Research: what can we learn from Filametrics and 3D Spool Tracker (AI-assisted filament managers)?

[sherrmann]

Context

During the LLM-integration brainstorm (docs/llm-integration-brainstorm.md on claude/llm-integration-brainstorm-qmjmhn), two commercial products surfaced that already ship AI-assisted filament management. Both validate the Scan-to-Spool direction — and they are interesting in general, beyond their AI features. Nobody offers a self-hosted equivalent, which is exactly the lane this fork can own.

Subjects

• Filametrics — "Add filaments by scanning the box label … let our AI slicer reader do the heavy lifting"; upload a slicer screenshot and usage data is extracted and applied to inventory; positions itself as a smart filament management and automation platform.
• 3D Spool Tracker — "Snap a label or box and brand, material, color, and recommended settings populate themselves"; mobile-first tracker.

What to analyze

• Full feature inventory of both, AI and non-AI: onboarding, mobile UX, printer/slicer integrations, notifications, automation rules, reporting, pricing/positioning.
• Exact UX of their AI flows: capture UI, review/confirm step, behavior when extraction is wrong, how latency is masked, what happens with unknown brands.
• Filametrics' slicer-screenshot usage extraction: which slicers, what data it pulls, how it maps to our A3 idea (slicer-screenshot usage logging).
• Data handling: do they persist photos, what runs client- vs server-side, any hints about providers/models used.
• Non-AI ideas worth adopting into Spoolman NG regardless of the LLM work (inventory UX details, low-stock/automation behavior, settings recommendations attached to filaments).

Deliverable

A short comparison write-up (issue comment or docs/) mapping each worthwhile capability to the brainstorm's idea clusters (A1/A3/B/C/D/E) with a build / skip / later recommendation for each — minding our constraints: self-hosted, provider-neutral (any OpenAI-compatible endpoint), photos ephemeral, AI invisible unless enabled, no emoji in UI.

Related

• docs/llm-integration-brainstorm.md — section 1 (prior art) and section 3 (idea clusters).

[sherrmann]

We just want to grab whatever ideas that we can, it's not just related to what's mentioned in the issue!

[sherrmann]

Findings (web research pass, 2026-07-23)

Sourced from both products' public sites. Neither has meaningful independent coverage (Reddit/blog searches surface Spoolman, Filadex, Spoolstock and friends instead) — both are young, small products: Filametrics claims "2,000+ makers", 3D Spool Tracker "14,200+ active makers". In-app UX details (review/confirm steps, error handling, latency masking) are account-gated; a hands-on pass with the free tiers is the natural follow-up and is cheap (3DST gives 10 free AI scans/month).

Feature inventory

Filametrics (cloud-only, web app; free tier / $3.99/mo / $199 lifetime):

• Inventory by brand/color/material/weight, low-stock alerts, QR/NFC/RFID scanning, weight history and analytics.
• AI Slicer Reader (signature feature): screenshot your slicer's usage screen, "Our AI reads it and maps it to your spools. Done." Claimed to support "most major" slicers (no list published).
• AI label/box scanning to add filaments.
• Kanban project management with estimated filament per job, 80+ project templates, out-of-stock alerts across queued jobs, 3MF upload for usage tracking.
• Shopify and TikTok order-import integrations (seller-oriented).

3D Spool Tracker (cloud, web + mobile app; free "Maker" tier / Pro $5/mo or $35/yr; hardware $99):

• Unlimited spools/projects, multi-location (shelves, dryers, rooms), low-stock alerts with one-click reorder links (Amazon affiliate funded), gcode file storage, CSV/JSON export, public REST API (Pro).
• AI photo identification of labels/boxes: "brand, material, color, and recommended settings populate themselves", "within 0.3 seconds". Free tier meters it at 10 scans/month; Pro is unlimited.
• Catalog of 2,000+ filaments with pre-loaded specs incl. drying schedules; community-contributed, reviewed before publishing.
• Spool Inspector hardware: $99 Wi-Fi scale (5 kg, ±1 g) with built-in QR scanner, real-time sync, drying-time tracking. Their stated business model is free software funded by hardware sales.
• Connected-printer support named for Bambu Lab, Prusa, Polymaker, Sunlu.

Analysis against the issue's questions

• AI UX flows: both describe capture → auto-populate; neither documents a review/confirm step publicly. The "0.3 seconds" claim strongly suggests their scan is (at least partly) a catalog lookup keyed by recognition output, not raw form-filling from OCR — i.e. both converge on the same design we chose: extraction is a query against known-good data, not trusted directly. Filametrics' slicer reader explicitly "maps it to your spools", so matching-to-existing-inventory is part of their extraction flow too — same conclusion as our decision ci: skip the redundant master build when the commit is a release tag #10 (own library first).
• Slicer-screenshot extraction (our A3): validated as a headline commercial feature. It serves exactly the users our Moonraker path misses (Bambu/SD-card workflows). No public info on which slicers or what fields beyond weight.
• Data handling: neither publishes whether photos are retained; both are cloud-only, so photos leave the user's network by construction. Providers/models are not discernible. Our ephemeral-photos + BYO-endpoint design (decisions feat: Spoolman NG release process (CalVer, latest/sha tags, fork identity) #8/fix(ci): remove duplicate body key (invalid workflow file) #9) is a clean differentiator we can state plainly in docs.
• Monetization insight: both meter or paywall AI scans. Self-hosted BYO-provider means we never need metering — a user's Ollama box or their own API key is their cost. Worth saying out loud in the eventual announcement.

Mapping to brainstorm clusters (build / skip / later)

Capability seen in the wild	Cluster	Verdict
Label/box photo → auto-populate against a catalog	A1	Build (agreed; design already matches both competitors' apparent architecture, plus our own-library-first refinement)
Slicer-screenshot usage extraction	A3	Build as fast-follow after A1 — commercially validated, low marginal effort on A1's plumbing
"Recommended settings populate themselves" (temps, drying)	A1	Build via SpoolmanDB match — we have 6,957 catalog entries to their 2,000
3MF upload for usage tracking	A3-adjacent	Investigate — we already ship 3MF import (client/src/pages/settings/threeMfImport.tsx); check whether extending it to per-project usage logging covers this without any AI
Metered AI scans / AI paywall	—	Skip — BYO endpoint makes it unnecessary
Wi-Fi scale + QR hardware (Spool Inspector)	outside AI track	Later, separate issue — Spoolman already has the PUT /measure API but no reference hardware; user story #15 already asks for ESPHome/Arduino reference firmware. A documented DIY recipe (ESP32 + load cell + optional QR/NFC) would answer their $99 product in kind
Kanban projects + "will queued jobs run out"	B3-adjacent	Later — insight-card material once usage attribution lands (user story #64 territory)
Low-stock one-click reorder links	—	Later, non-AI — natural extension of the existing low-stock page + shop/order tracking
Drying schedules / dryer tracking	—	Later, non-AI — fits extra fields + locations (dryboxes) we already have
Shopify/TikTok order import	—	Skip — seller tooling, out of scope for a self-hosted inventory

Suggested next steps

1. Hands-on pass with both free tiers to fill the UX-flow gaps (capture UI, confirm step, failure behavior) before finalizing the A1 review screen.
2. Keep this issue open until that pass is done; the table above then gets confirmed/corrected.
3. File the Spool Inspector-style reference-hardware idea as its own non-AI issue (ties to user story ci: bump GitHub Actions off deprecated Node 20 runtime #15).

Sources: Filametrics, 3D Spool Tracker, search cross-checks: Spoolstock, Filadex, filament-tracker, Spool Shelf.

---

Generated by Claude Code