Contract review for the lawyer, resume screening for HR, a one-page summary of a 200-page document for anyone. Ten ready agents: not an empty chat window, but a role with a clear input and a result at the output. Need something else — we build it around your process in two weeks, inside your perimeter.
FAQ
How is an agent different from a plain chat with an LLM?
An agent has a role, a ready scenario, access to your documents and a defined output format. The employee does not have to invent a prompt: they pick an agent, drop in a file and get a document in the right form. A plain chat answers "in general" — an agent delivers the work.
Why are there only ten agents in the catalog?
Because these are the ones that work on your documents from day one, with no integrations or rework: the employee attaches a file and the agent answers according to the department policies from the knowledge base. We do not publish a showcase of a hundred cards with nothing behind them. Everything else is built around your process: describe the task and a standard agent takes about two weeks. Where an engine is needed rather than text, our industry modules come in — MTR analogue search, commercial proposal analysis, pricing analysis; they already run for customers as separate products.
Can the agent invent a number in a report?
Where numbers are involved we do not let the model compute them. In reporting scenarios the model understands the question, then the data is pulled by queries and the report is assembled by code. After the data is retrieved there are zero calls to the model. With no data the agent answers "no data" instead of filling in something plausible.
How does the agent know our internal policies?
Knowledge is connected in three levels. First — the employee attaches a file in the dialogue, works immediately. Second — we load the department policies and templates, 1–2 days. Third — we connect the corporate knowledge base and the agent answers with a link to a specific clause of a specific document that you can open and verify.
The agent we need is not in the catalog. What then?
Describe the task in plain words: what goes in and what you want out. A standard agent with its own knowledge base and document template takes about two weeks. If you need integration with your system or formula-based calculations, that is a separate module quoted per task.
Where is it deployed and does data leave?
Whichever suits you: on-premise in a closed customer perimeter or in the cloud — the same platform. In a closed perimeter documents and correspondence stay inside and you can run fully on local models with no internet access. In the cloud we deploy an isolated instance for you. If you use external models, confidential data is masked before sending and restored in the response: only a de-identified request goes out.
Is this the same as Gatewarden?
No. Gatewarden is the protection layer that can be deployed into the perimeter separately as infrastructure for any AI services. Workspace is the workplace for employees on top of that protection: the agent library, the knowledge base and the interface. Under the hood Workspace uses Gatewarden.