Security Policy — AnyLLM for Jira & Confluence

AnyLLM is built entirely on Atlassian Forge. All backend code runs on Atlassian's infrastructure. maplee operates no server of its own in the data path: prompts flow from the Forge backend directly to the LLM endpoint that the customer's administrator has configured. maplee never sees, receives, or stores customer content.

1. The user writes a prompt in the app's UI (a Forge custom UI iframe inside Jira/Confluence).
2. The Forge backend reads Jira/Confluence content strictly with the requesting user's own permissions (asUser); the app never elevates privileges for AI actions.
3. The backend loads the API key from the Forge secret store (keys are never sent to the browser) and calls the customer-configured LLM endpoint over TLS.
4. The response is streamed back to the user's browser via Forge Realtime.

All persistent data is stored exclusively in Atlassian Forge hosted storage, encrypted at rest and isolated per installation:

• Chat history (max. 50 conversations per user; "temporary chats" are never stored), user memory, personal and shared agents and prompts, admin configuration, and dashboard report content are kept in the Forge Key-Value Store.
• API keys for the customer's LLM, web-search, and image providers are kept in the Forge secret store (encrypted, backend-only, never readable from the frontend).

There is no external database, no file storage outside Atlassian, and no analytics or telemetry. Data residency follows the Atlassian site's pinned location (Jira, as the required product).

The app's manifest declares no external endpoints. Every outbound destination — the LLM endpoint and, optionally, a web-search or image-generation provider — must be explicitly approved by the customer's administrator at runtime (Forge customer-managed egress). Sub-processors are therefore chosen and authorized by the customer:

• Atlassian (Forge platform: hosting, storage) — always.
• The LLM provider configured by the customer admin (for example OpenAI, Azure OpenAI, Anthropic, Google, Mistral, or a self-hosted gateway) — per customer configuration.
• A web-search provider (for example Tavily, Brave, Serper, or self-hosted SearXNG) and/or an image provider — only if the admin enables those features.

maplee adds no telemetry, tracking, or analytics of its own.

• All AI reads and writes to Jira/Confluence run with the requesting user's permissions (asUser).
• Write actions (for example creating a page or editing an issue) require an explicit user confirmation click; the server additionally validates the action against an allowlist, the admin policy, and field constraints.
• Admin settings are served by a separate Forge function that is structurally reachable only from admin modules and verifies admin permissions server-side (fail-closed).
• Scopes follow least privilege; write scopes are never used with user impersonation.
• Admin-configurable per-user rate limiting protects against abuse and cost overruns.

• Prompt-injection mitigation: content fetched by tools is passed to the model marked as data, not as instructions.
• External images in LLM responses are never auto-loaded; they are rendered as clickable links. This prevents silent data exfiltration.
• No inbound attack surface: the app exposes no REST APIs, webhooks, or web triggers; every invocation originates from a Forge module with a server-side user context.

Application logs remain within the Atlassian Forge platform and are never shared with third parties. In normal operation, logs contain only technical metadata and counters — never Atlassian account IDs. API keys are redacted from all diagnostics. In error cases, truncated, key-redacted excerpts of an LLM endpoint's response may be logged for troubleshooting.

• Users can delete any conversation at any time; "temporary chat" mode stores nothing.
• Uninstalling the app deletes the entire app storage (enforced by the Forge platform).
• A weekly automated job reports stored account IDs to Atlassian's personal-data API and permanently deletes all data belonging to closed Atlassian accounts (GDPR Art. 17). This job cannot be disabled.

Please report suspected vulnerabilities to security@maplee.de. We confirm receipt promptly and triage and remediate in line with Atlassian's Marketplace security requirements and bug-fix SLAs. Please do not publish details before a fix is available (coordinated disclosure).

We may update this policy; the "Last updated" date reflects the current version. Questions: kontakt@maplee.de. For privacy details, see our Privacy Policy at /legal/anyllm/privacy.