| name | xentral-business-model |
| description | Read and maintain an instance's Business Model — the canonical description of how the business actually runs (model, sales channels, fulfillment, automation, target market, pain points) — with the `xentral_business_model` tool. Covers the 6-dimension strict-enum profile that every other agent reads first, the markdown sections, starting from a library template, and the read/write/ apply_template/profile_set workflow. Use when onboarding an instance, when the user describes a strategic change ("we opened a B2C shop", "we moved to a 3PL"), or when another agent needs business context before acting — not to log day-to-day operations (Jobs) or track metrics (Dashboards).
|
| examples | ["Set up the business model for a new B2B wholesale instance.","We just launched a French webshop — update the business model.","Classify this instance's profile so the other agents behave correctly."] |
Xentral Business Model
Purpose
The Business Model is the canonical description of how this instance actually
runs: business model, sales channels, fulfillment posture, automation level,
target market, pain points. It is the source of truth every other AI feature
reads first to behave appropriately for this operator. The
xentral_business_model tool reads and maintains it.
What it is not:
- Not a financial plan / P&L. It describes the business, not its forecast —
no per-quarter numbers.
- Not a marketing brochure. Concrete facts an agent can act on, no fluff.
- Not an operations log. Day-to-day decisions go to Jobs / System Events.
The profile is the contract
The single most important part is the profile — a 6-dimension strict-enum
classification that every other agent keys off to pick behavior. Get it right
first; everything downstream depends on it.
| Field | Type | Drives |
|---|
business_model | single | b2c, b2b, d2c, marketplace, manufacturer, wholesale, mixed — invoice phrasing, payment-term defaults |
sales_channels | multi | ecommerce, marketplaces, pos, field_sales, wholesale, distributors — expected integrations |
fulfillment | single | in_house, 3pl, dropshipping, mixed — shipping-doc layout, returns flow |
automation | single | low, mid, high — how aggressively agents auto-act vs. ask |
size | single | small, mid, upper, enterprise — feature/plan suggestions |
complexity | single | simple, medium, complex — how detailed agent explanations are |
Always set the profile via profile_set, never via write. The write
path doesn't enforce the enums and lets a typo through silently, breaking every
downstream agent. Read the live enum list (profile_get / the tool's enum
schema) before setting — invalid values are rejected.
Start from a library template
For a fresh instance, seed the sections from a starter template instead of writing
each from scratch. The library ships business-model templates (b2b_wholesale,
d2c, ecommerce_boutique, manufacturer, multi_channel, generic, …), each paired
with a _configuration. Apply one with apply_template, then refine. The three
skeleton flavours are classic, agents, ai_first.
The workflow (every task)
action: help — the live manual (section semantics, profile enum values).
Read once per session; pass locale='de' for German.action: profile_get — read the 6-dimension profile (and the valid enums).action: read — without slug: list all sections with metadata; with
slug: one section's markdown. Always read before a replace write.- Seed / classify:
apply_template (classic|agents|ai_first,
overwrite=false by default — non-destructive), then profile_set for the
six dimensions.
- Edit:
write a section — mode='replace' (default) or mode='append'
(incremental adds). Never write the profile section.
Which reference to read
- Profile enums, the section catalogue, recipes & pitfalls — the full
section list (mission, target_market, channels, operations, pain_points,
tone, …), invariants, and the onboarding / strategic-update / quick-add flows →
reference/profile-and-sections.md
Don'ts
- Don't write the
profile section via write — use profile_set (enum-safe).
- Don't
replace a section without reading it first in the same turn — you'll
clobber the user's work.
- Don't
apply_template(overwrite=true) without explicit user consent — the
default false protects edited sections.
- Don't fill every section eagerly at onboarding — empty sections legitimately
mean "not relevant yet".
- Don't paste chat or marketing copy into a section — distill actual operations;
every section is read by other agents like a public API.
