Trim: Filament Calibration

The calibration wizard that knows your printer has more than one nozzle.

Trim walks one spool of filament through the tests that produce a filament preset — temperature, flow, pressure advance, retraction, max flow, shrinkage — then generates the preset for your Orca-family slicer and says exactly where every number goes. Unlike every other calibration tool, including the upstream project this forked from, it models a printer as having one or more nozzles, and each nozzle's feed path decides its ranges: a bowden auxiliary is never handed direct-drive numbers.

You run the test prints yourself, in your own slicer. Trim never talks to the printer.

It runs as a local-first web app and as a desktop app. No account, no backend, no telemetry.

Get it · Why per-nozzle · Your first calibration · Slicer support · Limits

What it asks of you. Nine calibration tests ship in the default order (ten with the dual-nozzle ooze step), each one a print you set up and run in your own slicer — a first pass is an afternoon's work, not five minutes. Trim never drives the slicer and never starts a print. It is built to be abandoned mid-flow: state is saved as you go, so you can close it while a print runs and come back days later to the step you left, with every carried-forward value still showing where it came from. It assumes you can find your way around slicer settings. It does not assume you know which calibrations exist, what order they run in, or which field each result belongs in.

Get it

Desktop app — no release has been published yet. The release workflow is written to build a Windows .exe (NSIS), a macOS universal .dmg, and Linux .deb and .AppImage from a tag, but the Releases page is empty, so today the only way to get the desktop app is to build it yourself: npm run tauri build. The desktop build is the one that can scan your slicer installation and write a preset into it — on Windows. Direct install is verified and enabled only on Windows today; on macOS and Linux the desktop build scans, generates, previews and exports, and the wizard points you at manual import. See Slicer support.

Browser — the app is a static bundle you can host yourself (Nginx, Docker) or run locally. There is no public instance to try — you host it or run it yourself. Everything works except writing into a slicer: the browser build downloads the generated .json preset for manual import instead.

The name

Trim is the aviation sense. Trimming an aircraft sets the controls so it holds its attitude without constant correction: you trim once per configuration. You calibrate once per filament and nozzle, and the printer then prints correctly without further intervention. Same structure. The app was called PerfectFit X2D up to 2.0.0; the X2D in the name claimed one printer, and it works with any of them. See CHANGELOG.md for what the 3.0.0 rename means for existing data.

This is a fork. Trim builds on PerfectFit by Aaron Taylor and adds per-nozzle calibration. It ships under its own product name and bundle identifier, so it installs alongside upstream rather than over it and keeps a separate data store. Every download link, issue link, and install guide in this repository refers to this fork; upstream is credited, not linked as a download.

Why this exists: the second nozzle

Trim models a printer as having one or more physical nozzles, and each nozzle's feed path decides its suggested ranges. On a Bambu Lab X2D the main nozzle is direct drive and gets automatic Flow Dynamics calibration from the printer; the auxiliary is bowden-fed from a rear-mounted extruder and gets none. Community reports put its K an order of magnitude above the main nozzle's — typically 0.5–1.0 against 0–0.1 — and its retraction correspondingly longer; Bambu's own machine profile ships 2 mm on the auxiliary against 0.8 mm on the main. Those are the numbers the suggested ranges are built from, and they are community-sourced except the shipped defaults.

If your auxiliary nozzle strings or oozes, that is the case this exists for. Its stock defaults ooze, and Bambu Studio has a reported bug — BambuStudio #10404 — where a filament preset whose Bowden Extruder retraction override is left unset silently falls back to the main nozzle's 0.8 mm default, far too little for a bowden path. Trim carries that warning into the suggested ranges, the retraction step and the ooze-control checklist. Reported on Bambu's own tracker, not documented in the wiki.

A bowden auxiliary is never handed direct-drive numbers here:

• A printer profile carries a list of nozzles, each with its own feed path and optional speed/acceleration caps.
• A calibration project targets one nozzle. Selecting a bowden auxiliary switches the pressure-advance range to 0–1.0 in 0.02 steps and retraction to 2–6 mm in 0.5 mm steps, and deliberately ignores the retraction range saved on the printer profile, because that range describes the main feed path.
• A guided session holds one independent working profile per nozzle, so a value measured on one nozzle never bleeds into the other.
• Nozzle count is read only from the printer profile's own nozzle list — never inferred from how many value slots a slicer preset happens to have.

NOZZLE 1 · MAIN — Main (direct drive). NOZZLE 2 · AUXILIARY — Auxiliary (bowden), with its own caps of ≤ 200 MM/S and ≤ 1000 MM/S². Machine limits above it: max nozzle 300 °C, max bed 100 °C, max flow unmeasured (—).

The X2D is the machine the research is based on, and the only machine whose nozzle layout ships as a template. The H2D and H2C appear in the printer database as machine specs only — the database has no per-nozzle field — so their nozzles and feed paths must be entered by hand. No calibration has ever been carried through to a print on any machine — see Limits.

Your first calibration

Launch it and the dashboard opens on a panel of six instrument dials. An unmeasured value is never rendered as zero — it reads an em-dash and its dial stays dark. That distinction runs through the whole app: "not measured" and "zero" are different facts.

Nozzle temp, flow ratio, pressure advance, retraction, max flow and shrinkage, unmeasured. Calls to action: 1 · ADD YOUR PRINTER, 2 · START CALIBRATING.

1. Add your printer

PRINTERS → + ADD PRINTER. Pick from 379 known printers across 64 manufacturers to fill the specs in automatically, or enter your own. Machine limits (max nozzle and bed temperature, max flow) are used to warn you before a suggested setting could exceed what the machine can safely do. Where the database publishes no limit, the form uses a stand-in — 260 °C nozzle, 100 °C bed — and labels it as not a specification.

Nozzles are a repeatable list: one row per physical nozzle you can fit a hotend into, each with its own Feed (direct drive, or bowden/remote) and optional max speed and acceleration. There is a one-click Bambu Lab X2D quick-fill that builds the two-nozzle layout for you.

NOZZLE 1 · MAIN — Main (direct drive), Feed "Direct drive", max speed and accel left at "no cap", help text "Feed path drives PA and retraction suggestions." NOZZLE 2 · AUXILIARY — Auxiliary (bowden), "Bowden / remote", 200 mm/s, 1000 mm/s². Below them, + ADD NOZZLE.

2. Create a calibration project

PROJECTS → + NEW PROJECT. A project is one spool on one nozzle. If the printer has more than one nozzle the form asks which — NOZZLE UNDER CALIBRATION — and the cards state what changes:

• PANEL 1 · MAIN — "Standard test ranges — the extruder motor sits on the toolhead, so the pressure response is short and predictable."
• PANEL 2 · AUXILIARY — "Wider K (0–1) and retraction (2–6 mm) ranges, and the ooze-control step is pre-selected below."

Picking the bowden auxiliary ticks the optional Dual-Nozzle Ooze Control step, which is a checklist and one verification print rather than a number to find. It checks levers — prime tower, travel speed, wipe, z-hop, the #10404 override — that Trim names but does not compute.

Chips: BOWDEN / REMOTE FEED, ≤ 200 MM/S, ≤ 1000 MM/S². A "Compatibility unknown" panel reads "PerfectFit could not determine whether PETG works on Auxiliary (bowden). Nothing was found to read, which is not the same as approval." — one of the rename leftovers noted under Limits.

Where installed slicer preset data can be read, Trim resolves the inherits chain and gives a graded verdict on whether that material suits that nozzle. It is a warning only — it never blocks a calibration, and unknown is explicitly not an all-clear. Override a warning and the verdict is recorded verbatim on the project, so later output cannot read as though the app had approved the pairing.

3. Work the guided session

One screen per test. The session header pins to the top of the window and carries the step number, the test name, a status, the step-dot track and how many steps are left, so the context does not scroll away while you read a long instruction page.

Each step opens with what is CARRIED INTO THIS TEST: a table of setting, reading, and where it came from. Every pre-filled number states its provenance — calibrated, user override, carried forward, printer default, material default, or unset. A material default does not satisfy a step's required input; it shows as a placeholder and the step still counts as missing it. Change an upstream value and downstream steps are marked stale with a reason, rather than silently keeping an obsolete result.

Where the shipped instruction data does not cover something, the step says so under NOT COVERED BY THE SHIPPED INSTRUCTIONS instead of inventing a menu path, and links the official documentation with the date it was verified.

Header: "Step 4 of 10 · PRESSURE ADVANCE · READY", the step-dot track, "6 STEPS LEFT". Below it a "NOT COVERED BY THE SHIPPED INSTRUCTIONS" panel — still saying PerfectFit, see Limits — and a CARRIED INTO THIS TEST table: nozzle temperature 240 °C and flow ratio 0.902, both badged CALIBRATED and naming the test that measured them.

4. Read the result, and check the arithmetic

Every calculation returns its inputs, the formula as a string, the substitution, the result and any warnings. Thirteen formulas ship this way, and rounding is disclosed. A user who distrusts a result can check it instead of accepting it.

RESULT 0.93. FORMULA "NewFlowRatio = OldFlowRatio + modifier", SUBSTITUTION "0.95 + −0.02 = 0.9300", RESULT "= 0.93", the note "Rounded to 3 decimals from 0.93.", and VALUES TO ENTER: "Flow ratio (decimal — not a percentage)" 0.93.

Back on the dashboard, a project's dials light as values land. The nozzle-panels strip shows which of the printer's nozzles this project owns and which are still untouched.

POLYMAKER PETG · PolyLite PETG · Galaxy Black · Bambu Lab X2D (0.4 mm) · brass, CONFIDENCE 42. Lit: NOZZLE TEMP 240 °C, FLOW RATIO 0.902, PRESS ADVANCE 0.36; dark: RETRACTION, MAX FLOW, SHRINKAGE. A NOZZLE PANELS · BAMBU LAB X2D strip marks MAIN (DIRECT DRIVE) — NO PROJECT YET and AUXILIARY (BOWDEN) — THIS PROJECT · BOWDEN · CALIBRATED. Steps 4 / 10, NEXT · FLOW RE-CHECK.

Two experience levels ship: Coach (plain-language callouts, good/bad examples, "I'm not sure" helpers) and Expert (ranges, formulas, destinations). Two working surfaces ship side by side: the guided session above, and the classic step-by-step wizard, reachable from every step.

The ten calibration tests

Nine ship in every project's default order:

1. Temperature
2. Flow ratio, pass 1
3. Flow ratio, pass 2
4. Pressure advance
5. Flow re-check
6. Retraction
7. Max volumetric speed
8. Shrinkage
9. Final verification

The tenth — Dual-Nozzle Ooze Control — lives outside the default order and is spliced in immediately before final verification for projects that opt into it, which in practice means bowden-auxiliary projects.

Each test carries its purpose, why it sits in that position, an expanded explanation, dependencies, prerequisites, per-slicer methods, an evaluation guide, result precision, the slicer destination, and version notes.

Where the shipped order contradicts vendor guidance, the app says so. It quotes Bambu's own sentence putting Flow Dynamics before Flow Rate, then explains why flow runs first here — and it labels the flow re-check as this project's own step, not sourced practice, because that is the step that closes the circular dependency between pressure advance and flow. Absences are explained too: bed temperature and first layer are stated as not-per-filament jobs, and retraction's late position is stated rather than left looking like an oversight.

Generating and installing the slicer preset

Generation is clone-and-patch: the base preset is deep-cloned, and only values backed by a completed calibration step with a final value are patched. Unknown fields, arrays and inheritance survive, and the count of preserved fields is reported. The keys written are nozzle_temperature_initial_layer, nozzle_temperature, filament_flow_ratio, pressure_advance (plus enable_pressure_advance), filament_retraction_length, filament_retraction_speed, filament_max_volumetric_speed and filament_shrink.

Five stages: 1. SLICER → 2. BASE PROFILE → 3. CONFIGURE → 4. PREVIEW & VALIDATE → 5. INSTALL / EXPORT. Trim ranks and recommends which stock preset to clone from, deterministically, with reasons.

Which value slot the calibration is written to is resolved from the base preset's own extruder-variant legend, read by name — never from the nozzle index. If the base cannot address the calibrated nozzle, the value is withheld and reported with a reason instead of written to a slot every nozzle reads:

"This base preset cannot hold a value for nozzle 2 — It carries only 1 value slot(s), but this project calibrated nozzle 2 (Auxiliary (bowden)). Orca-family slicers apply a single-slot value to EVERY nozzle, so installing it would give every nozzle — including the main one — nozzle 2's calibration. Go back and pick a base preset for this machine that carries 2 value slots." The Profile name field above it reads "PerfectFit - Polymaker PETG Galaxy Black @ Bambu Lab X2D 0.4" — the rename leftover noted under Limits.

Given a base wide enough, the preview lists every changed field with its old value, its new value and its slot number, and patches only the calibrated nozzle's slot. Validation runs before install, and warnings such as withheld fields or printer-limit breaches must be explicitly acknowledged.

Four changes, each naming slot 2: nozzle temperature 255 → 240 °C, flow ratio 0.95 → 0.902, pressure advance 0.02 → 0.36, and enable pressure advance 0 → 1. Below them: "16 field(s) preserved from the base profile. Identity fields updated: filament_id, filament_settings_id, from, name." and the top of a VALIDATION panel below.

Then export, install, or keep it in the project. In the browser build automatic installation is reported as unavailable and no button is rendered for it:

"1. EXPORT PROFILE FILE" with an EXPORT PROFILE button, "2. INSTALL AUTOMATICALLY" reading "Automatic installation requires the Trim desktop app." with no button, and "3. SAVE INSIDE TRIM" with a SAVE IN PROJECT button.

On the desktop build, direct installation writes into the slicer's user preset directory (e.g. %APPDATA%\OrcaSlicer\user\<account>\filament\). It backs up the affected files first with SHA-256 checksums, writes to a temp file, verifies, atomically moves it into place, re-verifies, and rolls back on failure. It refuses to run while the slicer is open, and will not replace an existing preset unless you explicitly confirm replacement. Backups are restorable from the app. The desktop build also snapshots the slicer's preset library before calibration starts, not only at install time; snapshots are listed, restorable and deletable from Settings.

Both behaviours can be switched off in Settings → Experimental features (Slicer profile generation, Automatic profile installation).

Bambu Studio and pressure advance

Bambu Studio does not write the native pressure_advance field into sliced G-code for Bambu machines — the printer's own Flow Dynamics owns K. That is this project's own finding from inspecting sliced output (checked 2026-07), not something Bambu documents. The only route Trim offers is optionally baking the measured K into the filament start G-code as M900 K<v> L1000 M10, which additionally requires you to set Flow Dynamics Calibration = Off in the send-print-job dialog yourself. Re-generating replaces the injected line rather than stacking a second one.

That bake is refused outright on any preset whose slots span more than one feed path — the X2D shape — because filament_start_gcode is a single string for the whole filament and has no per-slot form. The "apply to all value slots" checkbox cannot switch the refusal off. On an X2D-shaped preset the auxiliary nozzle's K has to be entered by hand in Bambu Studio.

Slicer support

Six install flavours across five families are modelled, each with its data directory, config file name, executable candidates and process names read off a real installation with a recorded verification date: Orca Slicer, Bambu Studio, Bambu Studio (Beta), Snapmaker Orca, ElegooSlicer, Flash Studio (Orca-Flashforge).

Direct install is gated per slicer + version + platform by an explicit registry. Scanning, parsing, generating and exporting stay available on any version; direct install is off unless that exact combination is verified.

Slicer	Version verified	Direct install	Last checked
Orca Slicer	2.4.2	yes, Windows	2026-07-19
Bambu Studio	02.07.01.62	yes, Windows	2026-07-19
Bambu Studio (Beta)	02.08.01.55	no — gated off	2026-08-01
Snapmaker Orca	01.10.01.50	yes, Windows	2026-07-19
ElegooSlicer	1.5.2.2	yes, Windows	2026-07-19
Flash Studio (Orca-Flashforge)	01.10.01.50	yes, Windows	2026-07-19

Every entry declares platforms: ['windows'], so direct install is Windows-only today. On macOS and Linux the wizard reports it as unverified and points you at export-and-import.

"Verified" here means a human ran a full manual end-to-end pass: transactional install, verified backup, the preset appearing in the slicer's filament list with the right values, a model slicing cleanly, and a backup restore returning the directory byte-identical to baseline. A slicer was verified. A printer was not.

That definition covers the five rows marked yes. The Bambu Studio (Beta) row is different: its date records a read-only inspection of a real 02.08.01.55 install — directory layout, config schema and preset schema were read, nothing was generated, installed, sliced or restored.

Slicer instruction data

Step-by-step instructions are version-aware data, not code: src/data/slicers.ts holds per-slicer, per-version content with a verifiedOn date and a docs URL. All ten calibration tests have their own menu path, numbered steps and save-to destination in each slicer, plus gotchas on every test that has any and a disable-first list on the tests where something has to be switched off first (the flow and pressure-advance tests). Updating for a new release means editing one data entry. Research notes with sources and verified formulas: docs/RESEARCH.md.

Currently Orca Slicer 2.4.x and Bambu Studio 1.7+, both verified 2026-07-23 against the official wikis. Note the two version schemes in play: the instruction data uses 1.7+ as a floor for Bambu Studio's documented UI generation (its own note records that the check was made against a 2.7.x build), while the install registry above keys on the build number the slicer reports (02.07., 02.08.). They answer different questions and are not in conflict.

Assumptions worth re-verifying when a new slicer version ships:

• Calibration menu still at top bar → Calibration (Orca) / Calibration tab plus the Develop Mode title-bar menu (Bambu Studio)
• Menu entry labels still differ per slicer: Orca Flow ratio / Retraction / top-level Max flowrate vs Bambu Flow rate ▸ Coarse-Fine / Retraction test / More... ▸ Max flowrate
• Temp tower still steps 5 °C per block; retraction/PA towers still step once per mm of height
• Flow YOLO modifiers still ±0.05 @ 0.01; Pass 2 still −9…0%
• Bambu Studio's Develop Mode still exposes retraction, Max flowrate and VFA calibration while a Bambu printer stays selected

The New Project screen defaults to Bambu Studio, because the bowden auxiliary that gets no automatic calibration from the printer is the case this product exists for.

Limits

What Trim does not do.

• It never talks to your printer and never starts a print. There is no printer-communication code of any kind in the repository. Every calibration test is printed by you, from your slicer.
• It cannot push a K value to a Bambu machine. See Bambu Studio and pressure advance. On an X2D-shaped preset even the M900 route is refused, and the auxiliary K must be typed in by hand.
• It has never been run end-to-end against a real printer. No calibration in this project's history has been carried through to a finished print on hardware. The CI gate proves the logic over temporary directories; the Rust tests that exercise install and restore against a genuine slicer installation are marked #[ignore] and never run in CI. A slicer has been verified by hand. A print has not.
• Automatic installation into Bambu Studio Beta (02.08.x) is deliberately gated off. That registry entry carries directInstallVerified: false and profileGenerationVerified: false, because no preset has been generated or installed against 02.08.01.55. Detection, scanning and export work there; automatic installation does not.
• Direct install is Windows-only today, and the browser build cannot install into a slicer at all — it downloads a .json for manual import.
• It does not prepare, configure or slice the calibration test prints. An earlier plan to drive the slicer automatically was removed before release; see What was removed, and what survived.
• A compatibility verdict never blocks a calibration. It is typed so that no caller can turn it into a block, and unknown is documented as not an all-clear.
• A per-nozzle install can be deliberately incomplete. When a base preset cannot address the calibrated nozzle, the value is withheld and reported as a skipped field with a reason, rather than written somewhere every nozzle would read it.
• The X2D is the only machine with a hand-made dual-nozzle template. The shipped printer database has no per-nozzle field at all — it carries an extruder count, not a nozzle list — so any other multi-nozzle machine's nozzles and feed paths must be entered by hand before per-nozzle behaviour applies.
• Suggested ranges are conservative starting points, not guarantees — spool labels and datasheets always win. The machine limits they are clamped against may themselves be app fallbacks rather than specifications, and are labelled when they are.
• It never calibrates a chamber temperature and never writes chamber_temperatures into a preset. Chamber is guidance only, and it refuses to name a number unless both a machine ceiling and a sourced material ceiling exist.
• It never writes travel speed, wipe or z-hop. Those are named as ooze levers to check, not values the app computes — there is no measurement behind them.
• Neither slicer generates a shrinkage test in-slicer, so that step needs an external tool or a large object of your own to measure; final verification needs a downloaded model or a part of your own too. See Model licensing.
• Orca's built-in calibration tests always target filament slot 1 and expose no extruder picker, so multi-tool printers need a manual filament reassignment — the wizard says so rather than pretending the limitation isn't there.
• Photos are stored and exported but never analyzed. The analysis field is typed null and reserved.
• No account, no backend, no cloud sync, no telemetry, and no network layer at all in the frontend. All data lives in IndexedDB and localStorage under one origin: clearing site data destroys everything, and serving the app from a different domain or port means starting fresh.
• The calibration card's QR does not embed the calibration. It links to the app URL plus the project id, so it resolves only on the same hosted instance and browser profile. The printed values are what make the card portable.
• No release binaries have been published. The pipeline has no code-signing certificate and no notarization step, so anything it does eventually produce will be unsigned — not code-signed on Windows, not notarized on macOS — and will trip Windows SmartScreen and macOS Gatekeeper.
• There are no users, customers, testimonials, download counts, reviews, stars or endorsements, and no company behind this. Nothing here should be read as claiming otherwise.

Known leftover from the 3.0.0 rename: the old name still appears in user-facing copy in about 67 places across src/. The most visible is the default name Trim proposes for a generated preset, which still begins PerfectFit - (an editable display string, not a stored identifier), but the compatibility verdicts and several guided-session instruction notes also still say PerfectFit — visible in four of the screenshots above. This is unfinished, not deliberate; the deliberately frozen identifiers are listed under Data storage & backups.

What was removed, and what survived

An earlier plan to have the app configure and slice each calibration test automatically was removed before release: it could only ever have driven OrcaSlicer, and a half-working path that silently produces the wrong test plate is worse than no path at all. The removal went down to the desktop commands — they are no longer registered, and the slicing-engine abstraction, Orca preset resolver, 3mf assembler and config merge are deleted rather than left unreachable.

Three things deliberately survived it, because a reader grepping the repository will find them:

• src/automatedCalibration/ — not the removed feature. It holds the guided session's step workflow registry, dependency graph, staleness detection and per-nozzle working profiles, and the session cannot run without it. The directory name is historical.
• docs/AUTOMATED_CALIBRATION.md — the record of what was removed and why.
• automatedCalibration: false in DEFAULT_EXPERIMENTAL_FEATURES — an explicit tombstone, commented as such. Nothing reads it.

The reasoning, and the condition on ever reviving it, is in docs/X2D_ENHANCEMENT_PLAN.md.

Other things that ship

• Confidence score (0–100) from weighted per-step contributions, scaled by your own confidence rating, reduced when a retest is flagged, and scored against the project's own step plan so optional steps only count for projects that carry them.
• Smart retest recommendations — a ranked list of suspects with reasons, never a verdict. Eleven failed-verification categories each map to ranked likely causes.
• A printable one-page calibration card: six drawn dials, a plain-text value table to type into the slicer, the confidence placard, the nozzle badge and a QR code. Plus a full printable report, and copy-final-values-to-clipboard.
• A per-project timeline logging every event: created, started, value-set, completed, retest, skipped, reset, note.
• 14 material presets with suggested temperature, tower and MVS ranges, and mandatory chamber guidance that distinguishes "run it hot", "keep it ambient" and "unknown" — the ambient case exists specifically to prevent heat creep on PLA, PETG and TPU.
• A 1.5 mm flexible-filament retraction cap, enforced from a single constant across the suggestion layer, the entry form, the desktop write path, the printed card and the report.
• Photos attachable to any calibration attempt from both the wizard and the session, stored locally and included in backups.
• Light, dark and auto themes (auto follows the OS and re-lights if it flips), a large-text mode that scales the whole type system, prefers-reduced-motion honoured globally, visible :focus-visible rings on every interactive element, a real radiogroup with roving tabindex for the panel-lighting switch, and dedicated print stylesheets that redraw the gauges as line work so the card and report print as real deliverables.
• A service worker, so the app works offline after first load and installs as a PWA when served over HTTP(S).

Data storage & backups

What	Where
Projects, printer profiles, photos	IndexedDB (perfectfit-db)
Settings, in-progress form drafts	localStorage (perfectfit. prefix)
Backups	JSON files you export (Settings → Backup)
Slicer preset backups (desktop build only)	{app data}/slicer-backups/{slicer}/{backup id}/ — e.g. %APPDATA%\io.github.espentruls.trim\slicer-backups\ on Windows. Written automatically before any preset install, restorable from Settings → Slicer profile backups.

perfectfit-db, the perfectfit. localStorage prefix and the perfectfit-filament-calibration-wizard marker inside every export file are not missed renames. They are the identity of stored data, deliberately frozen at 3.0.0: IndexedDB opens databases by name, so a new name is a new empty database, and changing the export marker would make Trim reject its own 1.x and 2.x backup files. The full list and the reasoning are in CHANGELOG.md under What deliberately did NOT get renamed.

• Backup: Settings → Export all data (optionally with photos, base64-embedded).
• Restore: Settings → Restore from backup. Import is all-or-nothing — every record is staged in memory and committed in one IndexedDB transaction, so a failure leaves the database untouched. Colliding project ids are imported as copies; colliding printer ids are left alone and reported as skipped; photos that cannot be decoded are reported rather than silently dropped.
• Backups are schema-versioned (v6 current) with forward migration from v1. A file from a newer schema is refused rather than half-read, and malformed nozzle lists or nozzle indices are dropped rather than guessed. The settings block is sanitized before it is applied — an out-of-range max-flow safety margin is discarded in favour of the default rather than trusted.
• Single projects can be exported/imported from the dashboard (printer profile embedded).
• Clearing browser site data deletes everything — back up first.

Development

Requirements, running the dev server, the test suites and what they do and do not cover, the architecture map, regenerating the printer database, Nginx and Docker hosting, the Tauri desktop build, and model licensing all live in docs/DEVELOPMENT.md.

npm install && npm run dev      # http://localhost:5173

Troubleshooting

• Linux: blank window on launch (Wayland). If the app opens to an empty window and, when launched from a terminal, prints Could not create default EGL display: EGL_BAD_PARAMETER, start it with WEBKIT_DISABLE_DMABUF_RENDERER=1 set — for example WEBKIT_DISABLE_DMABUF_RENDERER=1 './Trim_<version>_amd64.AppImage', substituting the filename your build produced. WebKitGTK's DMABUF renderer fails to initialise EGL on some Wayland setups; this makes it fall back to a working path. The app sets the variable itself (unless you set it yourself), so this workaround should not be needed on this build — it is kept here for anyone running an older one.
• The dev server serves a stale module — for example "does not provide an export named …" for an export that plainly exists, with the app rendering only its "Instruments powering up…" fallback. That is a stale Vite transform, usually after a file was saved mid-edit. Restart with npm run dev -- --force.

Future ideas

AI-assisted photo evaluation (the storage schema already reserves an analysis field), photo comparison, multiple printers per filament (multiple nozzles per printer already ships — see Why this exists), per-nozzle data in the printer database so machines other than the X2D fill their nozzle list automatically, printer API integration, community profile sharing, filament inventory with drying/spool tracking.

License

Copyright (C) 2026 Aaron Taylor — original PerfectFit Copyright (C) 2026 espentruls — Trim (fork of PerfectFit)

Trim is free software: you can redistribute it and/or modify it under the terms of the GNU Affero General Public License, version 3 as published by the Free Software Foundation. The full text is in License.

This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY — without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Affero General Public License for more details.

Why AGPL-3.0: Trim is built around the Orca-family slicers, and OrcaSlicer, PrusaSlicer, and Slic3r are all AGPL-3.0. Matching that license keeps the project compatible with the ecosystem it depends on — particularly as future releases integrate more deeply with the slicers themselves — and guarantees the work stays open: anyone may use, modify, sell, or host Trim, but derivative works must remain open source under the same terms, including when offered over a network.

Upstream PerfectFit used a custom non-commercial license (R3D-NC v1.0) before its 1.3.1 release. That license was incompatible with AGPL-3.0 code and was retired there. This fork's own history begins at 2.0.0 and has been AGPL-3.0 throughout.