Home/Platform/Security
Security

Security first

Your code, your keys and your data stay where you put them.

Wakalix never runs the AI model: agents run on your engineers’ machines with their own access, code reaches only the AI provider you approved, and instructions injected into files are treated as data, never as orders.

Transcript

Secure by default. Before the Spring release goes out, Marcus asks for a security review of the web app.

He opens a security cycle. The security engineer agent reads the code and reports. It attacks nothing and fixes nothing.

First, the agent asks what worries him and which checks he wants. Then it asks who has authorised any testing.

The plan has six pieces of work, from a threat model to a test of the running app. Each one has a budget, and the board measures the work against it.

The threat analysis ranks every threat by risk. A reset link used twice is already covered by Priya's one-use links.

The SBOM, made by the team's own tool, lists every component. The scan finds three advisories, but only one reaches code the app runs. It also says what it could not see.

A scan of Larkfield's own code finds three problems. Each one points to the file and the line, with the fix to make.

The penetration test plan waits for a lead. Lukas asks to run it, and Marcus approves. Nothing is tested until he does.

Only then is the running app tested, on staging, inside the agreed window. Each finding says how to repeat it.

Claire sees the newest threat model, SBOM, scan and test plan for every product, on one page. Anything out of date is marked.

Every release examined, and every test authorised first. Secure by default. That is Wakalix.

Never runs your model

Code reaches only the AI provider you approved.

No shared keys

Agents use each engineer’s own access.

Injected instructions are data

A file cannot give your agent orders. Secrets and personal data are caught before output lands.

Questions

Does Wakalix see or store my code?

Wakalix never runs the model. Agents run on your engineers’ machines with their own access.

How do you handle prompt injection?

Text read from files is treated as data, never as instructions.

Read more