How rush86999/atom works
Atom is a self-hosted, governed AI agent workforce. A user delegates a task in plain language; Atom builds and runs a workflow, with approvals and scoped governance around actions, then checks outcomes and records activity. The platform also ingests documents and business communications into searchable memory, supports integrations and local or hosted LLMs, and includes finance and analytics capabilities. The README documents the main product workflow and safeguards; source excerpts expose parts of ingestion and persistence, but not the full request-to-execution path.
Main entry points: agent_execution_service.py Agent execution · frontend-nextjs Web workspace · hybrid_data_ingestion.py Hybrid data ingestion · atom_communication_ingestion_pipeline.py Communication ingestion
Components 21
User interaction
- Web workspace
frontend-nextjs - Interactive canvas
archive/2026-07-consolidation/source/src/components/canvas/AppCanvas.tsx
Agents and governance
- Workflow automation
backend/ai/automation_engine.py - Agent execution
backend/core/agent_execution_service.py - AI trigger coordinator
backend/core/ai_trigger_coordinator.py - Governance and safety
backend/core/governance/policy_engine.py - Outcome verification
backend/independent_ai_validator/core/validator_engine.py - Human approvals
backend/api/approvals_routes.py
Data and memory
- Data ingestion API
backend/api/data_ingestion_routes.py - Document ingestion API
backend/api/document_ingestion_routes.py - Knowledge ingestion
backend/core/ingestion_pipeline.py - Hybrid data ingestion
backend/core/hybrid_data_ingestion.py - Communication ingestion
backend/integrations/atom_communication_ingestion_pipeline.py - Vector memory
backend/core/lancedb_handler.py
Integrations
- Gmail integration
backend/integrations/gmail_service.py - Ingestion webhooks
backend/api/routes/webhooks/ingestion_webhooks.py
Platform services
- Application database
backend/core/database.py - LLM provider credentials
backend/core/llm_oauth_handler.py - Finance operations
backend/accounting - Analytics
backend/analytics
Other components
- Business user
Connections 20
- Business user → Web workspace: delegates work
- Web workspace → Workflow automation: describes workflow
- Workflow automation → Agent execution: runs workflow
- Agent execution → Governance and safety: enforces boundaries
- Agent execution → Human approvals: requests approval
- Agent execution → Outcome verification: checks outcomes
- AI trigger coordinator → Agent execution: coordinates triggers
backend/core/ai_trigger_coordinator.py - Business user → Data ingestion API: configures sync
backend/api/data_ingestion_routes.py - Data ingestion API → Hybrid data ingestion: starts sync
backend/api/data_ingestion_routes.py - Document ingestion API → Knowledge ingestion: syncs documents
- Hybrid data ingestion → Vector memory: writes vectors
backend/core/hybrid_data_ingestion.py - Knowledge ingestion → Vector memory: stores knowledge
backend/core/ingestion_pipeline.py - Ingestion webhooks → Communication ingestion: triggers ingestion
backend/api/routes/webhooks/ingestion_webhooks.py - Gmail integration → Communication ingestion: feeds communications
backend/integrations/gmail_service.py - Communication ingestion → Vector memory: indexes messages
backend/integrations/atom_communication_ingestion_pipeline.py - Hybrid data ingestion → Application database: reads and writes
backend/core/hybrid_data_ingestion.py - LLM provider credentials → Application database: stores credentials
backend/core/llm_oauth_handler.py - Web workspace → Interactive canvas: presents workspace
- Business user → Finance operations: uses finance tools
- Business user → Analytics: reviews analytics