"Insider trading API" is one of the more consistently searched phrases among people building financial AI agents right now, and for good reason - insider activity is genuinely informative when it's read correctly. The problem is that most APIs answering that search stop at handing you a list of SEC Form 4 filings and calling it a signal. A list isn't a signal. Reading it correctly usually means reconciling it against something else.
Here's Apple, as of this week. Two things are simultaneously true:
An insider sold. SVP, General Counsel and Secretary Jennifer Newstead sold 1,439 shares on August 11 and another 1,439 on August 18, each around $307.50 - two transactions totaling roughly $885,000. Zero insider buys recorded in the same window. Read in isolation, that's a bearish data point.
Institutions bought, heavily. Across the same reporting period, institutional ownership rose by a net 331.3 million shares - Vanguard Group adding 26.9 million, BlackRock adding 18.3 million, State Street adding 12.8 million, and 2,863 total holders increasing their position against 3,201 decreasing. Total institutional ownership sits at 76.57% across 6,473 holders.
A raw insider feed reports exactly one of those two facts - the one it's scoped to - and reports it as if it were the whole picture. An agent that only sees "insider: SELL" and stops there has a materially worse read on the stock than one that sees both signals and understands they need to be weighed against each other, not averaged or ignored.
A single executive selling a few thousand shares over two dates is exactly the shape of routine, pre-scheduled diversification or tax planning - not a conviction signal about the company's prospects. Executives sell for reasons that have nothing to do with their view of the stock all the time: a mortgage, a 10b5-1 plan set months in advance, simple portfolio rebalancing. A feed that reports "insider activity: SELL" with no size context, no counterpart signal, and no distinction between "the CEO liquidated a third of their position unscheduled" and "a VP sold two routine tranches" is handing your agent a fact stripped of the context needed to weigh it.
| Field | Why it's necessary, not nice-to-have |
|---|---|
| Transaction-level detail: who, role, type, size, price, date | A Section 16 officer's trade means something different from a 10% beneficial owner's - role matters, and raw aggregate counts hide it. |
| The institutional trend, not just the latest snapshot | One 13F filing tells you where ownership is. The change - holders increasing vs. decreasing, and by how many shares - tells you where it's moving, which is the actually useful part. |
| A reconciled verdict when the two signals disagree | Insider and institutional data disagree often. Someone has to actually weigh them against each other - ideally before your agent burns a reasoning step doing it from scratch. |
| A confidence number tied to real signal strength | "Insider activity found: yes/no" is a boolean pretending to be a signal. A mixed, low-conviction read and a unanimous, high-conviction one shouldn't look the same. |
This is how Stocklake's insider tool answers it: the two sides are scored separately and then reconciled - in AAPL's case, insider_signal: NEGATIVE, inst_signal: POSITIVE, folded into an overall NEUTRAL read - alongside the buy/sell counts, institutional ownership (76.57% across 6,473 holders), and a plain-language summary that names the actual movers ("2 sells totaling $885K by SVP Jennifer Newstead... net increases from Vanguard (+26.9M sh), BlackRock (+18.3M sh)... offsetting declines"). The transaction-level Form 4 detail is what the synthesis is built from upstream, and the summary cites it by name where it matters - but what the agent gets handed is the reconciled read with its reasoning spelled out, not two disagreeing fields to reverse-engineer.
Insider-plus-institutional data isn't useful in isolation, either - it's one input into a broader read that also includes technicals, sentiment, and sector context, the same "don't hand the agent one fact and call it a decision" principle that shows up everywhere else in how Stocklake's tools are built. It's a Pro-tier field (get_insider_activity), reachable in the same call shape as everything else in the API - no separate SEC-scraping pipeline to stand up yourself.
No. A single executive selling a routine-sized tranche is often pre-scheduled diversification or tax planning, not a conviction call. What matters is the size relative to the company, whether other insiders are also selling, and what institutional ownership is doing over the same window - not the raw yes/no fact of a sale.
Insider activity is SEC Form 4 transactions by officers, directors, and 10%+ owners - individual, transaction-level trades. Institutional activity is the 13F-based ownership trend across large holders like Vanguard and BlackRock - a much bigger, much harder-to-fake signal about where ownership is actually moving. Stocklake's insider tool reconciles both into one verdict rather than reporting them separately.
Coverage depth follows the underlying filing regime: SEC Form 4 only exists for Section 16 officers, directors, and 10%+ owners of US-listed issuers, so a foreign-listed or thinly-covered name will naturally have less insider data than a mega-cap like Apple. That's a constraint of the filings themselves, not something an API should paper over with a fake result.
Not for the filing facts. Free and guest keys get the 5 most recent insider filings of the last 90 days with the buy and sell counts, in the same call shape as the rest of the tools. The scored insider and institutional read, the summary and the full year of filings are Pro.