Security & Data Handling

Last updated: 22 July 2026

1. Local-first architecture

Metadata extraction, the symbol index, and every query run on your own machine. Your PackagesLocalDirectory, X++ source, the contents of files the tools read or write, and the results of any data reads never leave your environment. What reaches us is limited to account identity, license state, and the privacy-safe operational records described below.

2. Zero data retention of your source

No feature of the Service stores your source code, query results, or business data on our infrastructure. The operational records we do keep are value-free by construction:

• Audit records identify which tool ran, when, and by which seat, with a cryptographic fingerprint of the arguments and a response byte count — never the arguments, code, field values, or record keys. For change-making tools we add a names-only summary (object and model names). Read events are kept 90 days; change events 180 days; then deleted automatically.
• Operational telemetry records tool, status, latency and byte counts — never argument or response content.
• The published index a team shares carries object names and paths only — never source or source-derived call data.
• Where an optional company feature briefly routes a custom-model code snippet through our service to your own online dev box, it is held only in memory for that one request and immediately discarded — never written to disk, logs, or a database. Microsoft standard code is never eligible for this path (see section 4), and the feature is off by default.

This is enforced in code and guarded by an automated regression test that fails our build if source text ever appears in an emitted record.

3. AI model processing runs through your own provider

The Service orchestrates AI features and supplies D365 metadata tools; it does not itself host or operate a large language model. AI inference is performed through the provider you configure — your own API subscription (for example Anthropic, OpenAI, or Microsoft Azure OpenAI) or, inside your IDE, your GitHub Copilot subscription. Your prompts and the code they reference go from your machine to that provider under the terms you hold directly with it, and are not stored on our servers. We do not use your data to train AI models.

4. Microsoft standard code

Microsoft licenses its standard D365 F&O source to your environment for internal use. When an AI feature needs to reason over Microsoft standard code, it is processed only through a channel you already license and that Microsoft operates — GitHub Copilot Business or Enterprise, or Azure OpenAI in your own tenant — both of which exclude your data from model training. Our optional company relay never transmits Microsoft standard source; it is limited to your own custom-model code. You are responsible for maintaining an account tier whose terms permit this, and for your organisation's data-governance requirements.

5. Isolation, access and licensing

Team data is isolated per tenant. Registration tokens and activation codes are single-use and expire in 15 minutes. Licenses are cryptographically signed, and each seat's capabilities are gated by tier and enforced server-side, not from editable local files: a developer seat cannot use the consumer chat, and a read-only super-user cannot change anything.

6. Transport and storage security

TLS in transit everywhere. Server secrets are held in managed stores. On your machine, the agent seals its database credentials in an encrypted local store and encrypts its backups. Hosting is on Microsoft Azure (UK South).

7. Responsible disclosure

Found something? Please write to security@tse-assistant.com. We welcome good-faith security research and will work with you on any valid finding.