Selecting first-line antipsychotic medication for first episode of psychosis patients is a very challenging task requiring the clinicians to empirically weight multiple criteria. Precision treatment rules developed using health records offer a pragmatic approach to support clinicians' treatment selection, however, they don't incorporate side effects and patient preferences. We used Electronic Health Records from Early Intervention for Psychosis services in South London and the Maudsley NHS Trust and followed the RECORD and TRIPOD + AI guidelines. Precision treatment rules were developed using causal machine learning methods and estimated effectiveness (change of medication, hospitalisation) and side effects (extrapyramidal side effects, hyperprolactinemia, sedation, sexual side effects, and weight gain) using clinical, demographic, symptom and substance use predictors. Patient preferences regarding side effects were incorporated by ranking method. 1709 patients (mean age 26.7 years and 64% male) were included. Aripiprazole was recommended to between 80 and 98% of patients depending on selected patients' preferences. Compared to the observed treatment decisions we estimated that under treatment rules recommendations hyperprolactinemia would be reduced by 4.7 percentage points (pp), sedation by 15.8 pp, sexual side effects by 4.3 pp and weight gain by 15.2 pp with no change in hospitalisation and change of medications outcomes. However, extrapyramidal side effects were estimated to increase by 5.5 pp. This study presents the first precision treatment rules for early psychosis that integrate effectiveness, side effects and patient preferences. Further research using larger data sets, more predictors and treatment options is suggested.