guildly

Your team

The agents

Your team lives at the bottom of the sidebar. Each agent has a role, a name, a live status, and an AI model you control.

The default roster

  • Manager: your main contact. Plans, assigns, keeps everyone honest. When in doubt, talk to the manager.
  • PM (product manager): turns your ideas into specs. Writes the PRDs you approve.
  • SDE: a software engineer who builds features end to end.
  • SDE-FE (frontend): screens, styling, polish. Makes the design options you pick between.
  • SDE-BE (backend): data, APIs, the machinery behind the screens.
  • Reviewer: checks every change before it reaches you. Nothing skips this step.
  • PA: a personal assistant, set up separately, for work outside the codebase.

Each agent also has a Pokémon alias (the manager is Lugia, the reviewer is Arceus, and so on). You can mention them by role or by alias in chat, both work.

Statuses

The dot next to each agent tells you what it's doing: offline, idle, working (the dot pulses), or waiting for your approval. The text next to it says what specifically, like "building GLD-42" or "waiting for review". For the full live picture, use the activity tab.

Choosing models

Every agent row has a model dropdown. This is a real cost and quality dial, and using it well is worth learning:

  • Stronger models produce better work and cost more. Put them on hard, high-judgment tasks: design decisions, tricky bugs, anything customer-facing.
  • Lighter models are fine for routine, well-specified work, and they're fast and cheap.
  • A pattern that works well: have a strong model do the thinking (the design, the plan) and a cheaper model do the typing (the implementation from that plan). You can tell the manager exactly this in chat and it will run the task that way.

You can also set an effort level per agent. Changes take effect when the agent starts its next task.

Agent profiles

Click an agent to open its profile. Here you can start or stop the agent, and see four panels: its system instructions (the standing brief that defines how it behaves, which you can edit), its skills (the playbooks it follows), its memory (what it has learned across sessions, including your corrections), and its secrets (encrypted credentials it can use). Custom agents also have a remove button here.

Editing system instructions is the most direct way to permanently change an agent's behavior. If you keep correcting the same habit in chat, put the rule in the instructions instead.