Dawn now includes an agent gallery so teams can create useful starter agents with recommended prompts, avatars, skills, and setup guidance instead of building every profile from scratch.
Dawn Team
Product Engineering
Dawn now includes an agent gallery. Instead of starting every workspace with a generic agent profile, owners and managers can choose ready-made agent templates and create useful specialised agents in a few clicks.
This is for the moment every new workspace hits quickly: you know you want Dawn to help, but you should not have to invent the first support agent, reviewer agent, testing agent, or knowledge agent from an empty form. The gallery gives teams a practical starting point and keeps the advanced configuration available for later.
The gallery presents templates for common jobs teams ask Dawn to handle: general workspace help, coding, code review, testing, support, knowledge work, and marketing. Each template comes with a recommended name, description, prompt direction, capability tags, and avatar.
That does not mean every workspace should use every template. A small engineering team might start with a code reviewer and a support guide. A product team might choose a knowledge agent and a marketing agent. A customer-facing workspace might start with a support agent and adjust the prompt around its own escalation rules.
The point is to make the first agent feel like a real assistant rather than a placeholder.
The gallery supports selecting multiple templates before creation. That is useful when setting up a workspace for the first time or expanding Dawn from one general agent into a small set of focused agents.
Focused agents are easier to trust. A review agent can be stricter about correctness. A support agent can be more careful about customer tone. A knowledge agent can stay closer to documentation and memory pages. A tester can think in edge cases and regression coverage. Keeping those responsibilities separate makes Dawn easier to operate than one all-purpose profile with too many jobs.
Gallery agents are starting points. After creation, each agent is still a normal Dawn agent profile. You can change the name, model, provider, avatar, prompt, skills, memory access, channels, and connected sources.
That matters because a useful starter prompt should not pretend to know your entire workspace. It should give you a sensible base and make the next edit obvious. If the support agent should only answer from approved memory pages, configure that. If the reviewer should only see GitHub pull requests and not customer conversations, assign those sources deliberately. If the marketing agent should use Google Analytics and Search Console, connect and assign those integrations.
The gallery gets you out of the blank-page state. Workspace configuration still belongs to the team.
The gallery is especially helpful when a workspace has no agents yet. Dawn can open the gallery during setup so owners can create a useful first profile before inviting the rest of the team.
That improves onboarding in a simple way: instead of showing an empty workspace and asking users to understand every setting up front, Dawn offers recognizable roles. Pick the ones that match how the team works, then refine them once real conversations show what needs adjustment.
The gallery also gives each starter agent a visual identity. That sounds cosmetic, but it matters in daily use. Once a workspace has several agents, avatars make it easier to scan the list and pick the right profile quickly.
They also work with the broader avatar support in Dawn. If the starter avatar is not right for your team, replace it with an uploaded image, an animated GIF, or a generated spritesheet using /avatar.
Open Dawn Web and go to Agents. Click New Agent to open the gallery, choose one or more templates, and create the agents you want. After they are created, open each profile and assign the model, sources, skills, channels, and memory access that match the role.
The fastest path is to start narrow. Create one or two agents for the work your team repeats most often, use them for real requests, then add more specialized agents when the need is obvious.