Open-source and low-cost autosamplers: what the literature shows

The published open-source autosamplers are very good at one thing: moving a needle, syringe, tube or pipette between containers cheaply, with reported hardware costs from about US$150 to a few hundred dollars for builds based on 3D-printer parts. Almost none of them inject into a pressurised HPLC flow path. Of the 14 papers reviewed below, the only one that does (Axtell et al., 2026) buys its six-port valve from VICI and estimates its total system cost at below US$20,000. If you need routine HPLC injections, the hard parts are the valve, volume precision, carryover and validation, not the XYZ motion.

Key points

• Most open designs reuse 3D-printer or CNC motion (stepper motors, lead screws, Marlin firmware) and are controlled by scripts: AutoIt, Arduino, Raspberry Pi or Python.
• They were built for formats that commercial autosamplers do not serve well: NMR tubes, solid capsules for elemental analysis, 40 mL field vials, bioreactor probes, 470 mL fraction bottles.
• Only one of the 14 is reported to inject into a pressurised LC flow path, and it uses a commercial valve. The others inject at low pressure (CE, GC syringe) or only collect samples.
• Reported precision ranges from under 1% error (microsyringe) to 25.7% volume RSD (peristaltic bioreactor sampler). Few papers report carryover.
• Build for unusual formats, teaching and research prototypes; buy for routine quantitative HPLC, regulated work and anything that must outlast the person who built it.

How the papers were chosen and checked

We looked for peer-reviewed papers (searched October 2026) that describe hardware for moving samples into or out of an analytical instrument, with either open design files or an explicit low-cost focus. Each reference below was checked on its DOI record and on the abstract or open full text (PubMed Central where available). Costs are the authors' own figures, in their currency and year, not adjusted for inflation. Where a cost or performance figure was not in the text we could open, the table says so.

This is not a systematic review. HardwareX dominates because it is a journal dedicated to open scientific hardware, so it is where most open autosamplers end up.

Published open-source and low-cost autosamplers compared

What the open designs have in common

Motion comes from 3D printers and CNC kits

Osmar was assembled from two G-code machines. Iara runs stepper motors on lead screws with parts listed from online marketplaces. Crandall et al. turn a hobby printer into a fraction collector and note that a capable printer could be bought for as little as US$100-200 in 2024. Positioning is rarely the weak point: in Murray et al. the 3-axis gantry repeated to within ±0.007 in (about ±0.18 mm), while the printed SCARA arm managed ±0.026 in (about ±0.66 mm) on one axis. A 2 mL vial septum is several millimetres across, so either is good enough to hit a vial; the gantry leaves more margin for smaller targets.

Control is a script, not a chromatography data system

Nearly every design here is driven by a script: AutoIt on Windows (Osmar, the solids sampler), an Arduino with a keypad (the CE sampler), a Raspberry Pi (BioSamplr, O'Brien's flow rig) or Python (OpenLC in Axtell et al.). That makes them easy to put into a self-driving lab loop and hard to put into a validated CDS sequence. If you are writing that glue code yourself, our guide to controlling serial lab instruments from Python covers the practical side.

They exist because the sample format is unusual

Look at what was automated: 5 mm NMR tubes, solid samples for elemental analysis, 40 mL field vials, bioreactor probes, 470 mL bottles. Commercial HPLC autosamplers already handle 2 mL vials and well plates. The open-hardware literature mostly fills gaps that the market ignores, rather than competing with standard HPLC autosamplers.

Why so few inject at HPLC pressure

An HPLC injection valve has to switch a sample loop into a flow path running at hundreds of bar without leaking, thousands of times. As a rule of thumb, that needs a lapped stator and a polymer rotor seal made to tight tolerances, which a 3D printer cannot make. Our article on the 6-port injection valve explains the rotor seal and its pressure limits.

So the published designs avoid the problem. The CE sampler uses hydrodynamic injection, where a small pressure difference pushes sample into the capillary. Osmar injects with a microsyringe. RotoMate moves NMR tubes with compressed air. BioSamplr and Iara only collect samples for later analysis.

The one exception keeps the pressure boundary commercial. Axtell et al. mount VICI valves and actuators on an OT-2, wire the actuator control pins to an ESP32 and coordinate everything with their OpenLC Python package. Their loop is a 13 cm length of 100 µm i.d. fused silica. You can check its volume with the cylinder formula:

V_loop = π × (d/2)² × L

Worked example: d = 0.100 mm and L = 130 mm, so V = 3.1416 × (0.050 mm)² × 130 mm = 1.02 mm³, which is 1.0 µL. The authors injected by aspirating either 500 nL of sample (partial fill, 0.5 times the loop) or 1700 nL (overfill, 1.7 times the loop). Partial fill saves sample but depends on the syringe; overfill relies on the loop volume for reproducibility. The same trade-off applies to any loop size, as explained in full-loop vs partial-loop injection.

Where DIY autosamplers struggle

Volume precision

The spread in the literature is wide. Osmar reports movement errors under 1% in repeated injections, the CE sampler under 2.7% RSD for peak areas, and the OpenSampler's nanolitre dispensing had CVs below 7% for volumes of 20 nL or more, with absolute errors below 15%. At the other end, BioSamplr's peristaltic sampling gave an overall volume RSD of 25.7% at a 0.75 mL target, and its authors recommend targets above 1.0 mL. As a rule of thumb, quantitative HPLC expects peak-area RSDs well below 1% for replicate injections of a stable standard, so a DIY injector needs its own precision study before its numbers go into a report.

Carryover

Carryover is rarely reported in these papers; the nanoLC OpenSampler paper, for example, does not report it. Iara is an honest exception: after three rinses following the most concentrated glucose solution, the next blank of ultrapure water clearly contained more dissolved organic carbon than the other blanks, although the difference across all blanks was not statistically significant. Run a blank after your highest standard on any sampler, built or bought; our article on autosampler carryover gives the test and the fixes.

Synchronisation with the chromatograph

A sampler must tell the pump and detector when it has injected, or be told when to start. Crandall et al. chose time-based fraction collection and note that using the HPLC's 5 V signal "greatly increases build complexity" through extra wiring and Arduino programming; they provide synchronisation only as an option. Expect the same work for an injector: contact closures, start signals and timing jitter have to be designed and tested.

Validation, documentation and upkeep

None of the designs reviewed here comes with installation and operational qualification documents, an audit trail or a driver for a commercial CDS. In a GMP or GLP lab you would have to write and run the qualification yourself (see autosampler qualification tests). Maintenance is the quieter problem: parts lists that point to marketplace listings go out of date, printed parts in the solvent splash zone need a material that tolerates your solvents, and the person who understands the firmware eventually leaves.

Robots, Opentrons and Jubilee: preparation more than injection

General-purpose liquid handlers appear in the literature mainly as sample preparation tools. Kverneland et al. use an OT-2 to digest proteins and load up to 192 samples onto Evotips in 6 h; an Evosep One LC then runs the samples from the tips. Li et al. list earlier DIY liquid handlers at under US$450 to about US$1,000 and put a basic Opentrons setup at about US$5,000 (2022 figures). The Jubilee platform has syringe and pipette heads, and the science-jubilee documentation reports that an OT-2 pipette mounted on Jubilee reproduced Opentrons' published accuracy and precision after gravimetric calibration. For robot arms, the 3D-printed digital pipette transferred liquid within 0.5% error. A 2025 review in Digital Discovery estimates that converted consumer 3D printers cut equipment cost by 90-99% compared with commercial alternatives (Doloi et al.).

These tools prepare and stage samples well. The injection itself still happens in a commercial instrument, which is how most self-driving labs handle sampling and injection.

When to build and when to buy

There is a middle path: commercial hardware with an open control interface. Elutrolab, which publishes this blog, is developing the AS-48, an engineering prototype that is driven over USB as a virtual COM port with a documented plain-text command set, so it can be scripted from Python or LabVIEW in the same way as the DIY builds above. Its 6-port valve has a 5000 psi (345 bar) design rating* (values marked * are design targets). It is not open-source hardware, and it has no CDS drivers; see the specifications.

How to judge a published design before you build it

1. Check the licence for hardware and software (RotoMate, for example, uses CC BY 4.0 for hardware and GPL v3 for software).
2. Read the bill of materials with its date. Prices and part numbers from marketplace listings change within a year or two.
3. Look for precision and carryover data with numbers, measured on the analytical instrument, not only positioning accuracy.
4. Confirm the wetted materials suit your solvents, and that printed parts sit outside the splash zone.
5. Check that the code repository still exists and builds.
6. Plan safety: needles moving near hands, and for CE, a sampler that controls high voltage needs interlocks.
7. Count your own hours. The hardware cost in a paper rarely includes assembly, debugging and the precision study.