Architecture map
How the repo is organized
Click systems in the graph to inspect responsibilities, important files, and nearby dependencies.
Diagram quality
Generated draft: use it for orientation, then verify components and edges in source before relying on the layout.
Layout checks
- - 22 node overlap risks detected.
- - 4 likely edge crossings may reduce readability.
Click a system in the map
Explore arklexai/arksim by selecting services, gateways, clients, and data stores. Each node shows the files, responsibilities, and neighboring systems an agent should keep in scope.
Pan, zoom, and use the minimap to keep large repos navigable.
Sign up to turn selected graph context into repo-aware agent sessions.
Key systems
What an agent should understand first
Command-line entry point for simulate, evaluate, init, ui, and setup-claude commands
Interactive browser-based control plane for job management and result viewing
YAML parsing, agent/simulation config validation, CLI override merging
FastAPI routes for simulation, evaluation, filesystem, and results endpoints
Multi-turn conversation orchestration with synthetic user LLM and agent execution
Creates agent instances from config: custom Python, Chat Completions HTTP, or A2A protocol
Scores conversations with quantitative metrics, detects errors, matches trajectories
Abstraction for OpenAI, Anthropic, Google LLM APIs (synthetic user and evaluator)
Evidence coverage
What this context is grounded in
Coverage
- Citations
- 7
- Findings
- 5
- Indexed entities
- Unknown
- Relationships
- Unknown