The AI workspace for growing businesses
New members for your team
Define the teammate you need. Write the job description in plain language: a system prompt, the skills it should master, and the working set of company knowledge it leans on. Fieldwork turns it into a colleague the whole team can call on.
Legal counsel
Contract questions grounded in our signed agreements and playbooks.
System prompt
You are our in-house counsel. Ground every answer in our signed agreements and playbooks, cite the clause you're relying on, and flag anything that needs a human lawyer.
The persona and method this agent brings to every conversation.
Working set
Pick the assets this agent focuses on. Context for the agent, never an access boundary.
Skills
Skills the agent can pull in when a task needs them.
Chat with them like colleagues. Pick an agent in the composer and the whole conversation runs with its instructions, skills, and knowledge. Ask the specialist, or stay with the generalist. It's one picker away.
QuickBooksCash & monthly burn17 months. Burn rises to £86k/mo from November; the model assumes the Meridian renewal lands.
I can model a January start date for comparison.
Finance analyst. Answers from your live numbers, grounded in QuickBooks and the plans in your knowledge base.
One-way in their favor: clause 2 only binds us. I'd counter with our standard mutual template.
I can draft the redline when you're ready.
Legal counsel. Reads the actual agreement, cites the clause, and counters with your standard playbook.
IntercomConversation history
ShopifyOrder #10391Drafted, with their order history attached. They qualify under the 30-day policy, so I'd approve it.
I can send it as soon as you approve.
Support lead. Pulls the conversation and order history, applies your policy, and drafts the reply for approval.
Give agents scheduled work. Put an agent on a schedule or a trigger and its runs arrive reviewed-and-ready in your inbox, Slack, or the knowledge base.
Security first
No training on your data
Neither we nor our model providers train on your data. Inference runs with zero retention wherever the provider supports it.
Access
Agents only see what the person asking can see. Conversations, files, apps and connections stay private until shared, and admins have no way into conversations.
Encryption and isolation
Data is encrypted at rest and in transit. Each user's code runs in its own microVM, and secrets reach models by name only.
Continuity
Continuous backups, replicated to a second region. Data is exported on request and deleted within 30 days of a contract's end.