How sourabhrajdev/atlas works
ATLAS is a local Mac AI assistant that turns user requests into tool-backed responses, while retaining memory and coordinating safety checks. Its broader documented capabilities include perception and a world model, retrieval, proactive signals, integrations, long-horizon planning, self-improvement, and specialist agents. The README supplies the high-level system relationships; sampled source confirms the CLI startup, engine dependencies, tool execution, and several subsystem internals. Many subsystem connections are documented rather than verified from their unsampled implementation.
Main entry points: orchestrator.py Orchestrator · engine.py LLM Engine · loop.py Autonomy Loop · cli.py CLI and Voice
Components 19
User Interaction
- Application Startup
atlas/main.py - CLI and Voice
atlas/interfaces/cli.py
Request Execution
- Orchestrator
atlas/core/orchestrator.py - LLM Engine
atlas/core/engine.py - Trust Layer
atlas/trust/layer.py - Tool Registry
atlas/tools/registry.py - Action Control
atlas/control/router.py
Knowledge Systems
- Persistent Memory
atlas/memory/store.py - Perception
atlas/perception/daemon.py - World Model
atlas/world/world_model.py - Production RAG
atlas/rag/retriever.py
Autonomous Systems
- Proactive Intelligence
atlas/proactive/engine.py - Autonomy Loop
atlas/autonomy/loop.py - Task Scheduler
atlas/scheduler/scheduler.py
Product Capabilities
- External Integrations
atlas/integrations/manager.py - Long-Horizon Planning
atlas/planning/engine.py - Multi-Agent Coordination
atlas/agents/coordinator.py - Self-Improvement
atlas/improvement/engine.py
Other components
- User
Connections 20
- User → CLI and Voice: submits request
README.md - Application Startup → CLI and Voice: starts CLI
atlas/main.py - CLI and Voice → Orchestrator: routes request
atlas/interfaces/cli.py - Orchestrator → LLM Engine: processes request
README.md - LLM Engine → Trust Layer: checks actions
atlas/core/engine.py - LLM Engine → Tool Registry: executes tools
atlas/core/engine.py - LLM Engine → Persistent Memory: stores responses
atlas/core/engine.py - Orchestrator → Perception: uses observations
README.md - Orchestrator → Long-Horizon Planning: coordinates planning
README.md - Orchestrator → Multi-Agent Coordination: routes agent tasks
README.md - Orchestrator → External Integrations: coordinates integrations
README.md - Orchestrator → Self-Improvement: coordinates learning
README.md - Orchestrator → Autonomy Loop: coordinates autonomy
README.md - Perception → World Model: updates context
README.md - World Model → Production RAG: supplies knowledge
README.md - External Integrations → World Model: informs knowledge graph
README.md - Autonomy Loop → Task Scheduler: checks scheduled tasks
atlas/autonomy/loop.py - Autonomy Loop → Tool Registry: executes tasks
atlas/autonomy/loop.py - Proactive Intelligence → Orchestrator: injects signals
atlas/proactive/engine.py - Trust Layer → Action Control: guards actions
README.md