C:\> security
Security & trust
Codepanion gives your support team read access to your codebase and production database, so we hold ourselves to the bar that access demands. Here is how we handle your data, who our subprocessors are, and how to report a vulnerability.
We never train models on your data
Your code, your data, and your conversations are never used to train any AI model. Codepanion uses your data only to answer the questions your team asks, within your own tenant.
The agent runs on Azure OpenAI (Azure AI Foundry). Microsoft is contractually committed to not using your prompts or the model's responses to train, retrain, or improve any of its models, and does not share them with other customers or with OpenAI. Azure OpenAI runs inside Microsoft's own Azure environment. Your data is not sent to a third-party model provider.
Under Microsoft's standard terms, Azure OpenAI may keep prompts and responses for up to 30 days solely to detect and investigate abuse of the service, stored in the geography of our Azure resource (the United States). Authorised Microsoft staff may review content that Microsoft's automated systems flag during that window. That copy is not used for training and is deleted when the window ends.
How your data is handled
- Tenant isolation. Every customer's data lives in an isolated tenant. Queries are scoped to the calling tenant at the data layer, so one customer can never reach another's code or data.
- Read-only database access. Database queries run through a read-only wrapper. The agent can read your production data to answer questions; it cannot modify it.
- Secrets stay in Key Vault. Your database connection strings are stored in Azure Key Vault and fetched at runtime. They never appear in application settings or source. The on-prem connector can keep credentials inside your network.
- Audit log. Every tool call the agent makes writes an audit row (the tool, the query it sent, the environment, and who approved any write), so you can review exactly what was accessed and when. The generated SQL and the rows it returned stay in the investigation's tool steps.
- Encryption. Data is encrypted in transit (TLS) and at rest using Azure's platform encryption.
Subprocessors
We keep our supply chain small. Codepanion runs entirely on Microsoft Azure; we do not use a separate analytics, logging, or model vendor outside it.
| Subprocessor | Purpose |
|---|---|
| Microsoft AzureUnited States (Central US) | Cloud hosting, compute, storage, and managed databases. The product runs entirely on Azure. |
| Azure OpenAI / Azure AI FoundryUnited States (Central US); model requests may be processed in other Azure regions | The AI models behind the agent (chat completions and embeddings). Operated by Microsoft within Azure; contractually no-train. Microsoft may keep prompts and responses for up to 30 days for abuse monitoring. Deployed on Azure's Global Standard tier, so Microsoft may route individual model requests to Azure data centers in other regions under Microsoft's product terms; data at rest stays in the US. |
| Azure Communication ServicesUnited States | Transactional email (invites, welcome messages, notifications). |
Reporting a vulnerability
If you believe you've found a security issue, please email security@codepanion.app rather than opening a public issue. We aim to acknowledge reports within two business days and will keep you updated through to a fix.
We operate in good faith with researchers. Security testing carried out in line with our disclosure policy is authorized, and we will not pursue legal action over it. Testing must avoid privacy violations, service disruption and access to other customers' data. The full policy and safe-harbor terms are published at /.well-known/security-policy.txt, with contact details in /.well-known/security.txt.
Security questionnaires
Running a vendor review or need more detail than this page covers? Email security@codepanion.app and we'll work through your questionnaire with you.