Overview

SQAI is the governed, read-only structured-data tool for AI agents — typed, policy-checked plans run on a SQAI runtime, every answer carrying replayable hashes.

SQAI (Structured Query AI) is the governed, read-only structured-data tool for AI agents. A natural-language or typed question becomes a typed, policy-checked plan, the plan runs on the SQAI runtime, and you get the answer plus full provenance. Query CSVs and databases — including Oracle — alongside 6,084 read-only compute capabilities, from stats.median to finance.npv to option_pricing.black_scholes_call. Every capability is side-effect-free, and every result carries hashes that replay it exactly.

Setup is one free device login. The query plane runs in-process — no separate service — and returns in sub-millisecond to low-millisecond time on typical files. Run sqai login once for the free sqai_developer key; the compute runtime then provisions once (the first-ever call downloads, extracts, and initializes the bundle in ~110s), then stays warm and resident: there is no cold start per query. Every later compute() is warm — finance.npv measured at ~0.9ms.

New to SQAI? Start with the Quickstart — connect a file and get a typed, hashed answer in minutes.

Start here

Quickstart

Install, connect a CSV, and get a typed, hashed answer.

Installation

npm, pnpm, yarn, bun, or pip. No separate service to run.

Connect your data

CSVs, SQLite, Excel/Parquet, and live databases — Postgres, Oracle, and more. Same API in Python.

Vercel AI SDK tool

Give any agent governed, read-only data and compute tools.

Concepts and governance

How it works

Question → typed plan → deterministic execution → provenance.

Compute & filtering

6,084 side-effect-free capabilities; bind arguments to columns.

Determinism & provenance

Replayable hashes; same input → identical hash.

Policy & governance

Allow-list sources, fields, and functions; read-only by construction.

Install

npm install @thyn-ai/sqai

pnpm add @thyn-ai/sqai

yarn add @thyn-ai/sqai

bun add @thyn-ai/sqai

pip install sqai

The query plane needs nothing else — no separate service. Run sqai login once, then the compute runtime auto-provisions on the first compute() and stays warm. For the AI SDK tool add @thyn-ai/sqai-ai-sdk; for the CLI add @thyn-ai/sqai-cli. See Installation.

A first look

Run a computation. The result is the answer plus everything needed to replay it — hashes, contract version, and the exact runtime it ran on.

import { createSQAI } from "@thyn-ai/sqai";

const sqai = createSQAI();

const result = await sqai.compute({
  module: "finance",
  function: "npv",
  args: [0.1, [-1000, 300, 420, 560, 680]],
});

console.log(result.value); // 505.020148896933

from sqai import create_sqai

sqai = create_sqai()

result = sqai.compute(
    module="finance",
    function="npv",
    args=[0.1, [-1000, 300, 420, 560, 680]],
)

print(result["value"])  # 505.020148896933

Both calls return the same result:

{
  "status": "ok",
  "value": 505.020148896933,
  "value_type": "float64",
  "module": "finance",
  "function": "npv",
  "invocation_hash": "b3ca3e925e4372a5f9365790a29c486afed868073d3fd301d03022f44689e423",
  "computation_hash": "b74f67d0d7a594aa7ac91f6291612452aa8ccdf603351ebc8d801a6fddd91bc8",
  "contract_hash": "sha256:31247fb219e656f349bb6d36fa65c2e2702e2688a2fddb4b16f1a9e09a02cdbb",
  "determinism": {
    "runtime_bundle_version": "0.1.0",
    "runtime_bundle_sha256": "4d64142e4c1ff63d299cfc8b172fdf97cb59169b545e1e02978e678a632ce6e1",
    "platform": "darwin-arm64",
    "architecture": "arm64",
    "kernel_build": "mojo",
    "precision_mode": "float64",
    "thread_count": 1,
    "input_hash": "aadddd29e6a8cc682c934233448531039e4f4777ad508d164734a0095c4e10f1"
  },
  "latency_ms": 0.93,
  "request_id": "sqai_eaab471b197880a4"
}

The value and both hashes are byte-identical in TypeScript and Python. Call it again and computation_hash is identical again — same input → same hash, forever. That is what makes an answer auditable and replayable. See Determinism & provenance.

What you can query

Data sources. In-process, with no key: CSV, TSV, JSON, in-memory row arrays, { records: [...] }, and SQLite files; Excel and Parquet in Python. Through a private SQAI deployment (SQAI_DEPLOYMENT_URL, with SQAI_API_KEY only if it requires auth) — a private SQAI deployment on your own infrastructure, no SQAI cloud: PostgreSQL, MySQL, Oracle (bundled Thin-mode driver), Microsoft SQL Server, Snowflake, BigQuery, ClickHouse, Amazon Redshift; S3 / GCS / Azure Blob; Redis / Neo4j / Elasticsearch; REST APIs; and Git repos. Access is governed, credentials stay on your SQAI deployment, and no data is copied. See Connect your data.

Compute. 6,084 capabilities in total, all read-only and side-effect-free; 5,790 exposed to the TypeScript, Python, and AI SDKs; 5,780 fully deterministic and 10 seed-required simulations. Breadth spans statistics, hypothesis testing, probability and distributions, time series, linear algebra, finance and risk, option pricing, ML and metrics, signal/DSP, information theory, number theory, calculus and root-finding, and optimization — plus an LLM-infra pack (flash_attention, kv_cache, moe_routing, rag_retrieval, embeddings) and 100+ applied-science and engineering domain packs. searchCapabilities returns a one-line summary for each. See Compute & filtering.

Bind a compute argument to a live column and the result carries binding provenance — the source, fields, schema_revision, row_count, and input_hash it read. That is how an agent's numbers stay traceable to your data.

NextInstallation

Last updated 2 months ago

Was this helpful?