CSA STAR Registry

Swarrm is an evidence and accountability layer for AI agents. It operates as an Orchestrated Service Provider (OSP), allowing customers to connect their own OpenAI-compatible model endpoints while retaining control over model selection, data, tools, permissions and business decisions.

Swarrm records AI interactions as signed, tamper-evident receipts containing cryptographic commitments, sequence and causal relationships, policy outcomes and relevant operational metadata. These receipts form append-only audit trails that can be exported as portable proof bundles and checked using an offline verifier. Selective disclosure allows authorised parties to verify specific facts without receiving the complete underlying interaction.

Swarrm does not train or host AI models and does not claim to evaluate their accuracy, safety or semantic behaviour. Its evidence plane is designed not to persist raw prompts, model responses or tool arguments. Its purpose is to make AI-agent activity traceable, independently verifiable and suitable for customer oversight, investigations, audits and compliance workflows.