CMP42 // CONTEXT · MODELS · PROCESSES // EDITION 2026.09ALPHA · DESIGN PARTNERS OPEN · BERLIN 52.52°N

CRM UIs are dying.
AI is the UI now.

Your team already opens ChatGPT before they open Salesforce. They ask Claude to summarise a call before they log it. They pipe HubSpot data through a Zapier hop into whichever agent shipped this week. The workflow has changed. The database has not.

Legacy CRMs were built to be looked at. Every field, every pipeline, every dashboard exists because a human was going to click on it. That assumption is now legacy. AI does not click. AI reads structured context and writes structured actions. When you feed it a schema built for humans, it hallucinates. When you feed it a schema built for agents, it works.

CMP42 is the schema built for agents. A headless CRM: no forced UI, no drag-and-drop pipeline builder, no "dashboards" that go stale on Tuesday. Just a well-structured Context + Models + Processes graph and a fast, boring API that any agent — ChatGPT, Claude, your team's Slack bot, the app you build next month — can read and write against.

// The metaphor

FROM SILOS
TO POSSIBILITIES.

CONTEXT.
MODELS. PROCESSES.

01
Context · The graph

People, Content and Knowledge as one connected object graph.

Contacts, companies, deals — yes. But also documents, meeting recordings, product briefs, playbooks, the messy Notion pages. Every object is versioned, every edge is typed, every field is addressable by URI. Agents read the graph like a database, not like a chat history.

  • Typed edges — a contact knows a company, owns a deal, authored a doc
  • Content-addressable IDs — stable across schema evolution
  • Full versioning + audit trail — every write is attributable
  • MCP tool surface out of the box: get_context, list_related, write_note
02
Models · The schema you can shape

Models are your ontology. Version them like code.

CMP42 ships with sensible defaults for the CRM primitives — contact, company, deal, task, activity. Then you extend. Models are declared in TypeScript-shaped YAML, migrated with a single command, validated at write-time. Your ontology is a first-class product artefact, not a hidden Salesforce setting.

  • Declarative model files (git-versioned)
  • Automatic MCP tool generation per model type
  • Validation with clear errors — agents self-correct
  • Backwards-compatible migrations — no schema surprise for the agents in production
03
Processes · The observable workflow

Processes are agents-plus-humans workflows, made observable.

A process is a named sequence of steps. Some run on humans (a partner reviews a case note). Some run on agents (Claude drafts the follow-up). Each step reads and writes to the graph, each transition is logged, each run is replayable. You audit an agent's work the way you audit a git branch.

  • Human-in-the-loop steps with approval hooks
  • Full run history — replay any process against a past snapshot
  • Attribution — every write records the human or agent that produced it
  • Cost accounting — token spend visible per process, per run

SHAPE THE CONTEXT LAYER.
NAME AN MCP TOOL.

Twelve companies will spend six months with us in Alpha. You get direct-Slack access to the founder, a monthly co-design call, priority feature requests, and the right to propose and name one MCP tool that ships to every future customer. In exchange we ask for real usage and honest feedback.

→ APPLY TO BE A DESIGN PARTNER

WHY I AM BUILDING THIS.

I have spent fifteen years wiring companies to HubSpot, Salesforce, Pipedrive, and the two dozen CRMs before those. The pattern is always the same: the data model is built for a human clicker, the users end up in a spreadsheet anyway, and the AI features arrive three years late and none of them work.

Then AI as UI happened. My clients started asking their agent for the pipeline instead of opening the dashboard. And the agent got it wrong — because the data model was still built for the human clicker.

CMP42 is what I want to hand a team on day one of their next re-platforming. Small enough to hold in your head. Big enough to feed the agents you have not chosen yet. Berlin-built, EU-hosted, GDPR-native, self-hostable. No dashboards until you write them yourself.