Studio
Where your people and AI agents work.
Your engineers work with AI agents beside them, in the terminal and editor they already use, and inside the client's own environment when the work is somebody else's. Each AI agent has a name, acts for a person you can point at, and never reaches further than that person does. Anything that cannot be undone stops before it happens.
The features
Everything in Studio
- StudioAgent runsGive an AI agent a job, then read back every step of it.Read the page
- StudioAgent securityA name, a limit, and somebody who can stop it.Read the page
- StudioAgent identityA person lets an AI agent in, and can stop it again.Read the page
- StudioWhat an AI agent may doIt never reaches further than its owner.Read the page
- StudioAgent sign-insA sign-in expires sooner the more it can do.Read the page
- StudioApproved modelsWhich models may run, who approved each one, and what it left behind.Read the page
- StudioScans and approved packagesApproved packages at the exact version, watched after that.Read the page
- StudioCommand lineThe terminal your engineers already use, on your rules.Read the page
- StudioMCP serverConnect any AI agent, and the rules come with it.Read the page
- StudioSDK and APIOne typed client for the whole platform, given to your team.Read the page
Work inside the client's environment
Meridian Systems does not send you their data. Your engineer signs in to their environment, works there with AI agents beside them, and Meridian can see exactly what that access is and switch it off without calling you.
- Nothing to migrate, and no copy of their work on your infrastructure
- The client grants the access and holds the switch
- Your engineer keeps the terminal and the editor they already use
Every AI agent has an owner
Noor Lindqvist leads quality assurance on your team. Her AI agent is called Quality assurance, because that is its job, and it checks a release against the playbook before it reaches Meridian. It reaches the tracker read only, and never further than Noor reaches, because the limit is read from her permissions rather than copied when the agent was made.
- An AI agent is named for its work, not given a personality
- It never reaches past its owner, and the limit is read, not copied
- Take somebody off the team and their AI agents narrow with them
Anything you cannot take back stops
Reversible work runs without asking. Deploying, publishing, sending to a client: those stop before they happen, go to a named human, and carry the reason that human gave. Nothing is estimated afterward, because the decision was written beside the work at the time.
- The class of every action is decided in advance, from a written list
- An AI agent can propose an irreversible action; it can never approve one
- The approval and the reason sit on the same record as the work
Approve a model once, for everyone
You decide in Studio what your organization may run: which models, and which skills. That decision reaches every project at once, Meridian's included. A team can narrow it, nothing can widen it, and a model nobody approved cannot be called from anywhere.
- One list for the organization, with who approved each model and when
- Skills are approved the same way, and scanned for known holes before anybody runs one
- Start from a library of ready patterns; they arrive as candidates, not switched on
- Your own provider keys, or a model on your own machines
- Enforcement is a switch, it starts on, and turning it off is on the record
One list, from every tool
An engineer signs in from the terminal and gets the approved models. Their AI agent connects over MCP and gets the same list, under the same limits, because the limit belongs to the person the agent acts for rather than to the tool it happens to be using.
- People sign in from the CLI, AI agents connect over MCP, one list behind both
- The same skills reach all three: the console, the API, and an AI agent's tools
- An AI agent never gets more than the person it acts for
See what the AI is actually doing
Which projects are using AI, for what, and through which surface. Coding through AI agents on MCP, document work by people in Studio, split by project and by kind of work, against the models you approved. That is the part most companies cannot answer today.
- Usage by project, by surface, and by kind of work
- What ran straight through, what waited for a person, what was refused
- The spend beside the hours, kept as two numbers and never added
Tell us what your AI should never do alone, and we will show you the controls.