HumaiPsy

A study interface for Human–AI interaction.

Two agents, two conditions, any number of counterbalanced trials. Every trial is timed, recorded, and written out in a form a statistics package can read.

Download HumaiPsy Source on GitHub

Windows · Linux · macOS

Design

The 2 × 2 design, and one participant’s twelve-trial sequence Left: a two-by-two table of agent by condition, with AGENT A and AGENT B as rows and COND X and COND Y as columns; COND X cells are filled solid and COND Y cells are hatched. Right: twelve trials in order, each filled by its condition, with trial 04 marked as the trial in progress. COND X COND Y AGENT A AGENT B A · X B · Y A · Y B · X A · X A · Y B · X B · Y A · X B · X A · Y B · Y 01 02 03 04 05 06 07 08 09 10 11 12 04 / 12 The 2 × 2 design, and one participant’s twelve-trial sequence Top: a two-by-two table of agent by condition, with AGENT A and AGENT B as rows and COND X and COND Y as columns; COND X cells are filled solid and COND Y cells are hatched. Bottom: twelve trials in order, each filled by its condition, with trial 04 marked as the trial in progress. COND X COND Y AGENT A AGENT B 04 / 12 01 02 03 04 05 06 07 08 09 10 11 12
Fig. 1 One participant’s run, shown at twelve trials. Each agent–condition pair appears equally often, in a randomised order. Solid cells are COND X, hatched cells are COND Y.

Materials

Hosts
Windows, Linux and macOS on Apple silicon. One native binary: no separate backend process, no bundled runtime.
Model
Ollama, either on this machine or on a host you choose. No account, no telemetry.
Storage
Stays on this machine. Master-password vault, loopback-only dashboard, air-gap option.
Licence
AGPL-3.0. Source and SHA-256 checksums with every release.

Procedure

  1. Set the design

    Agents, conditions, repetitions and the prompt set, defined once and reused across participants.

  2. Run the trials

    Participant ID, a countdown per trial, and timings captured per turn without being asked for.

  3. Close the run

    The event ledger is sealed with a summary and SHA-256 hashes, so a finished run cannot be edited quietly.

Output

exp_001.csv
One row per trial. Opens straight in R, Python, SPSS or JASP.
runs/<id>/events.jsonl
Append-only event ledger, hashed at close.
chat.sqlite
Turn-by-turn transcript of every trial, kept locally.

Researchers

Cognitive science, HCI, experimental psychology

Within-subject designs, an IRB-facing security note, and output that drops into the analysis pipeline you already have.

AI and ML research

Model-behaviour comparison, time to first token, and the same design run air-gapped on local models.