You are in Build the agent team · phase 3 of 6
Build the agent team
Agents in Buzz are members, not features: each has its own identity, its own channel memberships and its own audit trail. This section covers building that team — creating and configuring agents, giving them models and skills, organising them into a delegation pattern, and the two decisions that deserve the most care: what real systems an agent may touch, and what all of this costs to run.
-
3.1Create and configure an agent6 steps
Decide the agent's role, set defaults once, create it by dialog or by asking another agent, and give it instructions that stick.
-
3.2Connect a Hermes agent to a Buzz community8 steps
Choose Desktop, relay bridge or native gateway; give Hermes its own identity; and validate who can make it act.
-
3.3Add a model via a custom provider5 steps
Bring a model Buzz does not offer natively — API key, base URL and all — without an afternoon of debugging.
-
3.4Make skills available to your agents5 steps
Skills must be installed globally to be visible to Buzz agents; here is how to install, respawn and confirm.
-
3.5Set up a lead agent and delegation pattern6 steps
Map your team's roles onto agents, choose a coordination pattern, and add the task-checker that catches silent failures.
-
3.6Grant an agent access to real systems5 steps
Every grant is a human decision: narrowest access that does the job, a guardrail where it counts, and a register you review.
-
3.7Control agent costs5 steps
Pin cheap models to cheap work, cap parallelism, use shared compute where it fits, and watch the spend before the invoice does.
-
The ten agents that ship with Berd, what each refuses to do, and how they hand work to each other.