Muster

Frappe AI · ERPNext AI · governed automation

Frappe AI that understands DocTypes, fields, workflows, and permissions.

Muster gives Frappe and ERPNext teams an agent harness that can build context from sites, installed apps, modules, DocTypes, reports, scripts, and workflows before it acts.

The goal is practical: fewer generic ERP guesses, more permission-aware answers, scoped memory, token visibility, and auditable automation.

pnpm --package=@musterhq/cli dlx muster demo

Frappe context pack

A plugin-owned context layer, not a bloated core binary.

Muster keeps the main harness light. The Frappe / ERPNext pack is where site-specific context belongs: docs, installed apps, module metadata, DocTypes, fields, permissions, reports, scripts, and workflows.

Why Frappe AI needs a harness

ERP context is relational, permissioned, and constantly customized.

DocType-aware retrieval

Frappe answers need module, DocType, child table, field, custom field, and workflow context, not generic docs alone.

Permission-aware actions

Reads and writes should execute as the real paired Frappe user, with Frappe remaining the authority.

Token visibility

Long support and ERP investigation threads should show token cost and replay waste instead of hiding spend.

Scoped memory

Tenant, site, workspace, user, and session context must stay separated to avoid cross-customer leakage.

Eval-gated learning

New Frappe behavior should be backed by fixtures before it becomes durable assistant behavior.

Use cases

What Muster can help Frappe and ERPNext teams build.

support

ERP support assistant

Answer operational questions with site, module, workflow, and record context while keeping memory scoped.

developer

Frappe developer copilot

Inspect apps, DocTypes, hooks, client scripts, reports, and bench commands with an audit trail.

operator

Production runbooks

Guide deployments, migrations, scheduled jobs, and incident triage with receipts and token accounting.

automation

Permissioned actions

Create or update records only through explicit tools, setup checks, and user-bound permission context.

Build Frappe AI with controls first.

Muster is open source. The Frappe / ERPNext path is strongest when contributors add real fixtures, module packs, field-level retrieval tests, and production-safe adapter hardening.