> For the complete documentation index, see [llms.txt](https://docs.konvu.com/konvu/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.konvu.com/konvu/reference/data-collection-and-security.md).

# Data Collection & Security

What data Konvu collects, how code is handled, what's shared with AI models, and our compliance posture.

Konvu minimizes data processing and storage. When data is required, it's securely stored and isolated. This page covers what Konvu collects, how your code is protected, and what's shared with AI models.

## SOC 2 Type 2

Konvu is SOC 2 Type 2 compliant, certified by an independent audit firm. Security controls are continuously monitored through [Vanta](https://www.vanta.com/). A SOC 2 report is available upon request.

* **Trust Center**: [trust.konvu.com](https://trust.konvu.com/)
* **Data storage**: AWS us-east-2

## Code privacy

Konvu never stores your code.

When you use [Managed Analysis](/konvu/get-started/connect-your-environment/source-code/managed-analysis.md) (GitHub App), source code is handled as follows:

1. The integration uses minimal read-only scope. No GitHub refresh or access tokens are stored -- authentication uses temporary tokens generated by the GitHub App.
2. The repository is cloned into a transient Lambda function using temporary credentials.
3. Stack analysis runs precompiled static code. Nothing beyond the analysis binary executes.
4. Only the identified stack information (see table below) is sent back to Konvu.
5. The Lambda is destroyed immediately.

For full on-premise control, use [Self-Hosted Analysis](/konvu/get-started/connect-your-environment/source-code/self-hosted-analysis.md) -- your code never leaves your infrastructure.

## What data Konvu collects

| Data collected                                                                                                             | Source Code Analysis | AI Agents |
| -------------------------------------------------------------------------------------------------------------------------- | :------------------: | :-------: |
| Repository names                                                                                                           |          Yes         |     --    |
| Stack information (languages, frameworks, runtimes, OS, package managers, manifest locations, backend/frontend indication) |          Yes         |     --    |
| Dependency names and versions                                                                                              |          Yes         |     --    |
| App name from manifest file                                                                                                |          Yes         |     --    |
| Dependency reachability (boolean, file name, line number)                                                                  |          Yes         |     --    |
| Vulnerable function reachability (boolean, file name, line number)                                                         |          Yes         |     --    |
| Small code snippets highlighting vulnerable code (typically 5 lines per match)                                             |          --          |    Yes    |

## AI model data sharing

{% hint style="info" %}
This section applies if you have an [AI triage policy](/konvu/triage-and-remediation/exploitability-assessment/how-assessments-work.md) configured.
{% endhint %}

When Konvu's AI agents assess a vulnerability, the process works in a targeted loop:

| Step                     | What happens                                                                                                                                                                                                                                            | Code shared             |
| ------------------------ | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ----------------------- |
| **1. Initial prompt**    | Konvu sends vulnerability metadata (CVE ID, dependency, version, enriched metadata) to the model. The model returns search patterns (`grep`, `list file`, `read file`). Source code analysis runs them on a temporary copy in an ephemeral environment. | None                    |
| **2..n. Agentic loop**   | Search results and evidence (matching line + surrounding 2 lines + file path) are sent to the model. The model returns a new search pattern, which gets executed.                                                                                       | Up to 5 lines per match |
| **n+1. End of analysis** | Investigation steps, tools used, results, and evidence are persisted in the Konvu backend. The ephemeral environment is terminated and the temporary code copy is erased.                                                                               | None                    |

### What is and isn't shared

| Sent to the model            | Never sent                                  |
| ---------------------------- | ------------------------------------------- |
| Up to 5 code lines per match | Entire repository                           |
| File paths of matched files  | Files unrelated to the vulnerability        |
| Vulnerability metadata       | Secrets, API keys, credentials              |
| Dependency version info      | Customer data or proprietary business logic |

### Training opt-out

Konvu has opted out of OpenAI's data sharing program. **Your code is not used to train or improve OpenAI's models.** Code snippets are processed for immediate analysis and not retained. See the [OpenAI data usage policy](https://platform.openai.com/docs/models/data-usage-policies).

### Access controls

* **Read-only access** -- the agent can only read code, never modify it
* **Targeted searches** -- only patterns related to the specific vulnerability
* **Path restrictions** -- cannot access files outside the repository
* **No execution** -- code is analyzed statically, never executed
* **Ephemeral environment** -- code stored only during analysis, deleted immediately after

### Bring Your Own Keys

You can configure AI triage assessments to use your own OpenAI API keys or endpoint. Your OpenAI token is stored encrypted in Konvu's database. Configure this under **Configuration > AI Triage** in the Konvu dashboard.

## Secrets data handling

[Secrets triage](/konvu/triage-and-remediation/finding-types/secrets.md) handles credential findings separately from the vulnerability metadata flow described above. Findings sent to the backend identify the credential with a hash and include its provider, repository, file location, commit, and available validation evidence. The finding record uses a masked representation rather than storing the raw credential as its identifier.

When provider validation is enabled, the analysis process can use the detected credential in live requests to its provider. This is separate from sending evidence to an AI model. Detection can continue with provider validation disabled; the organization-level policy defaults to disabled for on-premises workspaces and enabled for managed workspaces unless overridden.

The Secrets detail panel displays masked matches and available evidence, including check times and known identity facts. A cached result represents an earlier check, and validation receipts may omit the exact request or response.

## Trust center

Review Konvu's security practices, request the SOC 2 report, and access compliance documentation at [trust.konvu.com](https://trust.konvu.com/).


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation by asking a question.

Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters:

```
GET https://docs.konvu.com/konvu/reference/data-collection-and-security.md?ask=<question>&goal=<user_goal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.