Skip to content

Agents

An agent is a custom AI teammate you create for a specific job. Where the Bridge copilot is a ready-made generalist, an agent is something you design once — its role, its personality, what it knows, and what it’s allowed to do — and then reuse again and again.

Think of building an agent like writing a really good job description, then having a tireless colleague show up to fill it.

Head to Agents → Library → Create Agent. You’ll set:

  • Name — what your team will call it (e.g. “Support Responder”).
  • Prompt — the agent’s instructions: its role, tone, and rules. This is where you describe how it should behave. Plain language works best.
  • Model — which AI model powers it. You can pick one or use your workspace default.

That’s enough to get a basic agent going. The rest is optional polish.

Every agent has a status:

  • Draft — private to you while you build and test it.
  • Released — published to the workspace so teammates (with permission) can use it.

When you’re happy, click Publish to move an agent from Draft to Released.

Some jobs need more than words — they need an agent that can actually run code, edit files, or execute commands. Turn on Computer use and the agent gets its own sandboxed virtual computer: an isolated environment where it can do that work safely.

Beyond the basics, an agent can be shaped with four building blocks. Each has its own page:

Building blockWhat it doesLearn more
KnowledgeDocuments, pages, and data the agent should use as contextTools & knowledge
Tools / MCPActions it can take, including external systems via MCPTools & knowledge
SkillsReusable bundles of capability it can call onSkills & rules
RulesConditional instructions for specific situationsSkills & rules
  • Customer Support Responder — knows your help docs, drafts replies, and can look things up in your CRM.
  • Release Notes Writer — assigned to a task, it gathers what shipped and writes the notes.
  • Data Analyst — answers questions about a database and summarizes trends.

Once an agent exists, you put it to work by assigning it to tasks, adding it to chats, or wiring it into a flow — see Agents in your workspace.