How aimhubio/aim works
Aim is a self-hosted ML experiment tracker: users instrument training to record run metadata, then query and compare runs through the SDK and web interface. Its main runtime spans framework integrations, run tracking, repository-backed persistence and indexing, an API server, and UI exploration; remote tracking, artifact storage, resource monitoring, and notifications extend that core. Relationships between sampled transport components are shown where the excerpts establish them; other links are documented by the README or left unverified where implementation was not sampled.
Main entry points: run.py Run Tracking · repo.py Repository · server.py Aim Server · adapters Framework Adapters
Components 21
Tracking SDK
- Framework Adapters
aim/sdk/adapters - Run Tracking
aim/sdk/run.py - Training Flow
aim/sdk/training_flow.py
Repository Data
- Repository
aim/sdk/repo.py - Run Indexing
aim/sdk/index_manager.py - Run Locks
aim/sdk/lock_manager.py - Artifact Storage
aim/storage/artifacts/artifact_storage.py - S3 Artifacts
aim/storage/artifacts/s3_storage.py
Serving and Remote
- Web API
aim/web/api/views.py - Aim Server
aim/ext/transport/server.py - Resource Handlers
aim/ext/transport/handlers.py - Remote Client
aim/ext/transport/client.py
Query and UI
- Query Analysis
aim/sdk/query_analyzer.py - Exploration UI
aim/web/ui/src/App.tsx - Notebook Access
aim/ext/notebook/notebook.py
Operations and Integrations
- CLI
aim/cli/cli.py - Tracker Conversion
aim/cli/convert/commands.py - Resource Tracking
aim/ext/resource/tracker.py - Notifications
aim/ext/notifier/notifier.py
Other components
- ML User
- S3 Service
Connections 22
- ML User → Framework Adapters: instruments training
README.md - Framework Adapters → Run Tracking: logs metadata
- ML User → Run Tracking: tracks runs
README.md - Run Tracking → Repository: stores run data
- Training Flow → Run Tracking: supports training
- Repository → Run Indexing: indexes run data
aim/sdk/index_manager.py - Run Tracking → Run Locks: uses run locks
- Artifact Storage → S3 Artifacts: implements storage
aim/storage/artifacts/s3_storage.py - S3 Artifacts → S3 Service: uploads artifacts
aim/storage/artifacts/s3_storage.py - ML User → CLI: starts tracker
README.md - CLI → Aim Server: starts server
- Aim Server → Resource Handlers: registers resources
aim/ext/transport/server.py - Resource Handlers → Repository: opens repository
aim/ext/transport/handlers.py - Remote Client → Aim Server: sends requests
- ML User → Exploration UI: explores runs
README.md - Exploration UI → Web API: queries API
- Web API → Repository: reads run data
- ML User → Query Analysis: queries metadata
README.md - ML User → Notebook Access: analyzes runs
README.md - ML User → Tracker Conversion: migrates tracker data
README.md - Run Tracking → Resource Tracking: tracks system usage
- Run Tracking → Notifications: alerts on progress