DevJock.aiBuild an Assistant with Multiple Agents and a Dashboard
A step-by-step guide to designing an assistant from several coordinated agents and reviewing its dashboard.
An assistant is the composite deliverable in DevJock: several agents working together, usually with a dashboard that shows their activity. The quickstart builds one end to end. This guide slows that process down so you can design each part deliberately.
1. Decide the agents and their roles
Start by listing the distinct jobs your assistant needs to do, then map each job to one agent. An agent is a single unit: one prompt plus one model, and optionally tools and skills. Keep each agent narrow. A support assistant might have a triage agent, a research agent that reads your docs, and a drafting agent that writes the reply. Narrow roles are easier to prompt, test, and debug than one agent that tries to do everything.
2. Write or reuse prompts
Each agent is defined by a versioned prompt. Open the Prompts area to see what already exists and reuse a prompt when one fits. When you write a new prompt, state the role, the inputs the agent can expect, and the shape of the output you want. Because prompts are versioned, you can revise wording later without losing the earlier version.
3. Pick models
Assign a model to each agent. You can set a specific model per agent, override the model on an individual call, or let an agent inherit the platform default so operators change one setting instead of many. Match the model to the work: a fast, inexpensive model for triage, a stronger one for reasoning or drafting.
4. Connect tools
If an agent needs to reach external systems, attach tools. DevJock exposes an MCP-over-HTTP API, and you register MCP servers and tools under the Developer group. Give each agent only the tools its role requires.
5. Assemble the assistant
Combine your agents into one assistant. Agents coordinate by sharing knowledge through memories on tasks and by spawning in parallel over the agent-to-agent protocol, so define how work passes from one agent to the next.
6. Review the dashboard
Open the assistant's dashboard to watch its agents run, follow the tasks they create, and read the memories and handoffs they leave behind. Use this view to spot where a prompt needs tightening or a role needs splitting.