Workbench
Every AI subscription in one place. Shared, governed and measured.
Today everyone runs Claude Code, Codex or Cursor on their own laptop, on their own plan. You can't share them, see what they cost, or control what they touch. Workbench puts your team's AI plans on one AgentWorks server. People work with shared Crew agents in chat or in the live terminal, and you see usage by person, Crew and model.
- 01Plans are scattered across laptopsEach engineer pays for, or expenses, their own Claude, ChatGPT or Cursor plan. Some seats sit idle while others hit their limits.
- 02No one knows what AI costsThere is no view of usage by person, team or project, so AI budgets are guesses.
- 03Agents run with full laptop accessA personal agent can see every file, token and repository on the machine, with no record of what it did.
Example goal
Know what AI costs, by team
of recorded runs, target 100%. Illustrative numbers.
How it works
One server for every agent. Chat or terminal, your choice.
Connect the plans once
An admin connects your Claude, ChatGPT, Cursor and other plans on the server as shared accounts. Anyone can still add a private account for their own work.
Give each team its Crew
Set up shared Crew agents with the skills, memory and tools each team needs. People reach them in chat, Slack or WhatsApp.
Chat or the live terminal
Work in plain chat, or switch to the vendor's own CLI running live in a terminal on the server, sandboxed to the folders it is granted.
Measure and govern
See recorded usage and cost by user, Crew, workflow and model. Roles, per-workflow access and a full run log keep it governed.
What you get
One server. Every agent, every plan, every run.
What IT and engineering leaders get from day one, instead of subscriptions scattered across laptops.
- Shared server accounts for Claude Code, Codex, Cursor, Pi and Muse
- Private accounts for people who bring their own plan
- Crew agents in chat, Slack, WhatsApp or the live terminal
- Usage and cost by user, Crew, workflow, project and model
- Roles, per-workflow access and SSO
- An OS-enforced sandbox per agent, a secrets vault and a full run log
Built for your security review
Runs in your cloud. Every action on the record.
Self-hosted
Deployed in your cloud account or data center. Evidence stays in your environment.
Approvals
Anything that changes production or reaches people waits for approval by default.
Audit trail
Every run, tool call, decision and cost is recorded per workflow.
Your models
Your enterprise Claude, ChatGPT or Gemini agreements, or private endpoints.
Scoped access
Tools, folders and secrets are granted per workflow, inside an OS-enforced sandbox.
SSO and roles
Sign-in through your identity provider, with roles and per-workflow access.
FAQ
Questions, answered.
Can we share one subscription across people?
That depends on the plan. Team, business and enterprise plans and API keys are made for shared use; personal plans usually are not, so check each vendor's terms. AgentWorks runs whichever accounts you connect.
Can people still use their own plan?
Yes. Anyone can add a private account for their own work, next to the shared server account an admin manages.
How is usage measured?
Every agent call is recorded with its user, Crew or workflow, provider and model, with token counts and cost. Subscription usage shows as a subscription-equivalent estimate, not as your bill.
What does the terminal mode give power users?
The vendor's own CLI, such as Claude Code, Codex or Cursor, running live on the server in a sandbox. They get the full tool without it touching their laptop, and every run is logged.
More use cases
Other goals agents can own.
Bring your plans in. See usage in week one.
We move one team onto a shared server with you, connect its plans, and show usage and cost by person and Crew from the first week.