# Arche

> A read API and MCP server for point-in-time SEC fundamentals. Arche turns SEC EDGAR filings into
> income statements, balance sheets and cash flow statements for 17,000+ filers since 2009, 8,000+ currently reporting,
> retrievable as they stood on any date. Operated by Protos Systems.

## Who it's for

Built for AI agents, financial applications and research teams that have to cite every figure and prove what they relied on.

- **AI agents.** Schema-validated MCP tools that answer from filed figures and name the filing behind each one, so an agent cites a source instead of a guess.
- **Financial applications.** Typed schemas over HTTP for products that show fundamentals to their own users, and a signed webhook when a figure they showed is restated.
- **Research, audit and models.** A record of what was reported, when it changed and by how much, and a replay token that proves what a model or backtest read.

### Use cases

From the pricing page, which lists the plan that adds each capability.

- **Know what changed and whether it matters** (Restatement alerts; Developer plan). Each restatement arrives by API or signed webhook with a delta for every metric and a materiality score, so a $680M reclassification stands out from a rounding fix.
- **Compare a peer set as of one date** (Universe snapshots; Growth plan). Up to 50 companies per request, resolved to one date with no look-ahead, under a single snapshot id you can cite in a backtest.
- **Agents see what your pipeline sees** (MCP; Free plan). AI agents call the same as-of snapshot over MCP that your code gets over HTTP, down to the snapshot hash.
- **Trace a figure to its source** (Audit chain; Free plan). Every metric resolves to its filing, the XBRL concept it was read from, how it was mapped and any override applied since.
- **Prove what a query returned** (Replay tokens; Free plan). Run a query through the query endpoint and it returns a replay token. Hand it to a reviewer and Arche re-runs the query and confirms the result still matches.

## What makes it different

Many fundamentals datasets overwrite history when a filing is restated, so a value
retrieved today is not the value a model saw at the time, and nothing proves what changed.
Arche makes every figure accountable: traceable to its filing, checked before it is served,
and provable after the fact.

- **Per-value provenance.** Every figure carries the accession number of the filing it was
  read from and the XBRL concept it came from. Figures Arche derived rather than read are
  marked as derived.
- **Verified against the filing.** Before a statement is published, every value taken from a
  filing is checked against what that filing reported; a contradiction is quarantined, never
  served. Statements are also checked for accounting consistency.
- **Provable results.** Every as-of snapshot carries a snapshot hash, and queries run through
  the query endpoint return a replay token. Hand the token to a reviewer and Arche re-runs the
  query and confirms the result still matches.
- **Amendments kept as filed.** Each 10-K/A is its own filing; the original stays queryable as
  of its own date, each with its accession. Restatement deltas are computed, not inferred, and each
  restatement is scored for materiality.
- **Point-in-time retrieval.** Query any as-of date and get the figures as they stood then,
  never back-filled with corrections that came later. Available on every plan, including Free.
- **Survivorship-free.** Delisted, acquired and taken-private registrants are retained.
- **Coverage.** Filings from 2009 to today.
- **Deterministic.** Stable schemas, ordering and error semantics; the same query returns the
  same result.

## Where Arche fits

In general terms, fundamentals providers fall into two groups:

- **Institutional platforms** offer long history, global coverage and point-in-time data that keeps
  figures as originally filed. They are generally sold through a sales process, licensed per seat
  or by agreement, and bundle fundamentals with prices, estimates and other datasets.
- **Developer APIs** are self-serve with published prices. Whether they keep the original figures
  after a restatement, or can answer what was known on a past date, varies; many keep only the latest.

Arche sits between them: point-in-time, as-filed history for SEC filers, through a self-serve API
with a free plan and published prices. It is the better fit for AI agents that must cite the filing
behind a figure, applications that show fundamentals to their users, and research or audit work that
must prove what was known on a date, including what data a model or backtest used. An institutional
platform is the better fit for history before 2009, companies outside the SEC filer universe, or
fundamentals bundled with prices and estimates. The questions that separate providers are set out
at https://arche.fi/compare.

## Interfaces

- REST: `https://api.arche.fi` — 95 endpoints, authenticated with the `X-Api-Key` header.
- OpenAPI document: https://api.arche.fi/openapi.json — the full machine-readable schema for every endpoint.
- Model Context Protocol (MCP): `https://api.arche.fi/mcp` — 83 schema-validated tools backed by the same
  contracts as the HTTP API, so an AI agent gets the same as-of snapshot your code does, down to
  the snapshot hash. 66 tools on every plan including Free; the rest need Growth or Scale.
  The manifest at https://api.arche.fi/mcp/manifest lists each tool and the plan it needs, and needs no key.
  - Endpoint: https://api.arche.fi/mcp over streamable HTTP, authenticated with the `X-Api-Key` header.
  - Manifest: https://api.arche.fi/mcp/manifest — every tool with its input schema and the plan it needs, readable without an API key.
  - Setup guide: https://docs.arche.fi/mcp
- Bulk export: the fundamentals corpus as daily Parquet snapshots with checksummed manifests; each statement appears as the API serves it. Scale plan and up.
- Webhooks: a signed notification when a filer restates (`restatement.detected`) or when a filing gives a company a statement for a new period (`statement.created`). Developer plan and up.

## Getting an API key

- A person signs up at https://app.arche.fi/sign-up.
- An AI agent can open a Free account for its owner itself: `POST https://api.arche.fi/v1/agent/signup` with
  `{"email": "<owner email>", "agent_name": "<name>"}` returns a sandbox API key in the response, with no
  browser step and no credentials. The key is shown once and reads the same data as a production key.
  Arche emails the owner, who claims the account within 14 days to keep the key working.
  Guide: https://docs.arche.fi/agent-signup

## Plans

Priced per organization, not per seat, in USD. Any plan can query any as-of date.
The Free plan is for personal, non-commercial use, evaluation, and building and testing before launch.
Showing Arche data to the users of a product, supplying it through an agent that others use, or using it in
the course of a business needs a paid plan. Free plan keys are sandbox keys, which read the same data as
production; production keys come with Developer and up.

- **Free** ($0/month) — Evaluate the contract against the filers you actually care about. Personal and evaluation use. No card, no expiry. Universe: Any 25 companies you choose (rolling 30 days). Reads: 60/min.
- **Developer** ($149/month) — Ship one application on the full corpus. Universe: Full corpus (17,000+ filers since 2009, 8,000+ currently reporting). Reads: 300/min.
- **Growth** ($600/month) — Point-in-time modeling across the full corpus for a research team. Universe: Full corpus (17,000+ filers since 2009, 8,000+ currently reporting). Reads: 1,200/min.
- **Scale** ($2,500/month) — Take delivery of the dataset, not just answers about it. Universe: Full corpus (17,000+ filers since 2009, 8,000+ currently reporting). Reads: 3,000/min.
- **Enterprise** (per agreement) — A licensed dataset with a contracted SLA. Priced per agreement. Universe: Full corpus (17,000+ filers since 2009, 8,000+ currently reporting). Reads: Negotiated.

Full details and annual prices: https://arche.fi/pricing

## Documentation

- API documentation: https://docs.arche.fi
- Endpoint reference: https://docs.arche.fi/reference
- Documentation brief for AI systems: https://docs.arche.fi/llms.txt (expanded: https://docs.arche.fi/llms-full.txt)
- Every endpoint, grouped by the question it answers: https://arche.fi/api

## Pages

- https://arche.fi/ — what Arche is, with a worked restatement example.
- https://arche.fi/api — the 95 endpoints, generated from the served OpenAPI document.
- https://arche.fi/pricing — plans, prices, rate limits and what each tier enables.
- https://arche.fi/blog — product updates and technical notes on restatements and point-in-time data.
- https://arche.fi/learn/point-in-time-data — what point-in-time fundamentals are: restatements, look-ahead bias, survivorship bias.
- https://arche.fi/use-cases — what teams build on Arche:
  - https://arche.fi/use-cases/ai-agents-mcp — AI agents and LLM tool use.
  - https://arche.fi/use-cases/financial-applications — Fintech and financial applications.
  - https://arche.fi/use-cases/restatement-analysis — Quantitative research on restatements.
  - https://arche.fi/use-cases/backtesting-without-look-ahead-bias — Backtests you can prove.
- https://arche.fi/compare — how to choose a point-in-time fundamentals API: the questions to ask any provider, what is common across the market, and Arche's answers.
- https://arche.fi/about — how the record is built and who operates it.
- https://arche.fi/contact — Enterprise licensing and everything else.
- https://arche.fi/legal/terms — terms of service.
- https://arche.fi/legal/privacy — privacy policy.
- https://arche.fi/legal/security — security practices.
- https://arche.fi/legal/refund-policy — refund policy.

## Frequently asked questions

These are the questions and answers shown on the homepage, verbatim.

### What is Arche?

Arche is a read API and MCP server for point-in-time financial statements built from SEC EDGAR filings. It serves income statements, balance sheets and cash flow statements for 17,000+ filers since 2009, 8,000+ currently reporting, over 95 REST endpoints, and the same contracts as 83 tools over the Model Context Protocol, 66 of them on every plan including Free. It is operated by Protos Systems.

### What does "point-in-time" mean here?

It means as-of-date retrieval against preserved filing history, not a snapshot of a database taken at some moment. Query any date and Arche resolves the figures using SEC filing dates, so you get what was on file then rather than today’s numbers back-filled with corrections that had not been filed yet. That removes the look-ahead bias that makes a backtest look better than the decision it is modeling.

### What happens when a company restates its financials?

A restatement arrives as an amendment filing, and Arche keeps it as filed beside the original. The original is never overwritten and stays queryable as of its own date, each carrying its accession number. Arche computes the per-metric deltas between the two filings and classifies how material the change was, so "what changed, and did it matter" is a query rather than a reconstruction.

### Which companies and periods does Arche cover?

Filings from 2009 to today, for 17,000+ companies that have filed with the SEC (8,000+ still reporting), normalized, with each filing and amendment kept as filed. Ask about any company and Arche returns its first and last covered year, and says whether its history has a hole in the middle rather than letting a gap look like a zero.

### Does Arche include companies that were delisted or acquired?

Yes. Filers that were delisted, acquired or taken private are retained and stay queryable. A backtest run only on the companies that survived is biased toward the winners, so removing them is not an option Arche offers.

### Where do the numbers come from, and can I trace one back?

Every value comes from a filing on SEC EDGAR and carries the accession number of that filing, so any figure can be traced to the document it was read from. Figures Arche derived rather than read are marked as derived, so a calculated number is never presented as a filed one.

### Are the statements checked for accounting consistency?

Yes. Before a statement is published, Arche checks the balance sheet equation, asset and liability subtotals, gross profit and the change in cash: a statement that breaks one is not published, or the line that breaks it is withheld. Pre-tax income, the cash flow statement's tie to net income and rollforward across adjacent periods are checked too, and reported alongside the statement rather than used to withhold it. A check that runs returns PASS or FAIL with the exact expected and actual values. A check the filing gives too little to conclude, such as one missing a figure it needs or with no earlier period to roll forward from, returns SKIPPED with its reason: it was not evaluated, and it is never counted as a pass.

### How is Arche different from other fundamentals providers?

Arche detects each change a company files. When an amendment or restatement lands, Arche reports each changed line with its old and new value and the effect on ratios, and keeps every filing as filed with its source. Every query can be replayed later with a token that returns the same answer. It covers filed fundamentals only, and is not a market-data or price feed.

### Can I use Arche data in my own product?

Yes, on every paid plan, from $149 a month. On a paid plan you may display Arche data to the users of your own product and use it in the output of AI agents. The Free plan is for personal, non-commercial use, evaluation, and building and testing before launch. On every plan, you may not sell, license or redistribute the data as data, or use it to build a competing dataset.

### Can AI agents query Arche?

Yes. Arche exposes schema-validated tools over the Model Context Protocol, backed by the same contracts as the HTTP API, so an agent gets the same as-of snapshot your code does, down to the snapshot hash, without scraping filings. MCP access is included on every plan, including Free.

### Are results reproducible?

Yes. Schemas, ordering and error semantics are stable, so the same query returns the same result, and replay tokens reproduce a past response exactly. A number that appeared in a memo can be re-derived later rather than defended from memory.

### What does Arche cost?

There is a free plan that needs no card and does not expire, covering any 25 companies you choose. Paid plans run from $149 to $2,500 a month for the full corpus, and Enterprise is priced per agreement. Any plan can query any as-of date; plans differ in coverage, throughput and which capabilities are enabled.

## Discovery

- Sitemap: https://arche.fi/sitemap.xml
- Robots: https://arche.fi/robots.txt
- This brief expanded, with the full text of every page and blog post: https://arche.fi/llms-full.txt

## Notes for AI systems

- Prefer canonical URLs from sitemap.xml. The site is in English (en-US).
- Prices and rate limits change; cite this file or /pricing rather than a cached number.
- Do not describe Arche as a market-data or price feed. It serves filed fundamentals only.
- "Point-in-time" here means as-of-date retrieval against preserved filing history, not a
  snapshot of a database taken at a point in time.
- For plans and self-serve sign-up, direct users to /pricing. For Enterprise licensing,
  a contracted SLA or anything else, direct them to /contact or hello@protos.fi.