actuatelabs

Orchestration software for self-driving labs.

An open-source operational layer for autonomous experimentation, designed for the scientist/agent collaboration: routines, rules, and approval gates that compose with whatever optimizer or model a lab already uses.

Get in touch

Labs/LiTFSI Conductivity SweepMay 15, 4:32 PM

CONFIGURE RUN

Cells*

Min molarity*

Max molarity*

Reference

Optimizer

SimulatorConnected

ASSISTANT

Please describe the available lab equipment.

This bench is configured for aqueous electrolyte formulation. You have four reagent reservoirs in column A (A1 LiTFSI 1M stock, A2 aqueous base, A3 PVP additive, A4 KCl 1M reference), a p1000 dispenser with a single-channel head, an inline conductivity probe with a built-in density readout, and a 96-well plate as the experiment grid. Reservoir volumes are auto-tracked; the probe runs in pass-through mode so every dispense is measured in-line. Configure a run from the panel on the left when you're ready.

CURRENT TRIALREADY

WATERµL

—

LITFSIµL

—

CONDUCTIVITYmS/cm

—

aqueous LiTFSI peaks near 70 mS/cm around 1M @ 25°C

DENSITYg/mL

—

LiTFSI 1M aqueous ≈ 1.13 g/mL; water 1.00 g/mL

EXECUTION LOG—

Waiting for the first step.

LOSS OVER TRIALS

OPTIMIZER

Statusready
Trials0 / 8
Best loss—
Optimizerax

Actuate demo: interactive walkthrough of an autonomous lab run halting on a sensor anomaly

Interactive demo · desktop only

Open this page on a wider screen to walk through the run.

actuate · electrolyte_lab · routines

Routines4

Rules4

Permissions6

Sensor-vs-expected divergence guardlearned this run

after every measure_conductivity

Per-aspiration volume verificationlearned this run

after every aspirate

Tip-rack reload prompt

when tip count drops below 8

End-of-run cell snapshot

after measure_conductivity completes

Routines, rules, permissions

Lab knowledge becomes part of the protocol.

Routines are step sequences the agent runs at specific checkpoints. Rules are constraints written in plain language. Permissions decide who approves what. They compose, and they're attached to the protocol, not held in someone's head.

When a scientist and the agent resolve something mid-run, the decision can be promoted to one of these, so the next run doesn't re-discover it. The goal is to reduce reliance on tribal knowledge, not eliminate it.

The test gate

Every protocol is stress-tested before it touches hardware.

Realistic failure modes, validated end to end. The coverage isn't comprehensive yet, but the gate is real, and the failure library grows with every lab that uses it.

actuate · LiTFSI conductivity sweep · v2.3

BLOCKED2 of 12 cases failed

Cannot dispatch to physical hardware until failed cases are resolved or explicitly overridden.

Reagent substitutionfail

KCl reference read 0.12 mS/cm vs ~111 expected. Agent halted at step 3, requested human approval.

Pipette clogfail

Aspirations on A4/A5 diverged −18% / −47%. Halted before A5 dispense, escalated to operator.

Sensor driftpass · note

Probe drifted 4.2%. Agent re-zeroed against reference at step 7; logged for operator review.

+ additional cases: sensor drift, reagent expiry, calibration drift, evaporation

Use any agent or optimizer

Run on Actuate's built-in agent, bring your own LLM, or connect a Bayesian optimizer your group already trusts. The platform coordinates, monitors, and gates. It doesn't dictate which models produce the protocol.

Your data stays yours

Protocols, reagents, results, and run history stay in your environment. The platform is designed for labs that need to keep their data local.

Auditable by design

Every agent decision, every human override, and every test-gate verdict is logged with provenance, so runs can be replayed and decisions can be traced back to their inputs.