Agent Profiles

Agent profiles define model choice, behavior, and access scope. A clean profile strategy is the difference between reliable operations and noisy responses.

What to configure

  • Provider and model (OpenAI, Azure OpenAI, Anthropic, OpenRouter, Codex when linked).
  • System prompt with explicit role, scope, and output style.
  • Channel bindings that determine where this profile responds.
  • Source bindings that determine what this profile can read/search.
  • Skill assignments for reusable behavior rules.
  • Memory access mode: No memory access, All workspace memory, or Only pages matching allowed tags.
  • Allowed memory tags when the profile should only use part of workspace memory.

Use the Agent Gallery when you want a curated starting point instead of configuring every field from scratch.

Profile design tips

  • Prefer one profile per business workflow (support triage, release assistant, docs assistant).
  • Avoid “kitchen sink” profiles with too many channels and unrelated sources.
  • Keep prompts concise and operational; do not duplicate tool docs in prompt text.
  • Verify plan/model restrictions so selected model is allowed by billing policy.
  • Leave integration write tools off unless the profile should intentionally create or update data in connected systems.
  • Use tag-scoped memory when one workspace supports multiple teams, products, or support domains.

Memory access guidance

  • Choose No memory access for agents that should work only from live prompts and connected sources.
  • Choose All workspace memory for broad general-purpose agents that should see every approved page.
  • Choose Only pages matching allowed tags when the agent should stay inside a defined knowledge domain.
  • When tag-scoped access is enabled, use allowed tags that overlap with the tags on the pages you want the agent to read and maintain.

Validation checklist

  • Profile responds in expected channels and nowhere else.
  • Profile can only access intended sources.
  • Profile memory access matches the intended scope: none, all, or tag-scoped.
  • For tag-scoped profiles, only pages with overlapping allowed tags are available.
  • Model selector reflects linked providers (including Codex when enabled).
  • Changing profile model/provider takes effect immediately in chat tests.
Validation Prompts

Use these prompts in Dawn chat to quickly validate this setup area. 

Review this agent profile and suggest prompt + binding improvements.
Check whether selected model/provider is allowed by current plan.
Completion Checklist
  • Agent role and system prompt are clear.
  • Channel bindings match intended scope.
  • Source bindings are minimal and relevant.
  • Provider/model selection validated in live run.