Authored by Samuel Glauser as CRM/Ideas on memex.systemorph.com (2026-08-28), reproduced verbatim. Idea 1 and idea 4 are what the first release ships (Crm/Client as a partition root with real children; Crm/Board as a live layout area); 2, 3 and 5 are open.

Five ideas where building the CRM on MeshWeaver gives us something a vendor product structurally cannot. Each names the mechanism it relies on, so we can judge feasibility rather than vibes. These are proposals, not decisions.

1. The customer graph is the CRM — accounts as mesh nodes with real edges

Model account, contact, opportunity as node types in the mesh, and let the relationships be actual graph edges to the things we already have here: the delivery space, the hosted instance, the course a client's staff attended, the support thread. In a vendor CRM those live in other systems and get linked by a copied-in id, at best.

Why it pays: "show me everything about Pearl Technology" becomes one traversal — pipeline, deployments, threads, training, all of it — instead of five tabs and a stale integration. Account health stops being a manually maintained field and becomes derived.

Mechanism: node types + the graph; existing spaces (e.g. the Portfolio, the Deployments) as edge targets.

2. Email and calendar as the activity feed — the log fills itself

Requirement 3 of the Design Brief is where every CRM dies. We already have mailbox and calendar tooling running as the user. Point it at the customer graph: inbound and outbound mail matched to account by sender domain and to opportunity by thread, meetings attached from the invitee list, all as activity nodes.

Why it pays: the completeness of the history stops depending on discipline. Nobody types a call report; the record exists because the interaction did.

Mechanism: the mail/calendar tooling + address-domain matching. OPEN: privacy boundary — whose mail is in scope, and how do we keep genuinely private mail out. Needs an explicit rule before any implementation.

3. Agentic pipeline hygiene — an agent that notices what nobody entered

A scheduled agent walks the pipeline and reports on absence: opportunities whose close date has quietly passed, proposals with no follow-up, accounts gone silent, stage/amount combinations that make no sense. It drafts the nudge — a follow-up mail, a task, a flag — and routes it via the notification rules the owner set for themselves.

Why it pays: vendor CRMs sell this as "AI insights" over data they only partly have. Our agent sits inside the data with the same access as the user and can act, not just chart.

Mechanism: agents + scheduled runs + notification triage and per-user notification rules.

4. Live pipeline views instead of a report builder

Build the pipeline, forecast and win/loss views as reactive layout areas over the opportunity data — embedded on the account page, on a team dashboard, in a QBR document, on a wall screen. Same underlying data, no export, no refresh button, always current.

Why it pays: the forecast in the board deck and the forecast in the CRM cannot diverge, because they are the same area rendered twice. Add the versioned snapshot the brief asks for and "why did the forecast move" becomes a diff.

Mechanism: layout areas + region embedding + node versioning for snapshots.

5. Vendor independence as a designed property, not a slogan

Make the data model the deliverable: documented node types, a documented export, no logic that exists only inside a UI form. Sync outward to whatever a client or a finance process requires, but keep the mesh as the system of record.

Why it pays: it is the whole justification for building rather than buying — and it is also the honest test of it. If we cannot export the CRM cleanly, we have rebuilt the trap with our own hands.

Mechanism: node types as published schema + content collections for export.

Add yours

One idea per node under this one, or comment here. The useful contribution right now is disagreement with idea 2's privacy boundary and with the build-vs-buy premise generally — if a vendor CRM plus a thin mesh integration wins, better to find that out on this page than after a quarter of building.