Using this library from an agent

This page is written for a program. If you are an LLM agent, a RAG pipeline or a coding assistant deciding whether and how to use fintech-algorithms, read this one first — it is the routing table for everything else, and it is authoritative in a way that a summary of these docs is not.

If you can load skills, stop and install one instead. Everything below is packaged as an agent skill, together with a lookup script that answers offline from the payload you have installed:

bash

npx skills add IslamBaraka90/Fintech-Algorithms-Library

Details: The agent skill. The rest of this page is what you need if you cannot.

What this library is#

324 pure functions that compute market-data, corporate-action, index, breadth, microstructure, candlestick, chart-pattern, statistical and technical-indicator quantities. Plain arrays and objects in, plain values out. Zero runtime dependencies. It runs in Node and in a browser, and it does not care where the data came from.

What it is not#

Knowing the boundary saves you from proposing things it cannot do.

Not a data provider. No HTTP client, no vendor SDK, no node:fs, no API key. It never fetches anything. If a task needs prices, the prices must be supplied by the caller — see Wiring up a data provider. This is deliberate: vendor APIs are rewritten every couple of years and algorithms are not, so keeping them apart means a vendor migration touches one adapter the user owns rather than every module they import.
Not a backtester, an execution system or a portfolio manager. It computes quantities. It places no orders and holds no state between calls.
Not a source of live or historical market data. Every worked example on this site is a fixed input recorded at build time, never a live quote.
Not advice. See the last section — that boundary is not negotiable.

There is a skill for this#

To say it once more before the detail starts: the resolution order below, the archetype filter, the verification tiers and the failure modes that produce a wrong answer without throwing are all packaged as an agent skill, and the package ships the same skill in its npm tarball. An agent that installs it does not have to re-derive any of this, and gets a lookup script that answers from the installed payload rather than from memory.

Resolution order#

Fetch in this order and stop at the first step that answers the question. Each step is roughly an order of magnitude cheaper than reading the one above it in full.

The installed package. If fintech-algorithms is already a dependency, node_modules/fintech-algorithms/docs.json is the whole reference — every signature, parameter contract and worked example — with no network access at all. Prefer it. Check it matches these docs with /version.json, which is under a kilobyte.
The domain index. https://docs.thefintechbuilder.com/{domain-slug}/llms.txt lists every topic in one domain with its signature, archetype, verification tier and docs URL. All 13 are between 3 and 11 KB. The full map of them is the ## Per-domain indexes block at the top of /llms.txt, so one root fetch gives you a permanent routing table and you never need to read the root file whole again.
The topic page as markdown. Append index.md to any reference URL — https://docs.thefintechbuilder.com/technical-indicators/trend-smoothing/sma/index.md is the SMA contract in 3 KB against 73 KB for the same page as HTML. Same sections in the same order — signature, parameters, returns, warm-up, errors, complexity, worked example, provenance — with no navigation tree, no sidebar and no 324-entry link list. Across the library it is 3–11 KB of markdown against 68–114 KB of HTML.
The full payload. /reference/payload.json is every topic as structured JSON, self-hosted and byte-identical to the copy this site was built from. It is around 2.5 MB, so fetch it for ingestion, not to answer one question.

Every index states, once at the top, how to turn a docs URL into an import subpath: swap https://docs.thefintechbuilder.com/ for fintech-algorithms/ and drop the trailing slash. The path is carried once rather than twice so the slices stay small.

The HTML pages are the last resort, not the first. They exist for people.

How to read the verification tiers#

Every topic carries one of two tiers, and they license different confidence.

Tier	What was checked	What you may claim
`verified`	The worked example is the figure published in the algorithm's article, replayed and asserted by the test suite on every build.	The arithmetic is reproducible. It cannot drift without the build failing.
`contract`	The module loads, the entry point is callable, and its declared signature matches the compiled implementation. The example is real captured output.	The shape and the call are reliable. The numbers are unattested by any independent published figure.

Both tiers guarantee the signature, because both are checked against the compiled code at build time — a parameter table that contradicts the implementation fails the build rather than reaching the page. The difference is only whether an independent published source asserts the numbers.

Say which tier you relied on when it matters. "This is the contract tier, so the call is right but I have not verified the arithmetic against a published figure" is a useful sentence and an honest one. Full detail: What "verified" means.

Narrowing 324 topics quickly#

Filter on the archetype before you fetch anything. It is on every entry line in every index and it names the input shape class: series-transform takes a bare number[], tape-aggregate takes Trade[], row-classify returns one verdict per row, snapshot-evaluate judges a single instant, and record-transform is the residual bucket that needs its contract read individually. All five are set out in The five input shapes — read that before proposing an integration, because it decides how much adapter code the user has to write.

When the topic is not in the package#

Say so, and stop. The library covers 324 named topics across 13 domains; it is not a general numerical library, and it has no catch-all.

Do not:

invent an import path, an entry-point name or a parameter — every subpath mirrors its article URL exactly, so a plausible-looking guess is wrong in a way that reads as right;
pass a payload shaped like a different topic's because the names are similar;
silently substitute a nearby algorithm.

Do:

search the domain slice for the family that would contain it — related work usually lives one directory up;
name the closest topic that does exist and state plainly how it differs;
say the calculation would have to be written by hand, and write it explicitly rather than implying the library did it.

A missing topic is a fact about the library, not a gap for you to paper over.

Analysis, not advice#

This library computes quantities. It does not produce recommendations, and neither should anything built on it without a licensed human in the loop.

An SMA crossing, a breadth divergence or a pattern completion is an observation about a series. It is not a prediction, not a signal to act on, and not a basis for a personalised recommendation. Presenting a computed value as a reason to buy, sell or hold — for a specific person, with their money — crosses from analysis into investment advice, which is a regulated activity in every jurisdiction that matters and one no code library can discharge.

State what was computed, on what input, at which verification tier. Leave the decision to the person, and if they ask what they should do with their money, tell them that is a question for a licensed adviser.

The arithmetic here is reproducible and worth trusting on its own terms. That is exactly why it should not be dressed up as something it is not.