to your data, and what the deployment looks like. Thirty minutes. No commitment.
Get in touch → or read the methodology