Agents is a private AI agent workspace for teams that La Boétie built in house on Hermes Agent from Nous Research. Staff sign in with Microsoft 365, talk to their AI colleagues, share a task board and create a new agent by describing the job. The workspace has no public signup or price.
Staff sign in with Microsoft 365 and create new agents by describing the job
Agents runs on Hermes Agent from Nous Research.
Staff sign in through Microsoft Entra with their Microsoft 365 account, and each person holds the Admin or the Member role. Members see the team agents and their own private agents. Admins have access to every agent, the vault and the Hermes console. Both roles can create an agent.
A person describes the job and answers a few questions, then reviews and approves a drafted personality. The recruiter then picks skills for that role. Each skill goes through the Hermes security check before installation.
For each agent, an admin ticks the team variables it receives. Only those variables enter the session and the container of that agent. The web app stores no secret of its own.
Agents hand work to each other on a shared kanban board. A dispatcher reads the description of each profile to choose who takes a task. Each agent has one main conversation and runs each task in a separate session.
La Boétie built Agents as in house work and offers no signup, download or price for it. The repository is private. A team planning a private agent workspace can describe its needs in a brief, and the studio will reply from this build.
An AI agent workspace is a shared place where a team uses AI agents that keep their own memory, skills and access. A chat assistant answers one person. A workspace gives the company a set of named colleagues: a sales assistant, an office manager, a research agent. Staff can talk to them, hand them tasks, and let them hand work to each other.
Agents is the private workspace La Boétie built on Hermes Agent, the open source engine by Nous Research.
| Question | Chat assistant | Agent builder | Agent workspace |
|---|---|---|---|
| Who uses it? | One person | A technical builder | The whole team |
| Memory | Per user | Per flow | Per agent |
| Setup | None | Flow design | A job description |
| Runs alone on a schedule? | Limited | Yes | Yes |
| Secrets per agent | No | Varies | Yes |
A small team with simple questions may need only a chat assistant. A workspace pays off when agents need different data, different permissions and a life beyond one conversation.
Skills can come from outside your team, so treat a skill like a software dependency. A security check reduces the risk of a bad one, and an admin should still read what an agent holds.
| Role | Sees | Can do |
|---|---|---|
| Member | Team agents and own private agents | Chat, assign tasks, create private agents |
| Admin | Every agent, the vault, the Hermes console | Manage access and secrets |
Staff sign in through Microsoft Entra with their Microsoft 365 account. When someone leaves the company and IT closes the account, access ends with it, so nobody needs a second list of users.
The usual mistake is one shared key that every agent can read. Here, the team vault holds variables, and an admin grants them per agent. Only those variables enter that agent's session and container. The web app stores no secret of its own.
| Agent | Example grant |
|---|---|
| Sales assistant | CRM key |
| Finance assistant | Accounting read key |
| Research agent | Search key, nothing else |
An injected instruction in a web page can then reach only what that one agent holds.
Agents hand work to each other on a shared kanban board. A dispatcher reads each profile's description to pick who takes a task. Each agent has one main conversation and runs each task in a separate session, so a long job stays out of the daily chat.
Clear profile descriptions matter, because the dispatcher reads them. Two agents with overlapping descriptions will compete for the same work.
| Good first agent | Why |
|---|---|
| Internal knowledge assistant | Low risk, read only |
| Weekly report writer | A clear, repeatable output |
| Lead research | Easy to check by hand |
Start with read access, review the first weeks of output, then add actions one at a time. Actions that send, delete or spend need a person's approval.
Send a brief with your team size, your tools and the first job you want to hand over. GateLLM can sit in front of the models, and the CRM SDK connects an agent to customer data inside an application.
An AI agent workspace is a shared application where a team creates AI agents, talks to them and follows their work. In Agents, each agent has a card with its role, plus its own memory, skills and conversations. The team shares a task board in the same web app. Members follow the tasks of each agent on that board.
Both the Admin and the Member role can create an agent. After a few questions, the recruiter drafts a personality for the member to review and installs skills for the job. Private agents stay visible to their owner and to admins. Team agents are visible to every member. Admins also see profiles without a card. Only admins can open the Hermes console.
The tools of each agent, such as its terminal and files, run in a rootless Podman container with no access to the Hermes folder or to other agents. A vault variable reaches an agent only after an admin ticks it. The whole box is capped at 3 GB of memory and 1.2 CPU.
The engine is Hermes Agent from Nous Research. We pin it to one version and add two patches. The first patch adds per person access control to profiles across the dashboard routes and the conversation socket. The second repairs conversation titles that come back with a badly encoded accent. The web app and the team plugin sit on top of that engine.
Sign in goes through Microsoft Entra, where the enterprise application requires assignment. The repository documents no other sign in path. An administrator adds each person in Entra and gives them the Member or the Admin role. A team on another identity provider should name that provider in a brief so the studio can say what would need to change.
Agents has no public offer. The repository is private with no license file, and La Boétie sells no seat, download or plan for it. Access to the workspace stays in house. Teams that want a private workspace for AI agents can send a brief from this page that names the roles, the tools and the sign in they need.