Agents, Memory, and Scheduled Tasks
Last reviewed: 2026-03-18
Owner: Product + support
Source of truth:
https://dawnhq.ai/docs/agents-memoryhttps://dawnhq.ai/docs/features/agentshttps://dawnhq.ai/docs/features/memoryhttps://dawnhq.ai/docs/features/skillshttps://dawnhq.ai/docs/features/scheduled-taskshttps://dawnhq.ai/docs/codex
Applies to: behavior consistency and automation guidance
Refresh cadence: monthly
Facts
- Agent profiles should have a clear role, scope, style, provider/model choice, and prompt instructions.
- Codex is available after the user links a ChatGPT account and the provider/model becomes selectable.
- Skills are reusable instruction packs and should be scoped and prioritized intentionally.
- Memory pages are for durable facts, runbooks, and policies; avoid storing secrets or ephemeral chat noise.
- Dawn memory pages can be tagged for organization.
- Scheduled tasks support both one-time and recurring automations with result delivery into the target conversation or channel.
FAQ snippets
- Why is behavior inconsistent? Usually prompt ambiguity, conflicting skill priorities, or excessive source/tool scope.
- What belongs in memory pages? Stable operational knowledge, not temporary chat artifacts or secrets.
- Can I organize memory pages? Yes. Dawn supports tags on memory pages.
- Why scheduled tasks? To automate periodic checks, summaries, reminders, and recurring operational workflows.
Escalation trigger
- Agent behavior regresses after prompt/skill changes and cannot be stabilized with scope and prompt cleanup.