Security & data handling
Clear boundaries for a security-sensitive tool.
Dokima runs in your workspace and delegates analysis to the AI runner you configure: Claude, Codex or OpenCode. Review state and local reports stay on your computer. Findings and status are uploaded for visibility to the rest of your team only if you explicitly approve private team publishing. With hosted models, source and prompts are processed by the configured provider. OpenCode also supports privately hosted local models; where data goes depends on your endpoint and configuration. Here is what your team controls.
Give agents read-only access to findings.
Organisation owners and admins can create named, expiring API keys for agents that need to review hosted findings. Each key can read reports from the selected organisation, including evidence and source excerpts. It cannot upload or change reports, manage members or access billing. Only a hash of the secret is stored, and admins can revoke access at any time. Read the agent API guide.