Prerequisites
The recommended system specifications for deploying a Virtual Signer
You need the following to run a Virtual Signer connected to your io.sign environment.
- Compute for the Broker and MPC Core, isolated according to your deployment architecture. The self-hosted Compose example runs separate containers; your production topology may use separate hosts.
- A hardened host. Turn on disk encryption and a firewall, and close all incoming ports except SSH or whichever port you need for remote access. Require SSH keys, set up SSH brute-force protection, and disable root logins.
- Virtual Signer credentials and your organisation ID from io.finnet support. See Deployment.
- The service URLs for your io.sign environment.
- An HTTPS policy service that handles approval calls from the Virtual Signer. This is separate from io.sign event webhooks. See the Custom policy server section for more information.
- An ML-KEM-768 disaster recovery key pair, generated by your Product Owner (Business User) before the Virtual Signer starts for the first time. See Disaster recovery key pair.
Disaster recovery key pair
Generate this before the first startThe Virtual Signer doesn't start without a disaster recovery public key, and that key is sealed into its configuration when it first registers. There is no supported way to change it afterwards.
The Product Owner (Business User) generates the key pair and keeps the private key offline. The IT/DevOps Engineer receives only the public key, and sets
IOSIGN_SIGNER_BROKER_DR_PK_PEM_FILEto its path on the Broker host.
Generate the key pair with the script shipped in the io.finnet disaster recovery tool, on a secure workstation that is not the Virtual Signer host. The script requires OpenSSL 3.5.0 or later:
./scripts/posix/gen_mlkem768.sh # macOS / Linux
.\scripts\windows\gen_mlkem768.ps1 # WindowsThe script produces mlkem768_pub.pem, which you deploy with the Virtual Signer, and mlkem768_priv.pem, which you store offline and never place on the Virtual Signer host. The private key is the only thing that can decrypt the signer's recovery bundles. If you lose it, you can't recover those bundles.
For the full procedure, including how the Product Owner verifies the deployed key, see Virtual Signer disaster recovery.
Deployment requirements
Use the compute, operating system, container images, and confidential-computing configuration qualified for your release. Confirm sizing against your expected vault count and signing workload; the Broker and MPC Core have different resource requirements.
For the self-hosted Compose guide you need Docker with Compose v2, persistent Broker storage, and certificate and key pairs for mutual TLS (mTLS) between Broker and MPC Core. Production MPC Core needs a supported confidential-computing environment; on the AMD Secure Encrypted Virtualization–Secure Nested Paging (SEV-SNP) path this includes a correctly configured guest exposing /dev/sev-guest. An ordinary container on a compatible CPU does not by itself establish that environment.
For AWS, deploy in eu-west-1 or us-east-2, the regions where AWS offers AMD SEV-SNP, on an m6a, c6a, or r6a instance. Get the deployment template from Deploy on AWS. Have your infrastructure team review the template's IAM permissions and scope the deployment role to the resources it creates.
For development hardware without the required confidential-computing support, enable simulated cryptography explicitly through the development-only settings. Never use this mode with production keys.
Networking requirements
Obtain the exact endpoints for your environment rather than copying a different environment's hostnames.
| Connection | Purpose |
|---|---|
| Broker to authentication service over HTTPS | Obtain credentials for the registered signer. |
| Broker to signing service over TLS HTTP/2 | Receive operation requests and coordinate participation; commonly port 50051. |
| MPC Core to MQTT over secure WebSocket | Exchange signing messages; use the configured host, port, and path. |
| Broker to MPC Core over mTLS | Internal cryptographic operations. Keep the RPC listener private to the Broker. |
| Broker to policy service over HTTPS, optionally with mTLS | Decisions for /keygen, /sign, and /reshare. |
| Services to configured storage and key-management endpoints | Persist and recover encrypted state, including cloud identity and Azure Secure Key Release (SKR) dependencies when used. |
Permit DNS and the platform-specific supporting services required by your release. Expose health and metrics only to your management network. Configure private CA trust where required; do not bypass TLS validation.
Before first registration
- Confirm the organisation, device name, and endpoints all belong to the intended environment.
- Choose the approval mode and policy URL and test the policy response contract.
- Mount the recovery public key and TLS files at the configured container paths.
- Configure encryption and persistent storage, including off-host backups.
- Follow Deploy self-hosted or Deploy on AWS, then Testing the deployment.
The approval mode, policy URL, recovery public key, and Broker–Core channel security are sealed at first registration. Check them before the signer joins a vault.
Updated 17 days ago