For open-source bill of materials (BOM) management across nested robot assemblies and build stock, evaluate InvenTree first. Its documentation covers multi-level BOMs, variants, separate part revisions and stock allocation. Evaluate Part-DB when electronics inventory, project BOM imports and component withdrawal are the immediate job. Part-DB also links a built project to a part. Test whether that relationship works at the scale of a complete robot and survives release and export. BOM workflow evidence.
To verify an earlier build package, test whether the export preserves the selected revisions, drawings, conditional substitutions and firmware references alongside the physical quantities.
Which editions, versions and licences are compared?
This comparison uses documentation and source for InvenTree 1.7.3 and Part-DB 2.20.7. Their official GitHub release records listed both as the latest stable releases on 20 November 2026. Both were published on 17 November. The licence review covers the self-hosted server code at those releases.
| Candidate | Release inspected | Core licence grant | Documented self-hosting route |
|---|---|---|---|
| InvenTree | 1.7.3 | MIT. Use, modify and redistribute the code, including commercially, with the required copyright and licence notice. | Official Docker images and Compose, or a supported non-container installation. |
| Part-DB | 2.20.7 | GNU Affero General Public License, version 3 or later (AGPL-3.0-or-later). Commercial use is permitted under its conditions. | Docker or a PHP/web-server installation with a supported database. |
Part-DB's AGPL matters if you modify the program and let users interact with that modified version over a network. Section 13 requires an offer of its corresponding source to those users. Check the obligations for the software you deploy or distribute, including dependencies and modifications. Part-DB licence evidence.
“Open source” describes the licence. “Self-hosted” describes who operates the deployment. A free hosted tier, source-available code with restricted rights, or an optional commercial module would require a different comparison. The core code reviewed here has identifiable open-source grants. Running it still requires operational ownership.
How this comparison was made
The audience is robotics teams moving mechanical, electronics and firmware work from prototypes into low-volume production. InvenTree and Part-DB were selected for their documented BOM and inventory workflows.
The comparison covers nested structure, selected part identity, variant isolation, stock visibility, conditional substitutions, revision recovery and manufacturer export. We inspected official documentation, tagged repositories, licence files and release records on 20 November 2026. The comparison describes documented behaviours and source-visible fields and export paths. The buyer exercises are proposed trials.
The existing buyer-guide methodology provides a reference for comparable evidence and unknowns. This article uses qualitative, observable criteria without numerical scores or those aerospace weights. Missing evidence stays an unresolved question. Arc is a software vendor with a commercial interest in software evaluations.
Can either tool represent two gripper variants correctly?
Use this fictional robot before choosing a data model. WP-200 has two orderable items. WP-200-J has a jaw gripper and WP-200-V has a vacuum gripper. Both have one CH-110 A chassis and one CTL-310 B controller bought complete. Each has one gripper assembled from its selected inputs. Record firmware as a controlled installation reference.
REL-WP-01 was approved on 30 October 2026. Five jaw robots, WP-001–WP-005, were built on 6 November using JAW A. VAC A was qualified in the same original package. Actuator availability then prompted ECO-027, approved on 18 November. REL-WP-02, issued on 19 November, applies JAW B to twelve robots in batch B-2412, WP-006–WP-017, while carrying VAC A for eight robots, WP-018–WP-025. Material issue is planned for 14 December. Earlier serials retain their as-built release, with no retrofit.
Table 1 gives the expected selections. Quantities are one of each listed physical item per applicable robot. The pump is nested under the vacuum gripper.
| Configuration and BOM | Gripper and its physical children | Firmware on common CTL-310 B | Build applicability |
|---|---|---|---|
| CFG-JAW A / BOM-JAW A | GR-J210 A contains ACT-410 A and ENC-510 A | FW-310 1.4.0 | WP-001–WP-005, REL-WP-01 |
| CFG-JAW B / BOM-JAW B | GR-J210 B contains ACT-420 A and ENC-520 A | FW-310 1.5.0 | Twelve JAW robots, WP-006–WP-017, B-2412 under REL-WP-02 |
| CFG-VAC A / BOM-VAC A | GR-V220 A contains PMP-430 A. No jaw encoder | FW-310 1.4.0 | Eight VAC robots, WP-018–WP-025, B-2412 under REL-WP-02 |
DW-GR-J210 A/B are controlled jaw drawings. TV-027 is approved fictional verification evidence for fit, encoder counts and direction, and encoder-fault response on the exact JAW B combination.
For the planned batch, expected physical totals are twenty controllers and chassis, twelve ACT-420 A actuators and ENC-520 A encoders, and eight PMP-430 A pumps. There are no prebuilt grippers. Assemble twelve GR-J210 B and eight GR-V220 A from those inputs. The earlier ACT-410/ENC-510 stack is excluded from this release's requirement. The broad BOM guide explains the parent-child roll-up and manufacturer manifest.
How InvenTree handles variants and revisions — InvenTree documents template parts with variant BOM inheritance. It also treats a new revision as a separate Part object with its own BOM and stock, linked back to the original. Its documentation states that existing stock and build-order references to the original part are retained when a new revision is created. Templates themselves cannot have revisions, although their variants can. InvenTree revision evidence.
That gives a documented route to distinguish the JAW and VAC variants and retain A/B gripper definitions. Before importing, write down which application record corresponds to WP-200-J, WP-200-V, GR-J210 A and GR-J210 B. Keep those mappings with the release manifest. A default or inherited row should not silently change an earlier build.
The BOM documentation says edits to inherited rows flow into variants. For a released structure, test whether copied references, permissions and locking preserve the accepted values. Part locking must be enabled to enforce its documented restrictions. A validated BOM flag and checksum indicate specified BOM data have been validated and have not changed since. They do not, by themselves, prove approval of FW-310 compatibility or applicability of TV-027. InvenTree BOM evidence.
How Part-DB links a project to its built part — Part-DB documents project BOMs that can report buildable quantities and withdraw components. Its 2.20.7 source represents a built project as a Part and permits that Part to be linked to a project BOM entry. Part-DB assembly evidence.
Test a robot project using the built gripper part, then verify how its child inputs appear in the requirement and export. With no grippers on hand, record whether the workflow expands the gripper requirement or requires a separate build step. Also test whether an accepted GR-J210 A definition remains recoverable when B is introduced. The inspected evidence does not establish a complete robot release snapshot, so plan a separate controlled manifest unless the evaluation proves one.
Compare the documented workflows against the same robot
Table 2 keeps documented features beside the exact uncertainty that could change the buying decision. CSV means comma-separated values. IPN is an internal part number.
| Buyer criterion | InvenTree 1.7.3 evidence | Part-DB 2.20.7 evidence | Shared WP-200 acceptance question |
|---|---|---|---|
| Nested assemblies | Native multi-level BOMs, nested view and BOM Exporter. | Project BOM and project-built part linkage. Recursive requirement/export behaviour needs demonstration. | Do twelve JAW B and eight VAC A robots retain the gripper paths and correct child totals? |
| Part and stock tracking | Separate Part/StockItem records, locations and serial/batch tracking. | Parts, multiple storage locations, inventory event log and project withdrawal. | Can a builder distinguish the selected revision from a similarly named physical stock item? |
| Substitutions | BOM substitutes can be allocated, and their stock contributes to availability. | Reviewed project entry model identifies a linked part. No equivalent conditional approved-stack workflow was established. | Is ACT-410 rejected or held for JAW B, while ACT-420/ENC-520/FW 1.5.0 remains the selected stack? |
| Earlier revision | Separate linked revision parts preserve original record references. | Event log and reverting parts to older versions are documented. A complete earlier-package snapshot is unresolved. | Can another engineer retrieve WP-005's exact drawing, structure, firmware and approval after B exists? |
| CAD/electronics import | Assembly BOM import supports CSV/Excel, with explicit mapping still required. | KiCad PCB, schematic and generic CSV project BOM imports, with field mapping and part assignment to review. | Does each imported row resolve to the intended internal item and revision, with no duplicate controller purchases? |
| Manufacturer export | CSV, XLSX and TSV plus a dedicated multi-level exporter. Field preservation varies by route. | Project table CSV plus general data exports in several formats and detail levels. | What survives as structured data, what is a name only, and what requires the companion manifest? |
Keep ACT-420 A as the only approved actuator in the JAW B test. InvenTree's documented BOM substitute mechanism allows stock from either linked part to be allocated. An unrestricted ACT-410 A substitute would exceed ECO-027's approval.
What survives the UI, API and CSV routes?
An application programming interface (API) can preserve relationships differently from a spreadsheet download. Table 3 compares those routes with CSV and other exports. It records documented or source-visible fields, with exact output still to verify in the trial. See the InvenTree export sources and Part-DB export sources.
| Route | Documented or source-visible data | Release information to test or supply separately |
|---|---|---|
| InvenTree ordinary BOM table export | Current-filter rows, quantities and item fields. Expanding a subassembly on screen does not include its child rows in this export. | Exported revision identity, every parent-child path and substitution conditions. Do not infer export completeness from the UI |
| InvenTree BOM Exporter | All-level option, BOM level and rolled-up quantity, with optional supplier/manufacturer, substitute and parameter data. The 1.7.3 substitute columns contain Part display names. | Exact revision fields, approved-stack constraints, drawing and firmware references, and approval evidence alongside substitute names |
| InvenTree REST BOM data | Parent/component record references, quantity and note, with optional detail/substitute data. Part detail includes IPN and revision. | Follow linked records and pagination, preserve their mapping and retrieve attached release evidence. A CSV conversion is a separate preservation test |
| Part-DB project CSV | Selected project rows with quantity, linked part ID, IPN, mount names and supported descriptive/stock fields. Export follows current search and column selection, using semicolon-delimited CSV. | The reviewed column mapping has no explicit item-revision, CFG, ECO, firmware or drawing-reference column. Retain those in a manifest and prove the linked definitions |
| Part-DB API or fuller data export | BOM-entry API exposes project/part links, quantity, mount names and comments. General JSON/XML/YAML exports and “Full” detail can include nested associated data and attachment records. | Complete historical package, downloaded attachment bytes and recursive gripper structure remain to verify. API docs label the interface beta. |
For both tools, retain separate source and export files, the selected record IDs, application version and export options. A filename or URL pointing to a current record does not preserve its past contents.
Run a reproducible manufacturer-package test
Run this plan in an isolated evaluation instance with synthetic data. The expected outcomes below are not reported test results. Use the versions above, or record the newer version and repeat the evidence checks before relying on this comparison.
- Create the original definition. Enter the two orderable items, common chassis/controller and JAW A/VAC A relationships. Select DW-GR-J210 A and FW-310 1.4.0 for JAW A. Save REL-WP-01's manifest and record the application IDs that map to each engineering item/revision.
- Record the earlier build. Associate WP-001–WP-005 with the original jaw release. Export the original package and retain its files. An editor and a separate read-only manufacturing user should be able to resolve WP-005 without private explanation.
- Introduce the bounded change. Add GR-J210 B with ACT-420 A, ENC-520 A, DW-GR-J210 B, FW-310 1.5.0 and TV-027. Record ECO-027 and REL-WP-02. Select twelve JAW B and eight unchanged VAC A units for B-2412. Confirm the earlier structure remains A and no VAC jaw encoder appears.
- Reconcile physical totals. Multiply each gripper path by its variant quantity. Expect twenty controllers/chassis, twelve replacement actuators/encoders and eight pumps. Check that complete-controller electronics are excluded and firmware generates no purchased-stock requirement.
- Compare export routes. For InvenTree, export the ordinary BOM table and the BOM Exporter with all levels and total quantities explicitly selected, then retrieve the same records through the documented REST schema. For Part-DB, export the project table CSV, export the relevant parts/project data at Full detail, then retrieve project BOM entries through its documented schema. Record filters, columns, part assignment and every page retrieved.
- Build a field-preservation ledger. Compare source to UI, API and exported files for item ID, revision, parent, quantity/unit, variant, substitute condition, drawing, firmware, approval and serial applicability. Classify each as structured and preserved, manually supplied, or unresolved. Count an attachment only after retrieving the intended file bytes, not merely its current URL.
- Recover the earlier package. Ask another engineer to select WP-005 and WP-006 after the change. WP-005 must resolve to JAW A, the old actuator and encoder, drawing A and FW 1.4.0. WP-006 must resolve to JAW B, drawing B and FW 1.5.0. Record whether this uses retained records, a saved manifest or reconstruction from history. Read historical copies without reverting the live inventory.
A satisfactory exit package lets the manufacturer perform those selections without your application account. If a field is manual, assign its owner and retain the method. If an export omits a required relationship, resolve that gap before treating the tool as the release authority.
Use the open-source PLM guide when approval, retained baselines and evidence applicability drive the purchase. Inventory history and a formal engineering release answer different questions. Use the open-source MRP guide when the next question is which receipts arrive before material issue and what must be bought or rescheduled. A current stock display alone does not establish dated planning.
What operating work comes with self-hosting?
Name an owner for the database, attachments, configuration and upgrades. Pin the evaluated server and container versions, record plugins, and test a restore in a separate environment. Keep an independent engineering package that a manufacturer can read without restoring your server.
InvenTree documents database and media backups, with restores tied to the same installed version. Its automatic daily backup option is disabled by default. Permissions are assigned through groups and roles for parts, BOMs, stock and builds. Confirm those controls under the builder and engineer accounts used in your trial.
Part-DB documents a full-backup command that collects database, attachment and configuration data. Its Docker guide requires persistent volumes and an HTTPS reverse proxy when exposed beyond a trusted local network. It documents granular permissions, while API access requires its own permissions and remains described as beta.
Include upgrade rehearsal, authentication, account removal, backup checks and recovery ownership in the adoption decision. Source access reduces dependence on a proprietary code grant, but does not supply those operations or a reviewed migration automatically.
Which should a robotics team try first?
Try InvenTree first when one workflow must connect nested robot assemblies to physical parts and build stock. Its documented model gives more direct evidence for that scoped job. The decision still depends on your test of released revision selection and the complete export package.
Try Part-DB first when the immediate need is a shared electronics inventory and project BOM workflow, and the team already has a controlled process for robot releases. Test project-built grippers and exact field preservation before extending that role to the whole robot.
For either tool, use the same WP-200 package and retain the preservation ledger. Choose the workflow you can operate and reconstruct, with every manual release step visible to the people who will build the robots.