acu configures Acumatica ERP from YAML files in a git repo (GitOps).
No UI clicks, no Configuration Wizard.
Tested against Acumatica ERP 26.101.0225 on Windows Server 2025, contract REST endpoint 25.200.001. Other versions will likely work, but only this combination is verified.
Acumatica configuration normally lives in the web UI: wizards, screens, and manual data entry that nobody can review, version, or reproduce.
acu moves that configuration into YAML files in a git repo, so a tenant can be rebuilt from scratch, audited in a pull request, and checked for drift like any other infrastructure.
YAML in git is the source of truth; the live tenant is the target. apply is the only writer.
| You can | Command | Effect |
|---|---|---|
| Seed a tenant from YAML | apply |
Idempotent PUT of config/ into the live tenant |
| Detect drift | diff |
Compare seed YAML to live (exit 2 when they diverge) |
| Pull live config into YAML | extract |
Inverse of apply — GET into config/{bootstrap,baseline,setup,master}/ |
| Run transaction scripts | run |
Forward documents from scenario/ (capital, buy, build, sell) |
| Capture derived balances | state |
Inquire trial-balance etc. into state/ (not seed) |
| Create and destroy tenants | tenant |
SSH control plane: list / create / delete / recycle |
| Publish the Bootstrap contract | bootstrap |
REST publish of AcuBootstrap (or --export zip for the UI) |
| Prove a cold rebuild | check |
create, then apply, then run, then diff on a fresh tenant |
| Snapshot a site offline | inventory |
SM203520 XML ZIP or ac.exe export xml writes inventory/ |
| Cross-check snapshot vs seed | reconcile |
inventory/ plus optional config/ writes findings/ only |
| Dump the contract schema | schema |
OpenAPI swagger.json for the pinned endpoint |
| Scaffold a data repo | config init |
Full config/, scenario/, matrix.yaml tree |
| Preflight a target | config check |
Discovery, secrets, matrix, REST, endpoints, SSH |
Seed YAML covers features, company, credit terms, subaccounts, chart of accounts, ledger, UOMs, financial year / calendar / periods, numbering sequences, inventory and distribution masters, roles, and users.
REST is the data plane (apply, diff, run, extract, state, bootstrap, schema).
SSH is the control plane (tenant *).
Hosted sites skip SSH.
uv tool install acumatica-cli
acu config init --host erp.example.com my-erp
cd my-erp # edit .env: set ACU_PASSWORD, ACU_TENANT
# pin+where = matrix.yaml cell (default_api + base_url)
# start from a brand-new empty tenant
acu config check # read-only preflight (incl. matrix.yaml)
acu tenant create --login DEV # create + bootstrap (SSH; --id optional)
# or hosted: acu --tenant DEV bootstrap
acu --tenant DEV apply config/ # seed config/{bootstrap,baseline,setup,master}/
acu --tenant DEV run scenario/ # once capital → buy → build → sell
acu --tenant DEV diff config/ # prove zero drift (exit 2 on drift)
acu --tenant DEV state # capture state/ trial-balance
acu check --yes --tenant DEV # cold lifecycle create→apply→run→diff (leave tenant)Bare apply / diff (no path args) also prefer config/ when those trees exist.
See docs/demo-seed.md for the entity map, once-guard, apply-order notes, NumberingSequence vs prefs *NumberingID, curated *Preferences field depth (V41), and Role/User + password seed rules.
Hosted Acumatica (no SSH): the tenant already exists; set a blank ACU_SSH= in .env.
The scaffold omits the key — without it, acu defaults to Administrator@<ACU_BASE_URL host> for SSH boxes.
acu config init --host customer.acumatica.com my-erp
cd my-erp # edit .env: ACU_TENANT, ACU_PASSWORD; add ACU_SSH=
acu config check # REST preflight; ssh probe is skipped
acu --tenant DEV bootstrap # publish AcuBootstrap via REST only
acu --tenant DEV apply config/
acu --tenant DEV diff config/
# offline UI fallback when REST publish is blocked:
acu bootstrap --export AcuBootstrap.zip # import + publish on SM204505acu [--cell ID] [--tenant NAME] [--url URL] [--ssh USER@HOST] [--api-version V]
[--username U] [--password P] [--version] [--completion [SHELL]]
│
├── tenant tenant CRUD (ac.exe over SSH — control plane)
│ ├── list CompanyID, sign-in name, internal CD, type
│ ├── create --login NAME [--id N] create + bootstrap; re-run to republish (SSH)
│ │ [--type SalesDemo|T100|U100] [--parent N] [--hidden] [--no-init]
│ │ omit --id → next free CompanyID (max list + 1)
│ ├── delete --id N | --login NAME [--yes]
│ │ delete the tenant and its data, recycle app pool
│ └── recycle [--yes] restart site app pool (tenant map + free API slots)
│
├── bootstrap [--export PATH] publish AcuBootstrap (REST); --export = offline zip
├── apply [--dry-run] [FILES...] push YAML via REST (idempotent PUT upserts)
├── diff [FILES...] drift check vs the live tenant (exit 2 on drift)
├── run [--dry-run] [FILES...] execute transaction scenario YAML (exit 1 on any miss)
├── check [--all] [--yes] [--tenant L] cold lifecycle create→apply→run→diff; leave tenant (V47)
├── state [--out DIR] [--diff] [--assert-unchanged] [--dry-run] [FILES...]
│ capture derived state into state/ (not seed)
├── extract [--out DIR] [--only NAME]... [--force] [--dry-run]
│ inverse of apply into config/{bootstrap,baseline,setup,master}/
├── inventory [--out DIR] [--force] [--dry-run] ARTIFACT
│ offline snapshot artifact → inventory/ (not seed)
├── reconcile [--inventory DIR] [--config DIR] [--out DIR] [--force] [--dry-run]
│ inventory/ + optional config/ → findings/ only
├── schema [--out DIR] dump the endpoint's OpenAPI schema (swagger.json)
│
└── config configuration ops
├── init [--host HOST] [DIR] scaffold full data repo (config/, scenario/, matrix.yaml)
├── show print the resolved config as a complete .env
└── check [--strict] preflight: discovery, secrets, matrix, REST, endpoints, SSH
When you omit FILES:
apply/diffpreferconfig/<name>/if any seed child exists underconfig/. Otherwise they walk rootbootstrap/,baseline/,setup/, thenmaster/when present. A path likeconfig/expands those nested seed dirs in the same order.rundefaults toscenario/.statedefaults toconfig/views/; writes go tostate/(--out).
Scenario YAML may use ${current_period} (host-local MMyyyy) on steps, expect params, and once.present params.
config/views and state keep Period pinned — see docs/demo-seed.md.
extract always writes under config/{bootstrap,baseline,setup,master}/ (catalog-driven; never root SEED_DIRS).
inventory and reconcile are offline (no REST, SSH, or password).
inventory turns an SM203520 Settings XML ZIP or ac.exe export xml folder into inventory/.
reconcile compares inventory/ to optional config/ and writes findings/ only.
It never writes seed.
Optional snapshot_map.yaml (data-repo root, or package defaults) maps DAC tables to catalog entities.
It normalizes the join: pad-trim, key/field aliases, Account/Sub FK CD resolve, enum label to code.
See docs/demo-seed.md.
acu --completion emits a completion script for bash, zsh, or fish — source it from your shell profile.
Run acu --help for the full mental model (workflow, planes, exit codes, command map).
That is enough for an agent to learn the tool without extra docs.
Run acu <command> --help (or -h) for flags, examples, and prerequisites.
Two read paths, one mutator (V35).
Do not confuse them with each other or with state:
| Command | Plane | Input | Writes | Role |
|---|---|---|---|---|
extract |
REST (live) | tenant via contract API | config/{bootstrap,baseline,setup,master}/ |
Inverse of apply — seed YAML |
inventory |
Offline | SM203520 Settings XML ZIP or ac.exe export xml folder |
inventory/ (summary.yaml + tables/) |
Full-table snapshot IR — not seed |
reconcile |
Offline | inventory/ + optional config/ |
findings/ only |
Cross-check gaps/deltas — never mutates seed or tenant |
state |
REST (live) | config/views/ |
state/ |
Derived balances/totals — not seed, not inventory |
apply |
REST (live) | seed YAML under config/ |
tenant | Sole tenant writer (keyed PUT) |
inventory/ and findings/ are engagement outputs: not SEED_DIRS, never loaded by apply/diff, not scaffolded by config init.
Binary .adb snapshots are rejected (XML only).
See docs/ac-exe.md for export / SM203520 notes and docs/demo-seed.md for the extract/state/inventory map.
Your configuration lives in its own git repo.
acu config init scaffolds a single full seed under config/ (features, company, credit terms, expanded COA, masters) plus lifecycle scenario/, observer config/views/, and README.
The Bootstrap endpoint contract is package SoT (bootstrap_project.xml inside the CLI — Bootstrap/1.4.0).
config init never writes project.xml, and data repos must not keep one (a present file hard-errors on bootstrap/publish).
There is no --flavor.
| Path | What it holds |
|---|---|
config/bootstrap/ |
virgin-tenant config: features, company, credit terms (no project.xml) |
config/baseline/ |
reference data: subaccounts, COA, ledger, UOMs |
config/setup/ |
one-time actions: financial year, master calendar, open periods |
config/master/ |
inventory/distribution masters: numbering (05-…) before prefs, warehouse, items, parties + Role/User (90-roles then 91-users) |
scenario/ |
lifecycle txns for acu run: once capital, then buy, build, sell |
config/views/ |
observer views for acu state (inquire: / entity: / gi:; not SEED_DIRS) |
state/ |
committed derived-state observations (evidence, not seed; money/qty fixed-point) |
inventory/ |
engagement: offline snapshot tables from acu inventory (not seed; not SEED_DIRS) |
findings/ |
engagement: acu reconcile cross-check output (never apply path) |
matrix.yaml |
multi-host pin+where: cells id+erp+default_api+base_url (V27); --cell selects |
.env |
secrets + optional where override (ACU_*); never Default API pin |
Legacy data repos may still keep root bootstrap/…master/; bare apply/diff prefer config/ when present and never merge both trees.
Files in each directory apply alphabetically; the numbered prefixes (10-, 20-, and so on) encode dependency order.
Commit matrix.yaml with the seeds so every clone knows verified ERP line, Default API half, and REST where per cell.
Seed YAML is state: apply upserts it, diff proves it.
acu extract is the inverse of apply: GET live tenant rows into seed YAML under config/{bootstrap,baseline,setup,master}/ (hard-cut).
Packaged seed_catalog.yaml is the sole extract registry (entity, endpoint, keys, file, strip/include, filter-split).
The demo entity map in docs/demo-seed.md mirrors those catalog paths.
Features synthesize to config/bootstrap/features.yaml.
Existing files skip unless --force; empty live sets skip.
Row failures continue (exit 1 only if any row failed).
Drift stays with diff.
acu --tenant DEV extract --out . --force # refresh config/** from live tenant
git diff config/ # review extract delta before commit
acu --tenant DEV apply config/ # replay extracted seed
acu --tenant DEV diff config/ # expect exit 0Dual-served entities (on both Bootstrap and Default) need an explicit endpoint: line.
| Value | Resolves to |
|---|---|
| omitted | Default/<api_version> for Default-only entities |
bootstrap |
active Bootstrap/<ver> from the packaged contract only |
default |
Default/<api_version> — tracks the resolved API version |
Bootstrap/1.4.0 or Default/25.200.001 |
literal pin (ignores the resolved Default version) |
api_version resolves as --api-version flag, else active matrix.yaml cell default_api, else code default 25.200.001 (never ACU_API_VERSION in .env).
base_url resolves as --url, else ACU_BASE_URL, else active cell base_url.
Prefer symbolic default over a pinned Default/25.200.001 so the seed tree travels with the dataset pin.
Requires Python 3.14 or newer.
uv tool install acumatica-clipipx install acumatica-cli and pip install acumatica-cli work too.
Or clone and install editable for development:
git clone https://github.com/kborovik/acumatica-cli.git
cd acumatica-cli
gmake install # editable install as a global uv toolVerify with acu --version.
Secrets live in one .env file (ACU_* vars).
Non-secret where and the Default contract pin live in committed matrix.yaml (cell base_url + default_api).
Optional ACU_BASE_URL overrides cell where for ad-hoc probes.
# ACU_BASE_URL optional when matrix.yaml cell carries base_url
ACU_TENANT=LAB5 # sign-in name of the tenant API sessions use
# ACU_SSH omitted → defaults to Administrator@ + resolved base_url host
# ACU_SSH= # hosted opt-out (blank key)
ACU_USER=admin # optional, defaults to admin
ACU_PASSWORD=... # required for live commandsThere is no ACU_API_VERSION env key (unknown ACU_* vars are ignored).
Ad-hoc override: acu --api-version 24.200.001 … (version half only, never
Default/25.200.001 — a full path would nest as /entity/Default/Default/...).
Committed matrix.yaml is the sole data-repo pin+where registry (1..N cells):
cells:
- id: "default"
erp: "26.101.0225" # claimed product line/build
default_api: "25.200.001" # sources Instance.api_version when --api-version absent
base_url: "http://acu-dev1.vm.internal/AcumaticaERP"--cell <id> selects a cell (omit means first cell).
When present, live commands source api_version from cell default_api and base_url from the cell when flag/env leave them unset.
acu config check reports ok matrix (cell=...; api_version from default_api=...; ...).
Missing matrix.yaml only warns on config check unless you pass --strict.
acu check (lifecycle) requires matrix.
One trunk seed in the data repo serves every host.
Version fan-out is not long-running product branches (acu-25r1, acu-26r1, …).
| Piece | Role |
|---|---|
| Trunk seed | Canonical config/ + scenario/ (newest supported matrix) |
matrix.yaml cells |
Each host: id+erp+default_api+base_url; --cell / acu check --all |
| Optional overlays | Surgical seed deltas keyed by Default half (e.g. overlays/default-24.200.001/) |
| OpenAPI | Live acu schema dump only (gitignored); never multi-version swagger trees in package or data repo |
Overlays live under overlays/default-<default_api>/ (scaffolded by acu config init).
No --overlay flag.
Bare compose (pin auto): when path args are omitted, acu apply / acu diff append overlay config seed dirs when present.
acu run replaces same-basename scenario files from the pin overlay.
Pin = resolved api_version (matrix cell default_api).
Explicit path args disable auto-compose.
# matrix cell default_api: 24.200.001 → uses overlays/default-24.200.001/
acu --cell lab25 apply
acu --cell lab25 run
acu --cell lab25 diff
# explicit path args (no auto) — later path wins same keys
acu apply config/ overlays/default-24.200.001/
acu diff config/ overlays/default-24.200.001/
# cold lifecycle every cell (SSH + tenant required); tenants left for inspect
acu check --all --yes --tenant LAB5Add a future half by creating overlays/default-<new-half>/ with the
minimal rewrite; no long-running product branches and no multi-version
OpenAPI trees (V11/V44).
Sibling data-repo retirement of release branches: acumatica-gitops#2.
Worth knowing:
- The
.envfile is found by walking up from the current directory, so any subdirectory of the data repo works. - Without a
.env, global flags plus the process environment (and matrix cell where) supply the configuration. - When
ACU_SSHis absent, acu defaults toAdministrator@+ the resolved base_url hostname. A present blankACU_SSH=is the hosted opt-out. Onlyacu tenant/acu checkrequire a non-empty value post-default. acu config showprints the resolved.env(password excluded; neverACU_API_VERSION) and comments active cell id/erp/default_api/base_url plus api_version source whenmatrix.yamlis present.- Redirect it to turn resolved state into a working config:
acu config show > .env.
Verify before touching anything live:
acu config check # discovery, secrets, matrix, REST, endpoints, SSH
acu config check --strict # missing matrix.yaml becomes fail
acu apply --dry-run # show what would be written, write nothingRequires GNU Make at least 3.82 — the Makefile uses .ONESHELL.
On macOS use Homebrew's gmake (brew install make); /usr/bin/make is 3.81 and fails the guard.
Elsewhere plain make is fine when it is GNU Make.
git clone https://github.com/kborovik/acumatica-cli.git
cd acumatica-cli
gmake install # editable install as a global uv tool
gmake check # offline gate: ruff, basedpyright strict, pytestThe default test suite is fully offline.
REST is faked with httpx.MockTransport, SSH with a monkeypatched subprocess.run — no live instance is needed.
gmake check must pass before every commit.
GitHub Actions runs the same gate on every push and pull request to main.
Human release notes live in root CHANGELOG.md (Keep a Changelog).
During development, append user-facing work under ## Unreleased in ### Added / ### Changed / ### Fixed as appropriate.
Empty Unreleased (no bullets) hard-fails the release — nothing to ship.
gmake release patch # or minor | majorgmake release is the sole release path (never local gh release create):
gmake check(ruff, basedpyright, offline pytest)- Fail if
## Unreleasedhas no bullets - Bump
pyproject.tomlversion (major|minor|patch) - Promote Unreleased body to
## [vX.Y.Z] - YYYY-MM-DD, leave an empty## Unreleased - Commit
CHANGELOG.md+pyproject.toml(+ lock if bumped) together, tagvX.Y.Z, push
GitHub Actions on tag v* re-runs CI, builds sdist+wheel, publishes to PyPI via OIDC trusted publishing, and creates a GitHub Release whose notes are the promoted CHANGELOG section for that tag (plus the artifacts).
gmake e2e runs the opt-in live tier against a real Acumatica instance (pytest marker e2e, deselected by the default suite).
Configuration is one file: a decrypted .env at the repo root names the instance — ACU_BASE_URL, ACU_TENANT, ACU_PASSWORD (and optional ACU_SSH; omitted defaults to Administrator@ + base-url host).
gmake e2e refuses to start without it.
The tier is self-contained.
Each run scaffolds a synthetic single-org company from the packaged acu config init templates into a temporary directory, copies the real .env into it, and runs the installed acu binary from there — no data repo, no pre-existing fixtures on the instance.
Scratch tenants (E2E, E2EA, E2EB, E2ESCEN) are created on the way in and always deleted on the way out, so nothing persists.
The packaged full config init seed (under config/) is the only scaffold.
gmake e2e # whole tier, about 20 minutes
gmake e2e FILE=test_provision_lifecycle # apply/diff focus
gmake e2e FILE=test_scenario_lifecycle # scenario + state focusThis project is licensed under the PolyForm Noncommercial License 1.0.0. Noncommercial use is free under that license.
Commercial use requires a separate license — contact lab5.ca.
Copyright 2026 Konstantin Borovik.