Human-led, AI-native security testing

A human security expert, with an agent’s reach.

At Twillsec, senior researchers lead every assessment and steer our in-house agents across more ground than any single tester or autonomous agent system could cover.

Code to exploit

Scenarios built from your code.

We read your code until we deeply understand your implementation, then hand our agents scenarios specific to it. What they find, we test against your running app and show where to fix it.

Built for depth

More than running an agent.

An off-the-shelf agent or autonomous tool only goes as deep as its model and context allow. Ours are built for security testing and steered by a researcher, whose experience and creativity surface issues an agent on its own won’t find.

Judgement on every finding

Precision over volume.

Our approach comes from building multi-agent secure code review and remediation at Meta. A researcher verifies every finding and rates it by how exploitable it is and what it would cost you.

01 / Methodology

Agents for breadth, researchers for depth.

We don’t outsource our thinking to agents. We use them to scale it: they do the exhaustive work, while researchers read the code, review every trajectory, and make every decision and every exploit.

  • Agents
  • Researcher
  1. Agents

    Map the attack surface

    They catalogue every page, endpoint and parameter your users can reach. With code access they also trace each route through the code to where its data ends up.

  2. Researcher

    Read the code, decide where to dig

    We review your code by hand alongside that map: how data really flows, what is valuable, where the logic is unusual. That becomes code-specific test scenarios for the agents, and the list of what we test ourselves.

  3. Agents, reviewed by a researcher

    Probe at scale, review every trajectory

    Specialized agents run those scenarios with the full context, across every endpoint and input we mapped. We review their trajectories step by step: what they tried, what they skipped, where they went wrong. Then we send them back in.

  4. Researcher

    Prove and chain by hand

    We verify every lead, exploit what we can and chain flaws together to show real impact. Every finding is rated by how exploitable it is and what it would cost you.

  5. Researcher

    Fix at the source, then retest

    Each finding comes with steps to reproduce, its business impact and a fix, pointed at the code when we have it. Once it’s fixed, we test it again.

02 / Coverage

One test follows your app into everything it connects to.

01

Web UI and business logic

02

REST, GraphQL and backend APIs

03

Authentication and authorization

04

AI features: LLM prompts, agents and their tools

05

Data exposure and tenant isolation

06

Cloud trust: roles, keys and service access

03 / Your data and the AI

What the agents see, and where it goes.

The agents read your code, signed-in pages and API responses. This is where that goes.


  1. 01

    Open-source models only.

    We use open-source models capable of real security work. Never a closed model behind an AI vendor’s API.

  2. 02

    Served by EU providers, never trained on your data.

    The models run with EU-based providers, in EU datacenters. Your code and data are never used to train a model.

04 / Compliance

A report you can hand to an auditor, a regulator or a customer.

Each report records the scope, the method and every finding, so it stands as evidence of independent testing.


  1. 01

    DORA

    Financial entities must test their ICT systems regularly. The report is the record of what was tested, how, and what was fixed.

  2. 02

    NIS2

    Essential and important entities must assess whether their security measures work. An independent assessment is direct evidence.

  3. 03

    ISO 27001, SOC 2 and customer questionnaires

    Auditors and enterprise buyers ask for a recent independent pentest. You get the full report to share with them.

05 / Engagement

Start with a conversation, not a scope.

You don't need to know what to test or where you stand. Most teams we talk to don't, and working that out is the first step.

Before we test

  1. 01

    An intro call

    Thirty minutes, free. Tell us what you’ve built and what’s on your mind. No scope or security background needed: we ask the questions.

  2. 02

    Where to start

    We work out where an attacker would get the most out of your app and recommend what to test first, and how deep. You get that with a fixed quote.

  3. 03

    Written authorization

    Nothing is tested until you sign an authorization that names the targets, the dates and who to call.

Testing

  1. 04

    Testing

    Inside the agreed window. If we find something critical, you hear from us that day.

After

  1. 05

    Report and debrief

    Each finding ranked by real impact, with proof, steps to reproduce and a fix. Then we walk your engineers through it, and through what to look at next.

  2. 06

    Retest

    Once the fixes are in, we test them again and update the report.

06 / Beyond code review

More than a code review.

AI review tools read your code and point out what might be wrong. Keep using them. We find what they miss, and show you what it means for your business.

01

We test the running app.

Some problems only show up when real users, roles and data meet. Reading the code can’t show them.

02

We look beyond your own code.

Some of the most serious problems sit in the frameworks and libraries your app is built on. Your team never wrote that code, and a review of your code never looks at it.

03

We show the real impact.

When small flaws combine into something serious, we show you how, so you know what to fix first.

First step

Talk to us.

Not sure where to start? That’s what the call is for. Tell us what you’ve built and what worries you.

Book a call with us