What we build
and give away.

Teams, roles, alerts, pre-sign checklists and SEAL-aligned governance workflows for the Safe multisigs an organisation actually operates — running on infrastructure you control, against a database you can read.

Convixa never holds key material and never executes a transaction. Connected wallets can propose owner changes; signing and execution stay in Safe{Wallet}.

• Inventory of Safes across networks, with threshold, signers, pending count and last activity read from the Safe Transaction Service.
• Organisations, teams and custom roles with granular permissions. The first account to register owns the org; there is no vendor console behind it.
• SEAL-aligned governance: configuration health, compliance scorecards, signer roster and affiliation checks, pre-sign checklists, out-of-band verification cases, incident reporting and readiness drills.
• Alerting on configuration drift, an audit log of who changed what, and CSV export for the evidence pack.

Structured around SEAL Secure Multisig Best Practices. Not affiliated with or endorsed by Security Alliance.

Everyone now has an agent on their laptop that can read files and run commands. These are prompts that use it to answer the question most organisations cannot answer on the day an advisory lands — is this machine actually affected? — without installing anything or sending anything anywhere.

Every prompt is read-only: no modification, no installs, no privilege escalation, no network probes from the audited host, and nothing about the machine leaves it. Findings are written down for a human; the agent never remediates.

• An endpoint audit for macOS, Windows and Linux: every trust store on the machine rather than the default one — each JDK, each certifi, Node’s compiled-in roots, WSL distributions, Electron apps that never see a browser revocation push.
• An EC2 and server audit covering the same ground plus containers, an outbound dependency map that separates AWS-terminated traffic from third-party, and whether a fix survives the next scale-out.
• Each prompt opens by verifying its own facts against primary sources and reports which it could not confirm — an agent repeating a version number from its prompt back as fact has told you nothing.
• Output is a findings table, a required NOT CHECKED section, a remediation plan, and a JSON block so a fleet of reports aggregates without anyone reading prose.

A prompt is a structured way to look, not an assessment. Output is only as good as the access the agent had and the facts it verified.