Agentis is a trust and identity platform for AI agents. Developers and organisations register on Agentis, verify their identity against Companies House or via email, and issue cryptographic decentralised identifiers (DIDs) to their agents. Each agent gets an Ed25519 keypair; the private key is returned once and never stored. Any counterparty can call a single public endpoint to verify who built an agent, what it is authorised to do, and whether it is active.

The platform has three layers working together. The identity layer ties every agent to a verified human or organisational entity; there is always an accountable party behind it. The request signing layer, handled by the companion library proveyouragent, gives agents a way to cryptographically prove on every HTTP request that the request actually came from them. DPoP (RFC 9449) binds each request to the agent's private key so stolen tokens are useless. Delegation chains let orchestrators pass scoped permissions to sub-agents with cryptographic linking, scopes can only narrow as they pass down the chain.

The audit layer records every agent action in a cryptographically chained ledger. Each record hashes the previous one. Altering any record breaks every subsequent hash. An external auditor can verify the entire chain without trusting Agentis as the operator. Developers can write their own events into the chain, invoice processed, data accessed, decision made, making it a complete tamper-evident record of everything the agent did, not just platform activity.

Agentis also supports A2A Agent Cards for structured capability declarations. Agents registered with a card URL have their capabilities imported automatically in a structured format with IDs and descriptions, making the registry queryable by capability.

The open source companion stack includes proveyouragent for request-level signing with DPoP and delegation chains, agentis-verify for FastAPI middleware that verifies incoming agent requests against the Agentis registry, and agentis-quickstart for a complete end-to-end working example.