Local / Remote / File Modes

The SDK supports two backends and three storage kinds.

Local mode (default)

Without a host, Tracker writes directly through PyO3 to local storage:

from trailer import Tracker
t = Tracker(project="demo")

• Zero external dependencies, zero network overhead
• Data lands in local SQLite (trailer.db) or file mode
• Runs belong to the admin user

Remote mode

Set a host (or TRAILER_HOST) to write to a running server:

import os
os.environ["TRAILER_HOST"] = "http://team-server:5120"
os.environ["TRAILER_TOKEN"] = "rt_xxxx"   # API token from /profile

t = Tracker(project="demo")               # data goes to the server

• Batches are msgpack/HTTP encoded to /api/v1/ingest
• Runs belong to the token’s account

Storage backends

Storage	How to trigger	Best for
SQLite	default	Single machine
File mode	storage="file", data_dir="data"	TensorBoard-style, versionable JSON

t = Tracker(project="exp", storage="file", data_dir="data")

Or via environment: TRAILER_STORAGE=file TRAILER_DATA_DIR=data.

Which mode to pick

Scenario	Mode
Solo local experiments	Local mode (zero config)
Team, centralized data	Remote mode against a shared server
Reproducible, backup-friendly runs	File mode
Multi-user with access control	Server mode + Remote SDK