NO FORCED UI.
AI IS THE UI.
Your team already uses ChatGPT, Claude, Slack. CMP42 does not compete with their interface — it feeds it. The admin surface is a data table and a migrations CLI. That is it.
CRM UIS ARE DYING.
Your team already opens ChatGPT before they open Salesforce. They ask Claude to summarise a call before they log it. A partner drafts the follow-up note with a paste of the meeting transcript and “make it sound like me.” 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.
Be honest about the role the CRM plays today: it is not the tool that runs the day, it is the database that gets consulted after the fact. If that is the honest role, let us build it for that role — a well-structured Context + Models + Processes graph and a fast, boring API, without a UI competing for your team’s attention.
LESS SURFACE. MORE STRUCTURE.
No drag-and-drop pipeline builder
Pipelines are part of your ontology, not a canvas. Stages are declared in your model files, versioned in git and validated on every write — so an agent cannot write a stage that is not in the pipeline.
No dashboards that go stale on Tuesday
Ask your agent for the pipeline and get an answer grounded in the graph, with citations. If you want a standing dashboard, build exactly the one you need against the API. No dashboards until you write them yourself.
A data table
The admin surface is a data table: a direct view of the objects in your graph, for the moments when a human needs to look at the records themselves.
A migrations CLI
Changes to your ontology happen in model files and are applied with a single command. The CLI is how the person who owns the data model works — reviewable, repeatable, in git.
The admin data table is a real working surface, not a viewer. It supports reading and editing — inline for single cells, in a detail view for the full record — filtering by any field with saved views, bulk actions, and CSV or JSONL export. Access control is per Model with optional field-level restrictions, so a role can be granted read on Contact but with the phone number field masked. No embedded analytics dashboard: usage numbers belong in your BI tool, not in a CRM sidebar.
YOUR TEAM ALREADY HAS AN INTERFACE.
It is called ChatGPT, Claude, Slack, or the app your engineers will build next month. CMP42 is designed to sit underneath all of them.
| Where your team works | How it reaches CMP42 |
|---|---|
| Claude Desktop, Claude Code, Cursor, Cline | Directly, through the MCP tools generated from your models. |
| A custom agent | Through MCP, using the MCP SDK — or through REST and GraphQL. |
| Agents built on ChatGPT or other models | Through the surface you wire them to: MCP where the client supports it, otherwise REST or GraphQL. |
| Your team’s Slack bot | A bot you run, reading and writing through REST or GraphQL. |
| The app you build next month | REST or GraphQL — the same operations, auth and audit trail as the MCP tools. |
Whatever the interface, every write is validated against your models and recorded against the human or agent that made it. All integration options →
WHAT HEADLESS DOES NOT MEAN.
Headless is a decision about where the interface lives — not a lack of visibility, control or care.
- Not “you cannot see your data.” The data table shows the records; your agent answers with citations to the source objects; every object is versioned.
- Not “no dashboards, ever.” You can build dashboards yourself against the REST and GraphQL API — the ones you actually need, not the ones a vendor shipped.
- Not “for developers only.” Most of your team works in the tools they already use. The CLI is for whoever owns the ontology; the API is for whoever builds on top.
- Not a black box. Every write is attributable, every process run is logged and replayable, and human approval steps sit wherever judgement matters.
- Not “start from zero.” CMP42 ships sensible defaults for contact, company, deal, task and activity, and generates MCP tools for them out of the box.
THREE ROLES. ONE GRAPH.
Ask and approve
Sales, partners and account managers ask their assistant for context and review what agents draft — at the human-in-the-loop steps you define in a Process.
Shape the model
Whoever owns the data model edits model files, reviews changes in git and runs migrations with the CLI. The ontology is a product artefact, not a hidden setting.
Build on top
Your engineers build the bot, the internal app or the dashboard against REST and GraphQL — or give an agent MCP tools and let it do the work.