Agent system / Overview
Architecture project
Owned workspace index
Your architectures
Open any architecture you own, or begin with a clean workspace.
No architectures yet
Create your first real architecture. Nothing has been generated or opened for you.
Your agent system, grounded in evidence.
Design architecture gates, connect research, and validate decisions before your next submission.
Architecture gates
Core decisions that shape your agent system
Research suggestions
Generated from your research library
Model comparison and architecture runs
Choose exactly one execution model for each custom prompt. Every output or provider error is saved to this project.
Enter a custom prompt to request a provider-backed architecture response. No sample output is substituted when a provider is unavailable.
System design
Architecture gates
Make every architectural choice explicit, testable, and traceable to a citation.
Decision map
Instruction design
Prompt evidence
Attach evidence to each prompt fragment so instructions stay defensible.
System prompt
● Backend persistedEvaluation
Sandbox scenarios
Run sample checks against realistic edge cases before review.
Backend-loaded evidence base
Research library
Curated reference data loaded from the Atlas backend; sources are not presented as user-owned research.
Change-aware review
System review
Review what changed since the last sign-off, then record a new decision.
Persisted review history
Backend-backed builder
Architecture agent
Chat with the configured model, or invoke explicit tools whose results are persisted against this architecture.
Agent conversation
Messages and provider errors persist on reload.