SEC data your code and agents can trust.
Point-in-time, survivorship-free SEC EDGAR fundamentals — every number traces to a filing and an acceptance timestamp. Pipe it into an agent, a backtest, or a cited thesis.
- Point-in-time, no look-ahead
- Every number cited
- Zero-retention BYO-LLM
Running flagship equity_research_brief SOP…
A plain-English question → typed tool calls → a cited answer.
Two datasets. One point-in-time record of US public markets.
We run the data and AI-infra layer beneath the answer: raw XBRL pulled from EDGAR, normalized into canonical concepts, kept point-in-time and survivorship-free — so every model you point at it inherits the same disciplined record.
Financial fundamentals
111M+ facts · 19,000+ companies · 1993–present
Income statement, balance sheet, cash flow, ratios and valuation — normalized from raw XBRL into canonical concepts across 12 core tables. Active and delisted names alike, drawn from 10-K, 10-Q, 8-K, 20-F, 40-F filings and their amendments.
Smart-money signals
78M+ rows · 6 tables · Forms 3/4/5/144 + 13F/13D/13G
Insider transactions and institutional ownership — who is buying, selling, and holding — under the same point-in-time discipline. Track filers and positions across the full survivorship-free universe.
Point-in-time
Every fact carries its EDGAR acceptance timestamp. Query any past date and see only what was public then.
No look-ahead, survivorship-free
Append-only — restatements add rows, never overwrite. Delisted, merged and bankrupt names stay in, so backtests can't cheat.
Provenance on every number
A deterministic fact_id pins each value to one SEC filing; verify_fact_lineage round-trips it back to EDGAR in a click.
Accuracy guardrails
A reported 0 stays distinct from missing; derived figures ship their formula and inputs; in the Workspace, untraceable numbers are blocked — enforced by a build-gating test.
Most “historical” data is a lie told in hindsight.
It silently overwrites with restatements and quietly drops the companies that failed. Both errors inflate every backtest. Here is what ours does instead — visibly.
| Metric | as reported | restated |
|---|---|---|
| Revenue | 10,388 | 10,388 |
| Operating income | 2,114 | 1,902 |
| Net income | 1,640 | 1,431 |
| Diluted EPS | 3.05 | 2.66 |
verify_fact_lineage() → both versions retained · each keyed to its filing
- ENRNbankrupt · 2001
- LEHbankrupt · 2008
- WAMUseized · 2008
- BBBYbankrupt · 2023
- SIVBfailed · 2023
- FTXcollapsed · 2022
get_pit_universe(date) → the index as it stood, not as it survived
The adoption gap
Everyone wants agents in the workflow. Trust is what's stopping them.
The blockers are well documented and they are all the same blocker: an answer your firm can't stand behind. Valuein removes the cause, not the symptom.
Agents invent numbers
Hallucination management is the #1 challenge for 55% of organizations piloting agents — a wrong figure can't go near a client or an IC.
Resolved
Agents call typed tools that return SEC-filed values. There is no free-text number to invent — the model selects a fact, it doesn't author one.
Nobody can defend the answer
86% of enterprise leaders name reliability and accuracy as the top blocker to deploying AI agents. Unauditable output fails compliance review.
Resolved
Every value carries a lineage envelope — EDGAR acceptance timestamp and identity hash — so any number traces back to the exact filing it came from.
You need an AI team you don't have
57% of family offices cite a lack of in-house AI expertise; most initiatives stall before production. No RAG to build, no scraping to maintain.
Resolved
Wire the AI clients your people already use in about a minute, or work in the browser. The hard parts — data, governance, provenance — are the product.
Your data and model leak into the vendor
Trust is the #1 barrier in finance; only ~13% of firms actually run agentic AI. Confidentiality concerns keep pilots from ever reaching production.
Resolved
Bring your own LLM. The key is sealed in a 24-hour cookie and never stored. Your data and your model never train ours; Enterprise adds zero-retention.
We didn't ask the model to stop hallucinating. We removed its ability to be the source of a number.
86% of leaders say accuracy is the #1 thing blocking AI agents. So we made the number un-inventable: tools return the figures, the model only arranges the words.
The model types the number.
Generated text
“Meta's FY2024 operating income was $70.2B.”
A raw LLM invents it. A data-API-plus-LLM returns a correct number, then nothing holds the model to it once it crosses into context. The digit is a thing the model said.
A tool returns the number.
The model can reason, narrate, and argue — it is never the source of truth for a digit. Every figure it shows is bound to a verified SEC filing, one click from the original.
Tools return the numbers. The model returns the words. The model never mints a number. We don't ask it not to hallucinate — we take away its ability to be the source of a digit.
Every figure ships its own receipt. Click the URL. Check the filing.
No re-running, no trust-me. A deterministic fact_id pins the value to one filing. Same fact, same hash, every time.
$69,380,000,000
FY2024 · as reported
- concept
- OperatingIncomeLoss
- period_end
- 2024-12-31
- unit
- USD
- availability
- available
- first_filed
- 2025-01-29T21:06Z
- latest_accepted
- 2025-01-29T21:06Z
- restated
- false
9f3c1b7e8a4d2056c1ae77b390f4e2d18c6b5a0f93e84217d6c0b1a59e7f3482
sec.gov/…/0001326801/10-KAn auditor clicks through and confirms the figure against the original filing. Nothing recomputed. Nothing to take on faith.
A reported 0 is a value, not a gap
A genuine reported zero comes back as available, value 0 — never silently nulled. Your model stops confusing “reported $0” with “we don't have it.”
Derived numbers carry their math
Anything computed in the Worker ships its formula and the input fact_ids that fed it, labeled derived. A derived number is never handed a fabricated source.
The wedge isn't “more data.” It's that the figure stays bound to its source inside the model.
A data API hands the model a correct number; nothing keeps it correct once it crosses into free text. We carry the source the whole way — and in the Workspace, anything that loses its trail is held out of the export.
Numbers come from a real filing
- Raw LLM
- Data API + LLM
- Incumbent terminal
- Valuein
The number stays bound to its source inside the model
- Raw LLM
- Data API + LLM
- Incumbent terminal
- Valuein
fact_id + one-click SEC lineage on every figure
- Raw LLM
- Data API + LLM
- Incumbent terminal
- human UI only
- Valuein
Point-in-time, append-only, zero look-ahead
- Raw LLM
- Data API + LLM
- Incumbent terminal
- Valuein
Survivorship-free (delisted names included)
- Raw LLM
- Data API + LLM
- Incumbent terminal
- premium
- Valuein
Deterministic gate blocks unbound numbers in artifacts
- Raw LLM
- Data API + LLM
- Incumbent terminal
- Valuein
- Workspace
Agent-native (typed, tier-gated tools)
- Raw LLM
- Data API + LLM
- Incumbent terminal
- Valuein
Trust guarantee enforced by CI
- Raw LLM
- Data API + LLM
- Incumbent terminal
- Valuein
One token. Three channels. The same point-in-time truth.
One subscription, three ways in — an AI agent, a browser workspace, or your own code. Pick the surface that fits the desk; the data, the lineage, and the tier are identical underneath.
Wire the agent you already have to data bound to its filing.
Connect Claude, Cursor, or ChatGPT to 57 typed tools — fundamentals, ratios, DCF, smart-money, forensic audit. The agent chains them into claims, theses, and reports; every result arrives cited. For the engineer wiring it and the PM reading the answer.
Connect your AI agentCalling Valuein MCP…
Ask in plain English. Get a number you can hand to the IC.
A browser research environment with BYO-LLM chat — bring your own Anthropic or OpenAI key, sealed for 24 hours and never trained on. Run a research SOP, compose claims into a scored thesis, export a cited report. No prompting PhD, 10x the coverage.
Try the WorkspaceRunning flagship equity_research_brief SOP…
Point-in-time SEC data, dropped straight into your stack.
pip install, one token, and stream 111M+ standardized facts through DuckDB — locally, out-of-core, survivorship-free. For quants who want the raw truth in their own code, not a black box.
Read the SDK docsfrom valuein_sdk import ValueinClient # Point-in-time SEC fundamentals — sample tier, no API keywith ValueinClient() as client: df = client.run_template( "fundamentals_by_ticker", ticker="META", ) print(df.head())Real SEC data, real MCP calls — no signup, no key.
These run against mcp.valuein.biz right now. Click a preset and watch the cited response stream back.
Try it live
● Sample tier · no tokenReal SEC data, one click away.
These are real MCP tool calls running against mcp.valuein.biz — no signup, no key.
Want the full universe?
Open the MCP playgroundPick a query on the left to call the MCP server. Real response will appear here.
Ask like you'd ask a junior analyst — get the work back from a senior researcher.
One login, every surface, the same underlying record. The figures come back audit-ready — so you move faster on work you can put your name on.
Spend your hours on judgment, not data entry.
Pull standardized fundamentals, ratios, and comps in seconds — already cited, already point-in-time, ready to drop into the IC memo. Spreading a filing stops being an afternoon and becomes a sentence.
Cover more names without cutting corners.
Screen 19,000+ active and delisted companies, surface what changed since last quarter, and trace every claim straight back to the 10-K behind it. Depth at the pace of a much bigger desk.
Backtests that can't see the future.
Append-only history (restatements add rows, never overwrite) and as_of_date reconstruction with zero look-ahead, across a survivorship-free universe back to 1993. Stream 111M+ facts through DuckDB locally. Cache, retry, replay — same result every time.
Wire SEC data your agent can't fabricate.
57 typed, tier-gated tools and 22 agentic SOPs over SEC-filed data. Connect Claude, Cursor, or Codex in minutes — skip the RAG pipeline and the scraper entirely. Whatever your agent reports comes back with a source attached.
Independent analyst or solo investor? The same platform, the same data, the same per-seat pricing are yours too — start free on the S&P 500 with no card, and scale to the full universe only when the work does.
AI you can put in front of a risk committee.
Your team wants AI. Their compliance team only signs off if every number is auditable — and “trust the model” doesn't pass a DDQ. Only about 13% of finance teams run agentic AI today; the gap is confidence, not appetite. Valuein is the layer that's safe to put in front of real money.
Every number is auditable
Each figure an agent shows carries a fact_id and a clickable SEC EDGAR link. Your compliance team verifies any answer back to the original filing — no "trust the model" required.
Unbound numbers are blocked
In the Workspace, a figure that can't be traced to a source is blocked from the exported report — and a build-gating CI test fails if that ever stops being true.
Your data and model stay yours
Bring your own LLM key — sealed in a 24-hour cookie, never stored, never trained on. Enterprise adds a zero-retention option. A trust story that passes a DDQ.
Under 5% of a terminal seat
Institutional-grade, point-in-time, survivorship-free data for a fraction of legacy terminal pricing — one token across MCP, Workspace, and the SDK, billed per seat.
The business case
What a switch is worth
Hard data-cost saving / year
$57,630
$72,000 (3 Bloomberg seats) → $14,370 (3 Valuein seats) · 80% less
Estimated analyst time reclaimed / year
662 h
≈ $49,650 in reclaimed labor (estimate — your inputs)
Bloomberg seat list price ~$24,000/seat/yr (industry-cited). Valuein Institutional is $4,790/seat/yr — both scale per seat. Labor figure is an estimate from your inputs, not a guarantee.
Hand your risk committee a guarantee, not a hope.
Every figure bound to its filing. The unbound ones blocked. Enforced by a build that fails without it.
See the full case for finance leadersWorks in the tools your team already opens every morning.
No rip-and-replace. Connect Valuein to the assistant on your desk and it instantly gains SEC-filed, point-in-time financial data, each figure carrying its EDGAR provenance.
- Add one server to Claude, Cursor, or ChatGPT — nothing new to learn, no platform to migrate.
- Prefer code? The Python SDK and bulk API serve the same point-in-time data to your models and notebooks.
- A single subscription spans all of them at your tier — no per-channel procurement saga.
Drop the Valuein block into your config — same pattern in Cursor, Codex CLI, Windsurf, and any MCP-compatible client.
{
"mcpServers": {
"valuein": {
"url": "https://mcp.valuein.biz/mcp",
"headers": {
"Authorization": "Bearer YOUR_VALUEIN_TOKEN"
}
}
}
}Start on the house. Scale when the work does.
Begin on the S&P 500 with no card, then open the full universe and smart-money signals as the work grows. Paid plans are priced per seat — per person, never per firm.
Free — S&P 500
$0no card
Begin on the S&P 500, 1993–present.
- Full S&P 500 history
- MCP, SDK & Workspace
- One token across channels
Pro
$49/mo per seat
The full universe for analysts & developers.
- 19,000+ companies, incl. delisted
- 15-year point-in-time window
- 57 tools + 22 SOPs + SDK
Institutional
$499/mo per seat
Full history, smart-money, redistribution.
- 1993–present, all amendments
- Insider + 13F smart-money data
- Webhooks, SLA & redistribution
Start with the free tier.
Full 1993–present history on the S&P 500, no credit card. Connect an agent in 30 seconds or open the Workspace in your browser.