An AI API for security work

Security teams were early adopters of language models and are still some of the heaviest users: reading unfamiliar code, triaging scanner output, explaining a packet capture, writing a detection rule at two in the morning. Cloakray gives that work one API for the models that are good at it, with a balance you top up in crypto and no card on file.

Models that read code and logs well

• DeepSeek V4.1 Flash Cyber is a DeepSeek V4.1 Flash variant tuned for security tasks, with a context window of about a million tokens, reasoning, tool calls and image input. It is cheap enough to point at a whole repository or a day of logs.
• Claude Opus 5.5 and Claude Sonnet 5.5 for careful code review and long reasoning over a large diff.
• GPT-6 Sol and Gemini 3.5 Flash as a second opinion, or for fast, cheap first passes.

Because they all sit behind the same API, you can run a cheap model over everything and send only the findings worth a closer look to a stronger one, without a second account or a second SDK.

What teams use it for

• Code review and audit. Feed a module, a pull request or a whole service into a long-context model and ask for injection points, missing auth checks, unsafe deserialization and secrets in code.
• Vulnerability triage. Turn a scanner report with hundreds of results into a ranked list with the reasoning behind each ranking, and drop the false positives.
• Log and incident analysis. Summarise authentication logs, proxy logs or an EDR timeline into what happened, in what order, and what to check next.
• Detection engineering. Draft Sigma, YARA or Suricata rules from a description of the behaviour, then have a second model review them.
• CTFs and training. Walk through a challenge binary or a web app with a model that explains each step.

Structured findings

Ask for findings as JSON with structured outputs, so the result goes straight into your tracker instead of being copied out of a chat:

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://api.cloakray.com/v1",
  apiKey: process.env.CLOAKRAY_API_KEY,
});

const response = await client.chat.completions.create({
  model: "deepseek-v4.1-flash-cyber",
  messages: [
    { role: "system", content: "You are reviewing code for security issues. Report only real, exploitable problems." },
    { role: "user", content: sourceCode },
  ],
  response_format: {
    type: "json_schema",
    json_schema: {
      name: "findings",
      schema: {
        type: "object",
        properties: {
          findings: {
            type: "array",
            items: {
              type: "object",
              properties: {
                file: { type: "string" },
                line: { type: "integer" },
                severity: { type: "string", enum: ["low", "medium", "high", "critical"] },
                issue: { type: "string" },
                fix: { type: "string" },
              },
              required: ["file", "line", "severity", "issue", "fix"],
            },
          },
        },
        required: ["findings"],
      },
    },
  },
});

Add tool calling and the model can grep the repository, fetch a file or query your SIEM itself instead of waiting for you to paste the next piece.

Your prompts stay yours

Security prompts tend to contain the things you least want to leak: source code, internal hostnames, credentials that turned up in a log. Cloakray does not store your prompts or responses or use them for training, and the console shows only usage and cost. See privacy for the details.

Each model is also subject to its provider's own usage policy, so keep the work to systems you own or are authorised to test.

Get started

1. Create an account. Your first project and API key are ready straight away.
2. Top up with USDT, USDC, BTC, ETH or another coin you hold.
3. Point your OpenAI or Anthropic SDK at https://api.cloakray.com/v1 and pick a model from the list.