Open it to everyone
- Personal accounts and API keys pasted into chats
- Nobody knows what ran or what it touched
- The bill shows up at the end of the month
AI agents for the whole company
Support, product and sales ask in plain language. Agents do the work in GitHub, Linear, Intercom and Pitch. Vessel checks permissions, holds risky steps for approval and records every action.
Engineering reviews it like any other pull request.
Your teams get AI that does real work. Leadership sets the rules, the budget and what needs approval.
The problem
By team
Real work in the tools you already use. Each request runs with the permissions of the person who asked.
Why does the export fail above 10,000 rows?
Answer with file and line citations
Read the mobile checkout issue, plan it and fix it.
PR opened · lint passed · waiting for review
Update the billing article with this week's changes.
Change waiting for approval
Proposal for Acme with the sales template.
Presentation created
Control
Vessel applies them to every request, before anything reaches a model or a tool.
Start from ready-made roles. Give one person an extra permission with a reason and an end date.
Risky actions wait for a yes from the person or team you choose.
Code changes run on runners you host, using their own credentials.
Lint in an isolated container, secret scanning and protected CI files.
Set token and spend limits for each person and each model.
Every request, tool call and approval is recorded. Reading a transcript takes a reason.
How it works
Plug in GitHub, Linear, Intercom and Pitch, plus Claude, OpenAI or your local models.
Start from the default roles, then adjust for a team or a single person.
People ask in plain language. You follow every run, cost and approval in one place.
Book a demo
See Vessel working with your tools and your rules, and leave with a plan for your first teams.
The booking link will be added here soon.