Skip to content
Product · OCO

A CVE lands. Two days later you still cannot say whether it touches you.

The public reference layer under OCO is open to AI agents as Lattice, in private beta. See Lattice

OCO is a cyber knowledge graph. It links a vulnerability to the exploited-list entry that points at it, the techniques it is used with, the detections written for them and the groups on record using them, and every record carries the public source it came from. It runs on local models: no external AI API, and no telemetry leaving the site.

Most people meet OCO without touching it. It is what we watch your estate on when we run the IT and security for a small firm, the reason that service can be priced for a twenty-person company rather than a two-hundred-person one. This page is for the reader who wants to know what is underneath.

  1. 1 · CVE record
    CVE-2021-44228
    Apache Log4j2, as published by NVD
    Source: NVD
  2. 2 · Exploited-list entry
    Log4j2 remote code execution
    The entry CISA added to its exploited-vulnerabilities catalogue
    Source: CISA KEV
  3. 3 · Techniques
    T1190 · T1059
    Exploit Public-Facing Application; Command and Scripting Interpreter
    Source: MITRE ATT&CK
  4. 4 · Detections
    Rules that cover T1190
    Microsoft Sentinel and Elastic rules, and the investigation guides written beside them
    Source: Rule sets
  5. 5 · Who is on record using it
    Groups and campaigns
    Threat groups and named campaigns that use T1190, each with its report
    Source: Public reporting
A real path, read from the running graph on 29 October 2026, for one well-known CVE. Each link is a record with its source attached. The graph holds the connections; whether any of these records matters to your estate, and what to do about it, is a decision for a person.
CVE + KEV
Vulnerability records, linked to the exploited-list entry, the weaknesses and the techniques they connect to rather than listed beside them
ATT&CK
Techniques, with the detections written for them, and the groups and campaigns on record using them
D3FEND
Countermeasures on that same technique space, so a record leads to something you can do
No external AI API
Local models only. Your estate data never leaves your site

Every record in the graph carries the public source it was drawn from. The local-model probe and the coverage-score probe were run on 29 October 2026 and passed. This page describes what is running on that date, not a roadmap.

What it looks like

Pick a record. See where it came from.

The Graph Explorer opens onto the whole graph. Search a CVE and its record opens beside it, with the source it was read from and the path to that source, so you can check the record rather than trust it.

The Graph Explorer with the record for CVE-2021-44228 open in the side panel: its identifier, labels, the source it came from (NVD), the source's own identifier and the address of the public data file it was read from, then its title and description. Above the graph are filters for each kind of record, from advisories and attack patterns to Sigma rules, countermeasures, threat actors and weaknesses.
The Graph Explorer, cropped to leave out the graph's size, which we do not publish. Note the panel on the right: the record names its source and the file it was read from. The filters across the top are the kinds of record the graph holds, each one a place a connection can lead.
What is new today

The day’s signals, from public feeds.

The Threat Intelligence Hub reads the feeds the graph is built from: the CVE record, GitHub and vendor advisories, and CISA’s exploited-vulnerabilities list.

Vendor Watch: five vendor cards ranked by the number of CVEs in the last ninety days times their average CVSS, each showing its CVE count, average CVSS, time of last update and highest severity.
Vendor Watch, on 29 October 2026: the vendors with the most CVEs over the last ninety days, ranked by CVEs times average CVSS. A vendor whose CVEs carry no score shows no score rather than an invented one.
Exploit Activity: a list of CISA exploited-list additions, newest first, each with its CVE number, the product, the actor shown as unattributed, and its severity.
Exploit Activity: the newest additions to CISA's list, newest first.

What to notice

  • The actor reads unattributed, because CISA’s list names none and we do not guess one.
  • Every row is a record with a source, so the panel is a way into the graph: from a CVE here to its techniques, detections and groups there.
  • The list is CISA’s, not ours. A CVE being on it means exploitation has been seen; whether it matters to you depends on what you run, which is the question the rest of this page is about.
  • Where we have written a note on one of these, it is in Insights, with its analysis report.
Ask in words

Ask in words. Get back the records the answer came from.

A question about your own estate, answered against your own data, with the source records listed underneath so you can check the answer rather than trust it.

The SOC Copilot screen: the question 'which hosts expose public facing service?' answered with five services and their host, port and protocol, and beneath it the five source records the answer was drawn from.
The answer panel on a demo tenant, cropped to leave out the navigation. The addresses are RFC5737 documentation ranges rather than anyone's real estate: no customer data goes into a delivered artifact, and that is enforced by one of the six checks rather than by a policy. Note what is under the answer: the specific records it came from. Being able to check an answer is the whole point; an assistant that cannot show its working is a confident stranger.
What is in the graph

Cross-linked, not just collected.

Each of the six kinds of record below is read from public sources: the CVE record and NVD, CISA’s exploited list, MITRE ATT&CK and CWE, D3FEND, Sigma and vendor rule sets, and advisory feeds from vendors, GitHub and the research blogs. The connections between them are what a list cannot give you.

Vulnerabilities

CVE records and exploited-list entries, linked to each other, to the weaknesses behind them and to the software and configurations they affect.

Techniques and who uses them

ATT&CK techniques, tied to the threat signals that mention a CVE, and to the groups and campaigns on record using each technique.

Detections

Sigma, Sentinel and Elastic rules, with their investigation guides, connected to the techniques they cover and, importantly, the ones they do not.

Countermeasures

D3FEND countermeasures mapped against the same technique space, so a finding leads to something you can do.

Signals

Advisories, vendor bulletins and research posts as they arrive, kept with the date and the feed they came from.

Your estate

Assets and what runs on them, held per tenant, so a question about a CVE can be put to the machines you have and not to the world.
Who it is for

Three ways to meet it.

A small firm

You never open it. We watch your estate on it as part of running your IT and security, and write to you in plain language.

A security team

You want the graph itself, against your own estate. Sandbox access is granted case by case; a person reviews every request and says so when it is not a fit.

An AI agent

The public, read-only part of the graph is open to AI agents over MCP, so an agent can look a CVE up in sources instead of in its memory.
Properties

Each of these has a probe behind it.

Runs fully local

Local models only. Your data never leaves your site, verified by a probe against the running system, not asserted in a brochure.

Honest coverage

The system reports what it cannot cover as well as what it can. A coverage score you can query is more useful than a dashboard that only shows green.

Multi-tenant from day one

Tenant scoping is in the data model, not bolted on. One core, many estates, no cross-tenant read path.

What it does not do

OCO tells you what a record connects to. It does not hand you a ranked remediation plan, and it does not take any action on your systems. Deciding what to fix first, and doing the fixing, both stay with your team. If what you need is something that closes tickets on its own, that is not what this is today.

Each property corresponds to a capability probe that must pass before the statement may appear on this page.

Want to see it against your own estate?

For most firms the answer is to have us run it: that is the IT and security service, and it starts with an asset review. If you run your own security team and want the graph itself, sandbox access is granted case by case; a person reviews every request, and we will tell you when it is not a fit.

Request access