Skip to content
ActionDesk is open: the whole SEC universe, free, with an event timeline for every issuerLearn more
The Consentra platform

Know what happened. Know what it affects. Take action.

The AI-native platform for securities events — from market intelligence and portfolio monitoring to institutional corporate-action execution.

Why not the obvious names

  • Corporate actions AI

    Describes a feature, and hands the comparison to the incumbents who already own it.

  • Financial AI platform

    Means nothing, and competes with everyone at once.

The seam is the category: the systems that tell you what happened and the systems that process it have never been the same system. An agent is the first thing that can be both.

The spine

Nine steps, whichever document they start from.

A corporate action, a shareholder meeting and a fund registration are the same nine steps pointed at different sources. Consentra builds them once; a product is that engine aimed somewhere specific.

  1. 01

    Source documents and data

    running

    Primary filings, custodian notices, agent messages and registry files — fetched, rate-limited, and archived exactly as served so everything above can be re-run.

  2. 02

    Normalise the event or obligation

    running

    Six documents become one object with an identity, a kind, terms, conditions and a chain. A later document adopts the object an earlier one opened rather than starting a second copy of the truth.

  3. 03

    Identify the affected entities

    running

    Which issuer, which security, which fund, which share class, which accounts, which holders. The entity graph is one graph, not one per product.

  4. 04

    Determine deadlines and rules

    running

    The date that binds, and the rule that set it — an offer expiry, a broker cut-off two days earlier, a record date, a statutory period counted in the working days of the market that owns it.

  5. 05

    Assign the required actions

    in build

    An event is not a notification. It is a set of things somebody has to do, each with an owner, a due date and a consequence if it slips.

  6. 06

    Human and agent approvals

    in build

    The agent proposes; a person with authority disposes. Approval is a record on the event, not an email thread beside it.

  7. 07

    Execute the workflow

    in build

    The election, the instruction to the broker or custodian, the filing, the notification — sent the way you already send it, and acknowledged.

  8. 08

    Collect the evidence

    running

    What was sent, to whom, under which rule, on which date, with the document. Assembled as a by-product of the work rather than afterwards from memory.

  9. 09

    Audit trail

    in build

    Append-only, hashed, and exportable whole. Correcting a mistake adds a record; it never rewrites one.

Steps one to four and step eight are running in ActionDesk today, against live SEC filings. Five to seven and nine are the enterprise half, and are what a design partner is buying into rather than buying.

Entity graph

Issuers, securities, funds, holders, accounts

Document ingestion

Fetch, archive verbatim, re-parse from the archive

AI extraction

Structured where it exists, prose where it does not

Rules engine

Legal basis, clock length, who starts it, what stops it

Deadline and calendar engine

Working days of the market that owns them

Agent runtime

Watches with a job description, not a subscription

Workflow engine

Actions, owners, states, escalation

Approvals

Authority, delegation, sign-off as a record

Evidence and provenance

Every figure traced to its sentence

Audit log

Append-only, hashed, exportable

Permissions

Per entity, per fund, per client. No shared boundary

Notifications

Email, webhook, and into your own systems

Integrations

Custodians, prime brokers, administrators, OMS

Four layers

One graph, four products, one direction of travel.

Each layer is a product in its own right, and each one makes the others worth more. The order is not a roadmap — it is a dependency chain. Intelligence is worthless without the graph beneath it, and execution is what produces data the graph cannot get anywhere else.

01DataFoundation and moat

Securities Event Graph

One normalised object per event, assembled from filings, notices and positions. Not a document store with search over it — a graph in which an event has a status, a set of deadlines, a set of options, the securities it touches and the decisions taken against it.

02IntelligenceDistribution

Event Intelligence

The public surfaces: radars, calendars, company timelines, ownership. Free, indexable, and sourced to the filing. This is how the company is found, and it costs almost nothing to run because it reads the same graph the enterprise product writes to.

03AgentsEngagement

Portfolio and event agents

An agent with a job description rather than a subscription: watch these positions, tell me if the offer moves, if a second bidder files, if a regulatory condition changes. Memory, state, triggers, tools, permissions — and an audit trail.

04ExecutionWhere the contracts are

ActionDesk

The half nobody finishes: position matching, the election, the approval, the instruction to the broker or custodian, the acknowledgement, and reconciliation of what actually settled. This is the enterprise product, and it generates data no public source contains.

The loop

A free page ends in an enterprise contract, and the contract improves the page.

The free surfaces are not marketing. They read the same graph the enterprise product writes to, which is why they cost almost nothing to run and why they get better every time a customer completes a workflow.

  1. 01

    A search finds a free event page

    Indexable, sourced, no account

  2. 02

    They follow the event

    The first thing worth an account

  3. 03

    They add positions

    Now the events are about them

  4. 04

    An agent watches it

    The product stops being a website

  5. 05

    The team wants shared coverage

    Seats become a desk

  6. 06

    The COO buys ActionDesk

    The workflow, end to end

  7. 07

    Execution generates operational data

    Decisions, exceptions, resolutions

  8. 08

    That data improves the graph

    Which improves every layer above

The last two steps are the ones that matter. Everything above them is available to anyone who wants to build it; nothing above them compounds.

The unit

The entity is the event, not the document.

EDGAR contributes documents. Custodians contribute notices. DTCC contributes announcements. Customers contribute positions. Agents contribute workflow state. The graph is what connects them, and the connection is the asset.

12/19fields populated today. The rest need customer positions and a completed workflow, which is exactly why they are the defensible ones.
FieldWhat it holdsSourceStatus
issuerThe filer the event belongs toEDGARpopulated
securitiesCUSIP, ticker, classEDGAR · reference datapopulated
event_typeTender, exchange, merger, rights, consent, spin-offDerivedpopulated
statusLive, pending close, completed, terminatedDerivedpopulated
announcement_dateWhen it became publicEDGARpopulated
effective_dateWhen it takes effectEDGAR · noticespopulated
deadlinesExpiry, withdrawal, election, record dateOffer documentspopulated
optionsWhat a holder can actually electOffer documentsnot yet
considerationCash, stock, ratio, prorationOffer documentspopulated
counterpartiesBidder, target, agent, trusteeEDGARpopulated
conditionsMinimum tender, HSR, vote, financingOffer documentspopulated
regulatory_stateWhat has cleared and what has notFilings · pressnot yet
source_documentsEvery filing on the chainEDGARpopulated
changesWhat moved between filings, and the sentence saying soDerivedpopulated
affected_positionsWhich accounts hold it, and whereCustomer positionsnot yet
decisionsWhat the manager chose, and who approvedWorkflownot yet
instructionsWhat was sent, to whom, by whenWorkflownot yet
acknowledgementsWhat came back, and what did notWorkflownot yet
settlementWhat actually arrivedWorkflownot yet
The moat

Public data is not a moat. The last seven links are.

SEC filings are not a moat: anyone can download them. Language models are not a moat: everyone has the same ones. Even a good interface is copied in a quarter. The last seven links in that chain exist in no public source, and they only accumulate for whoever is actually doing the work.

  1. 01Announcementpublic
  2. 02Multiple source interpretationspublic
  3. 03Extracted termspublic
  4. 04Correctionspublic
  5. 05Affected positionsproprietary
  6. 06Human decisionproprietary
  7. 07Electionproprietary
  8. 08Custodian responseproprietary
  9. 09Settlementproprietary
  10. 10Exceptionproprietary
  11. 11Resolutionproprietary

anyone can reconstruct exists only where the work is done

What accumulates

  • 01Which events produce exceptions
  • 02Which brokers respond slowly
  • 03Which fields require human correction
  • 04Which notice discrepancies actually matter
  • 05Which event structures produce errors
  • 06How deadlines differ between custodians
  • 07Which actions lead to settlement breaks
Open Securities Event Schema

One way to describe a securities event, that brokers, custodians, agents, funds, developers and AI systems can all read. If other people start using these event identifiers, that is worth more than any amount of traffic.

Open

  • The schema
  • SDKs
  • Validation tooling
  • Evaluation datasets
  • MCP tools

Closed

  • The normalised event database
  • Workflow history
  • Custodian mappings
  • Decision data
  • Reconciliation data
  • Enterprise orchestration
  • The network itself
One question

Does this help us own the lifecycle of a securities event?

Every product decision runs through it. It is the only thing standing between this and financial-feature soup.

Yes

  • A Form 4 tracker

    It feeds the event graph, and it is a search entry point.

  • A 13D radar

    Activism is a situation, and situations are the wedge.

  • A tender tracker

    It is the wedge, rendered publicly.

  • An enterprise election agent

    It is the lifecycle.

No

  • A generic P/E screener

    Nothing about it touches an event.

  • A crypto news feed

    Different market, different workflow, no contribution to the graph.

  • An AI stock picker

    Informing a decision is the product; making one is not.

Derived from public SEC EDGAR filings. Extraction from prose is heuristic and carries a confidence; every figure links to the document it was read from. Not investment, legal or tax advice.