# Offensive-security platforms

> Pentest, attack-simulation and attack-path platforms that want source-grounded evidence behind every path.

Source: https://keryn.dev/partners/offensive-platform/

**Ground every attack path in the code and the agents behind it.** Keryn is pre-release; every partnership starts as a time-boxed pilot.

## The problem

Your platform shows what an attacker can reach. Customers then ask where in the code it comes from, whether the fix really closed it, and which AI agents, tools and MCP servers sit on the path. That context lives in source and configuration, not in traffic.

## What Keryn gives you

- Code-side evidence for attack paths: a trust graph of services, agents, tools, secrets and data flows, each edge with a proof level and a source location.
- AI and agent surface: an AI-BOM of models, SDKs, agents, prompts, MCP integrations and workflow authority, exportable as CycloneDX.
- Verified closure: re-run on the fix commit. An unchanged snapshot costs nothing; a changed one costs only its delta.
- Local first: scans run on the customer side by default and nothing uploads unless someone chooses to.

## How it connects

Diagram: Offensive-security platform integration. An offensive-security platform sends findings and attack traces into Keryn through SARIF or capture import, or calls the customer API and the MCP server. Keryn returns source-grounded evidence and retest results.

- Your findings in: SARIF, Burp or ZAP XML, HAR (keryn dast import).
- Keryn evidence out: SARIF, CycloneDX, keryn-export/1.
- Call us: the hosted customer API with @keryn/sdk and an organisation key; `@keryn/mcp` and `@keryn/skills` for your own agents.

## What is available today

| Surface | Status |
|---|---|
| SARIF, Burp/ZAP XML and HAR import | Available |
| SARIF, CycloneDX and keryn-export/1 output | Available |
| Customer API with organisation keys | Available, bounded scans first |
| `@keryn/sdk` TypeScript client | See [Download](https://keryn.dev/download/) for npm availability |
| `@keryn/mcp` (stdio and streamable HTTP) and `@keryn/skills` | See [Download](https://keryn.dev/download/) for npm availability |
| Active testing in the hosted service | Not available yet |

See [deployment models](https://keryn.dev/partners/deployment-models/) for who operates what.

## Proof you can check

- Parity: on a public corpus Keryn's findings agree with the reference engine (Semgrep CE 1.157.0) on 10,306 of 10,306 findings, with one extra (receipt 2026-11-02).
- Detection: recall is 0.34 on 582 real-fix pairs, published with its denominator and its gaps (scoreboard 2026-10-28).
- Known fixes: 3,156 signatures in the known-fix database (2026-10-28).
- Local first: the CI copy-paste run made no network attempts (2026-11-02).
- Every evaluation suite, with inputs, denominators and limits, is at keryn.dev/evals.

## Commercial terms

Pilot first on the starter credits ($10 for every new organisation, plus $10 with a design-partner code). Then Keryn Cloud credits at a partner rate by annual commit (Silver, Gold or Platinum: 10, 20 or 30 percent off list) and, if you co-sell, a 15 percent referral on first-year purchases by the customers you introduce. See partner terms. Contracts are drafted per partner and reviewed by counsel before signature.

## First 30 days

| Days | Keryn | You |
|---|---|---|
| 1-5 | NDA, one technical owner each, one pilot question and a decision date | Pick one representative target and one sanitised finding object |
| 6-12 | Map the object to the canonical result, publish fixtures, list unknowns (schema, auth, identity mapping, dedupe, lifecycle) | Answer each unknown with the smallest artefact |
| 13-20 | Run Keryn on the target source; produce evidence for three attack paths | Review: useful, wrong or missing? |
| 21-27 | Retest after a fix; measure time saved per path | Score against the pilot plan |
| 28-30 | Written decision: extend, productise or stop | Same |

Exit criteria: Mapping loses no id or provenance; at least 70 percent of reviewed paths rated useful by the partner (a target to confirm with them, not a promise); nothing called integrated without live acceptance.

## What we ask of you

One sanitised finding or attack-trace object from your platform. We map it, show the mapping as a candidate rather than your official schema, and agree what accepted means before anyone claims an integration.

[Become a partner](https://keryn.dev/partners/apply/?track=offensive-platform) or write to [keryn@keryn.dev](mailto:keryn@keryn.dev).
