What SwarmSpace collects, what it does not collect, and how data is handled when plugins execute. Last updated June 2026.
SwarmSpace is a plugin marketplace for personal AI agents — a discovery, trust, and execution layer between AI agents and third-party capability providers. Most users interact with SwarmSpace indirectly, through LUMARA or another AI companion, rather than directly. This policy explains data handling across both surfaces.
We collect the minimum necessary to operate the platform. The categories below cover everything.
| Category | What we collect | Purpose |
|---|---|---|
| Account data | Email address, authentication method (email/password, Google, or GitHub OAuth), account creation timestamp, subscription tier. | Account identity, authentication, billing tier enforcement. |
| Usage data | Plugin slug and call timestamp, credits consumed per call, daily call count (reset midnight UTC), tier-level quota tracking. We log that a call happened — not what was said. | Quota enforcement, billing reconciliation, and aggregate product analytics (e.g. which tools are used most). Query content is never logged. Your user ID is stripped from activity logs after 90 days by an automated daily process; the anonymized record (tool name, tier, result, timestamp) is retained for product improvement. |
| Developer data | Submitted manifest fields (name, description, endpoint URL, trust tier, pricing, tags), submission timestamp and review status, developer UID linked to your account. | Plugin review, listing, and catalogue management. |
| API keys | Each account is issued an ss_ prefixed API key on signup. Keys are stored as a lookup reference, not in plaintext alongside your account document. Regeneration is atomic — the old key is invalidated in the same operation that creates the new one. |
Authentication for direct API and MCP access. |
Three things happen at execution time that are directly relevant to your privacy.
privacy_data_required). LUMARA extracts only those fields client-side. The plugin sandbox receives nothing else through the supported path. This is enforced structurally at the V8 sandbox boundary, not by policy. See the PRISM documentation for full detail.fetch, the credential is attached transparently.network_domains). All other destinations are blocked at the platform layer.SwarmSpace exposes its research and workflow tools via the Model Context Protocol (MCP). When an AI agent calls a SwarmSpace tool:
SwarmSpace hosts the index and trust layer. Plugin execution calls the developer's endpoint. SwarmSpace does not audit third-party APIs or certify their upstream data handling. The Developer Agreement each plugin developer accepts covers data handling obligations, prompt injection liability, and third-party API terms compliance.
If a plugin receives user context (declared via privacy_data_required), that context is subject to the developer's own privacy practices in addition to the constraints SwarmSpace enforces structurally. Review a plugin's manifest and Swarm page before authorizing it for sensitive use cases.
The following third-party services process data as part of operating SwarmSpace. No data is sold or shared beyond what is described here.
| Service | What they receive | Why |
|---|---|---|
| Firebase (Google) | Email address, authentication credentials, account data, usage counters, plugin activity logs. | Authentication, database, and Cloud Functions runtime. |
| Stripe | Email address, payment card details (processed directly by Stripe — never stored by SwarmSpace), subscription tier. | Payment processing and subscription management. |
| Cloudflare | Request metadata (IP, headers) for traffic routing; OAuth token data in KV storage; plugin execution workloads. | Worker runtime, MCP server, plugin sandbox execution. |
| Google Gemini API | Query content for AI processing when Gemini-powered tools or the SwarmSpace discovery agent are invoked. | LLM inference for platform-level AI features. |
| Groq | Query content for AI processing when Groq-powered tools are invoked (using a SwarmSpace project-level key, not your key). | LLM inference for platform-level AI features. |
| NewsData.io | Search query terms for news retrieval. No personal account data is sent. | News briefing content sourcing. |
| User-configured LLM providers (OpenAI, Anthropic, Cloudflare Workers AI, and others) | If you configure a bring-your-own API key, your queries are routed to the provider you selected under your own key. SwarmSpace acts as a pass-through; the provider's privacy policy governs that data. | User-selected LLM routing. |
SwarmSpace sets no tracking cookies and loads no third-party analytics scripts.
sessionStorage for a single purpose: preserving a pending workflow chain parameter during the signup redirect flow (pendingChain). This is cleared immediately after use and never sent to any server.document.cookie values on your browser.SwarmSpace itself sends no email. You do not need to accept marketing email or transactional email from Orbital AI to use the platform.
anonymizeActivityLogs, runs 02:00 UTC). The anonymized record — tool name, tier, result status, and timestamp — is retained indefinitely for aggregate product analytics. Query content is never stored at any point.SwarmSpace uses Firebase Auth for authentication. Firestore stores account and usage data. Cloudflare Workers handle plugin execution with V8 isolate sandboxing. Stripe handles payment processing under Stripe's own privacy and security policies. We do not store payment card details. For the full security architecture, see the Security & Trust Architecture page.
Material changes will be noted at the top of this document with an updated date. Continued use of SwarmSpace after a material change constitutes acceptance of the revised policy.