SAME DATA.
GROUNDED AGENTS.
CMP42 is built for teams whose workflow already moved to AI — and whose CRM did not. Four kinds of teams are the first we are building with. What they share: agents that read a real graph before they write a word.
THE DATABASE STAYED PUT.
A year ago, a sales manager checked the pipeline by opening the CRM. Today the honest answer is often: "I ask my agent to summarise it, and if the summary looks off, I open the CRM." A partner in a consulting practice drafts the follow-up note in Claude, with a pasted transcript and a request to make it sound like them.
The workflow has changed. The database has not. Most teams we speak to have either wired their CRM into an agent through Zapier and a lot of Python — with inconsistent results, because the schema was never built for agent reading — or given up and paste transcripts into ChatGPT. CMP42 is built for the teams who want neither.
THE FIRST TWELVE.
The Design Partner Program takes twelve companies into Alpha. We expect them to come from four places. Each page explains the problem, how CMP42 approaches it, and an illustrative scenario.
Boutique consultancies
Your competitive advantage lives in the framework you keep re-explaining. CMP42 makes the framework the context an agent uses on the next engagement — not a slide you drag into every deck.
Framework as context →
Product-led B2B teams
You want an agent that reads a real customer's history before it drafts the renewal note. Not the generic ChatGPT summary. That agent needs a graph, not a HubSpot export.
Renewal notes from real history →
Agencies with AI ambitions
You are already selling AI to your clients. Sell them the version that grounds on their actual accounts — not the demo that hallucinates a competitor.
Grounded client agents →
Anyone leaving HubSpot
You want lean data plus a real API, not another platform to migrate onto next year. CMP42 is deliberately small, deliberately obvious, deliberately headless.
What a move looks like →
THREE PATTERNS, ONE GRAPH.
The four segments look different on the surface. Underneath, they use CMP42 in the same three ways.
Grounded agents
An agent that answers from typed objects and cites its sources. Connect Claude Desktop, Claude Code, Cursor, Cline or your own agent built on the MCP SDK: it reads the Context graph through MCP tools instead of guessing from a paste.
Ask "who at Aspen owns the Q4 renewal?" and the agent walks named edges — contact owns deal — and points to the objects it read.
Drafts from real history
Before an agent drafts the renewal note or the client follow-up, it reads the account: who owns the deal, which documents were authored, what the last notes said.
A Process can hold the draft for a human to approve, and every write records whether a human or an agent produced it.
Frameworks as context
The playbook your team keeps re-explaining becomes objects in the graph, linked to the engagements and accounts where it was applied.
Declared in Models and versioned like code, it gives the next agent your method as a starting point — not a blank prompt.
SMALL PARTS. DELIBERATELY.
None of the use cases needs a special edition. They run on the same three movements — Context, Models, Processes — and the same tool surface. What differs is the ontology each team declares and the processes it chooses to make observable.
That is intentional. A CRM small enough to hold in your head is one you can explain to your agents, your auditors and your next hire. Why we built it this way →
- Context — people, content and knowledge as one graph with typed edges, versioned and attributable. Context →
- Models — your ontology in git-versioned, TypeScript-shaped YAML, validated at write-time. Models →
- Processes — human-plus-agent workflows with approval hooks, run history and replay. Processes →
- MCP tools generated per model, with REST + GraphQL parity for everything else. MCP tools →
- Your hosting choice — managed by us in the EU, or self-hosted on your Postgres and object store. Self-hosting →
GROWTH FROM THE SAME DATA.
Most teams do not have a data problem. They have a reading problem. The contacts, deals, call notes and proposals already exist — scattered across a CRM, a drive and a wiki, readable by the humans who remember where things are, and by agents only as pasted text.
In CMP42, growth does not come from collecting more. It comes from the same data becoming usable by more readers. Once an account is a typed object with named edges to the people, documents and notes around it, every agent you connect can work with it — the one drafting follow-ups today, and the one your engineers build next quarter.
Why the effect compounds
Every agent reads the same graph
Connecting a new MCP client does not mean a new export, a new sync or a new prompt template. It reads the context that is already there — with the same auth and the same audit trail.
Every approved note becomes context
Notes an agent writes back are versioned and attributed. Once approved, they are part of the graph the next run reads — so the second renewal note starts from more than the first did.
Every model adds tools
When you declare a new model, CMP42 generates MCP tools for it. Extending your ontology extends what agents can do, without a separate integration project. Defaults stay lean; extension stays explicit.
We publish no growth figures. CMP42 is in Alpha. Any percentage on this page would be invented — and invented numbers are exactly what we are building against. Measured outcomes will come from Design Partner case studies, as described below.
RESULTS, PUBLISHED WHEN THEY ARE REAL.
The landing page talks about real results. Here is what that means today: there are none to publish yet, because Alpha begins in October 2026. We would rather show you an empty shelf than a fabricated testimonial.
The first Design Partner case studies are planned for Q4 2026, alongside the stabilisation of the MCP tool surface. They are optional for partners: if a partner agrees, we co-write the case study together; if not, we acknowledge them quietly and never name them without permission. We do not ask for testimonials, referrals or logos.
Where results come from
- Now
- Design Partner Program open. Twelve slots, applications close December 2026.
- Oct 2026
- Alpha begins with the first Design Partners. Weekly release cadence, real usage on real data.
- Q4 2026
- First case studies published. MCP tool surface stabilises.
- Q1 2027
- Beta opens to a broader waitlist. Self-hosted image published.
- H1 2027
- General availability. Pricing published.
Be one of the first case studies
If one of the four use cases describes the mess on your desk, come and Alpha it with us. The application starts with a 15-minute call.
→ Apply to be a Design Partner