Security and privacy
Strict about who sees what.
Velkri remembers what your team decides, so it has to be precise about access. These are the rules it follows, and where they are enforced.
Direct messages never feed AI
DMs stay between the two people in them. They are never indexed, searched by Velkri or used in project memory.
Private by default
Projects are visible to their members only, and Velkri answers each person only from sources that person can open.
Enforced in the database
Row-level security scopes every query to your workspace and your access, so a bug in the app cannot show another team’s data.
People confirm, Velkri proposes
Nothing enters the record or changes a task until someone on your team confirms it. Every change is in the audit log.
Dedicated tenants
Need isolation? Run Velkri in your own deployment with its own database and storage, managed by us.
No data kept by AI providers
Velkri runs on open-weight models through providers that keep no prompts and never train on them. Dedicated orgs can use their own endpoint instead.
Who can see what
Every workspace is its own organisation. Inside it, each project has members, and only members can open the project's channels, calls, files, docs and timeline. Direct messages are visible only to the two people in them.
These rules are not only checks in the app. Velkri's database applies row-level security to every query, scoped to the workspace and the person asking, so data from another workspace or a project you are not in is never returned, even if the app has a bug.
Clients and other outside reviewers see work through share links that show one artifact, read-only, and can be revoked at any time. They do not get access to the project.
How Velkri uses AI
- Answers respect access. When someone asks Velkri a question, it searches only the sources that person can open, and every claim in the answer links to its source.
- Direct messages are excluded. DMs are never indexed, searched or used to build project memory. Velkri replies in a DM only when someone asks it to, and only from sources both people can see.
- Velkri proposes, people decide. Decisions and action items captured from conversations are proposals until a person confirms them. Velkri never edits the record on its own, and every confirmation is in the audit log.
- Your workspace does not train models. Velkri does not train or fine-tune models on your data.
- Usage is visible. AI usage is metered per workspace, project and person. Admins set budgets and decide what happens at the limit; chat, calls, files and tasks always keep working.
Where your data lives
Most teams run on Velkri's shared platform, where every workspace is separated by row-level security and has its own address, such as acme.velkri.com.
Teams that need isolation can choose a dedicated tenant: their own deployment of Velkri with its own database, search index and file storage. It runs the same software as the shared platform and is managed and upgraded by us.
All traffic between your devices and Velkri is encrypted in transit with TLS.
AI providers
Velkri runs its AI on open-weight models through inference providers that keep no prompts and never train on them. If one provider is down, Velkri moves to another that follows the same rules.
On the Dedicated plan, admins can run Velkri on their own OpenAI-compatible endpoint instead, such as Azure OpenAI, vLLM or LiteLLM. Velkri checks the endpoint for the capabilities it needs and runs each job on a model that can do it; those jobs never reach Velkri's providers.
Reporting a vulnerability
If you think you have found a security issue, email [email protected] with the details and steps to reproduce. Please give us a chance to fix it before telling anyone else. We will confirm we received your report and keep you updated.
Questions about privacy or your data go to [email protected]. The full privacy policy explains what we collect and why.