Developers

1. Get a key

Sign in, open Settings → API keys, and create a key. Keys are scoped to your workspace and draw on its credits — buy credits from the workspace menu, or contact us about a pilot. Each key is shown once; the API stores only its hash.

2. Install

pip install archeai-sdk

Python 3.9+. pandas is optional and turns results into DataFrames.

3. Three lines

from arche_ai import ArcheClient

client = ArcheClient("arche_sk_...")          # Settings → API keys
hist = client.behavioral_history("AAPL")      # one row per executive per quarter

Each row carries seven communication-style dimensions and Big Five estimates for that executive in that quarter, computed only from that quarter's calls — so the history is append-only and safe for point-in-time research.

4. Event study against your own returns

from arche_ai import event_study as es

feat = es.point_in_time_features(hist)      # self-baseline deviations from prior quarters only
df   = es.join_events(my_events, feat)      # your events: ticker, call_date, your returns + baselines
res  = es.horserace(df, target="car_1", baseline_cols=["mom_21", "sue"],
                    train_quarters=train_q, test_quarters=test_q)
print(res["baseline"], res["baseline_plus_archeai"], res["lift"], res["baseline_definition"])

The kit reproduces the published construction: self-baseline deviations from the executive's own prior quarters, a within-quarter rank IC, and a baseline-vs-baseline-plus-ArcheAI horserace. You bring the returns and the baseline columns; the result names the baseline it was measured against.

What you are working with

Corpus

560,000+ unique corporate speech events (one copy per event) from 19,000+ companies, 2002–present

Executives scored

150,000+ named executives, 1 million+ executive-quarter profiles, one row per executive per quarter

Fraud-language risk

a quarterly series for every covered company, by ticker, with within-quarter and all-history percentiles

Validation

every model tested on data it never saw; model cards and test results under NDA

Every result

names the artifact that produced it (artifact_version + sha256, verifiable against the published manifest)

IP

US Patent 12,763,096 B2 (issued Oct 13, 2026; 19 claims) + continuation application 19/768,904 (filed Aug 7, 2026, pending) + recorded assignment to Arche AI, LLC

Links

• PyPI — archeai-sdk — Install, release notes, and the rendered SDK docs.
• API reference — OpenAPI for every route, including result schemas and error codes.
• Quant demo notebook — examples/quant_demo.ipynb in the package: install → key → 50 tickers → event study, with a synthetic-returns template.
• Quick start — session analysis — examples/session_demo.ipynb: one multi-speaker recording → diarized transcript → per-speaker style + personality → familiarity between the two main voices → one-screen summary.
• Quick start — executive monitoring — examples/exec_monitoring.ipynb: behavioral_history() → each executive's departure from their own prior 8 quarters → an "unusual quarters" read, designed to pair with the fraud-language line.
• FAQ — What the models measure, how accuracy is reported, how data is handled.