Lattice Analyst

New · for SOC specialists

The analyst assistant that shows its work

Paste an alert, an email, logs or a list of CVEs. Get a verdict, the evidence behind it and the next steps. Facts are written in code from a cyber knowledge graph, and a person always decides.

Alert triage in the Lattice Analyst: a Contain now verdict with reasons and the techniques the text shows

Show the work

Every answer comes with its query plan and its evidence

Open the query plan to see each graph call and model call and how long it took. Open the evidence pane to see the exact nodes the answer rests on, why each one is there and how many hops it sits from the question.

The query plan for CVE-2021-44228 with four graph calls, a code step, one model call and a check of the answer
Query plan. Every graph call, code step and model call with its time. The facts are worked out in code; the model call shows its tokens; the answer is then checked against the evidence.
The evidence pane with the nodes behind the answer, why each is there, and the graph path to the techniques
Evidence pane. Each node with its kind, why it is there and its hop, the share of cited ids verified, and the graph path from the CVE through the weakness to the techniques.

Why it exists

Analysts need an answer in seconds, and a reason to trust it

Generic chat assistants sound sure and can invent a version, a date or a verdict. Lattice Analyst takes the opposite approach: the facts and the decisions come from code and a graph of public sources, and the language model only turns them into readable steps.

Verdict first

Every answer opens with a clear label and the reasons, so you can act or escalate without reading a wall of text.

Evidence you can check

A query plan lists every graph call and model call with its time, and an evidence panel shows the nodes behind the answer.

Honest about what it cannot see

It never says “safe” and never claims something happened on your systems. Each answer lists what is missing.

Local model, no external AI

Answers are written with a local language model. No external AI service is called, so your pasted text stays in the service.

How it works

One recipe for every activity

Every activity follows the same five steps, so answers stay correct and the interface stays simple.

1

One input

Tap a recipe or paste. Pasted text suggests the matching recipe and never runs by itself.

2

Read in code

Indicators, timestamps, headers and behaviours are extracted by rules. Addresses are shown defanged.

3

Facts from the graph

Exploitation status, fixed versions, techniques, groups and detections come from the graph, never from the model.

4

Verdict by written rules

Weights and named exceptions, reviewed like detections.

5

Show and hand over

Verdict, evidence, steps, and drafts that say they are drafts.

Ask about any CVE

The verdict line is written from CISA's record in code: whether it is exploited, whether ransomware use is recorded, the deadline and the fixed versions. The model then explains in plain words, and its steps are checked before you see them.

  • Query plan with timing for every call
  • Evidence panel with the source of each link
  • Follow-up actions on the CVE: checklist, hunt, containment
Analysis of CVE-2021-44228 with a verdict line and a query plan

Triage a paste

Alerts get a label (contain now, escalate, investigate, monitor or need more information) from behaviour rules, with the technique behind each finding and what to check next. Emails are judged by structure, such as failed sender checks and look-alike domains, not by wording alone.

  • No model is used for the verdict
  • A claim of approval is to be verified, never believed
  • Containment steps are suggestions for a person to approve
Alert triage verdict with reasons

Draft, review, copy

Detection rules, messages and incident reports are drafted from reviewed templates and the facts you pasted. Anything unknown stays a visible blank. Code blocks have a Copy button.

  • Sigma rule drafts with their false positives
  • Management, user and customer message drafts
  • Post-incident report in a copyable block
A Sigma rule draft with a Copy button

What it does

The activities an analyst repeats every day

StageRecipes
TriageToday's briefing · Triage an alert · Triage an email · Triage scanner output
InvestigateBuild a timeline from logs · Investigation checklist · Hunt hypotheses · What usually follows
RespondDraft the messages · Containment plan · Ask the runbook · Which CVEs to patch first
ImprovePost-incident report · Draft a detection rule
PersonalMy products: save what you run in your browser, then ask “Does this CVE affect me?”

Everywhere you work

Built for the desk and the phone

Installable and responsive

The chat works on a phone as well as on a desktop, follows the light or dark system theme, and can be installed as an app. Your conversation stays in the page: go Home to the watch lists and come back to it.

Alert triage on a phone in dark mode
Landing page with recipe buttons and the zero-day and ransomware watch lists

Trust and safety

Designed so a person stays in charge

What it will not do

It will not say something is safe, close a ticket, send a message, or run an action. It cannot see your systems, so every answer says what to verify. Drafts are labelled as drafts and unknown facts are left blank.

Privacy by design

The list of products you run is kept in your browser and is sent only inside a question about exposure. The service does not keep what you paste.

Public sources, with notices

Answers rest on public cybersecurity sources, and each link states which source said it or that it was inferred.

Advisory only

Check anything you act on. Containment steps list what they affect and who should approve them.

FAQ

Common questions

Does it connect to my SIEM or scan my network?

No. It works on what you paste and on public data. It cannot tell you whether you are affected, exploited or covered by a rule.

Which AI model does it use?

A local language model, with no external AI service. The model words the explanation; verdicts, facts and structure are written in code.

Can I use the graph from my own AI agent?

Yes. The same graph is served as an MCP server. See the Lattice MCP page for the install guide.

How do I get access?

Open the chat and sign up for a trial key. Keys are rate limited, and verifying your account raises the limits.

Is it production ready?

It is in private beta, provided as is, with no warranty and no service level, under the Lattice beta terms.

See it on your own alert

Paste something you triaged today and compare.