Quickstart

This page tracks one run end to end on a local database, with no server and no extra packages. For remote logging see Log from a remote client.

Open the local database. It is created on first use.

using DearDiary
DearDiary.initialize_database()

Create a project and an experiment to hold the runs.

julia> project_id, _ = create_project("Iris classification")(id = "46d86f22-1e1c-425e-8190-f64c8af29e76", status = DearDiary.Created)
julia> experiment_id, _ = create_experiment(project_id, DearDiary.IN_PROGRESS, "Decision-tree sweep")(id = "d8232d05-77ec-4dc2-aae6-04e893ef2738", status = DearDiary.Created)

Open an iteration and log parameters and metrics. with_iteration marks the iteration SUCCEEDED on a clean return or FAILED on a throw, and snapshots the environment.

julia> with_iteration(experiment_id) do iter
           create_parameter(iter.id, "max_depth", 7)
           create_metric(iter.id, "accuracy", 0.96)
       end(id = "9c5f51a7-5ec2-48c2-94aa-9f6e5c73058a", status = DearDiary.Created)

Read the logged data back.

julia> iteration = last(get_iterations(experiment_id))Iteration(6980c2c9…, SUCCEEDED)
  id                   "6980c2c9-f4f0-430d-aee3-473de175ecfc"
  experiment_id        "d8232d05-77ec-4dc2-aae6-04e893ef2738"
  notes                ""
  created_date         2026-10-30T16:40:50.940
  end_date             2026-10-30T16:40:51.156
  parent_iteration_id  nothing
  status_id            2 (SUCCEEDED)
  error_message        ""
  julia_version        "1.13.3"
  git_sha              "27e16ee163b517ff48ad44d29d1d9da690ec7462"
  git_dirty            true
  entrypoint           "docs/make.jl"
  project_toml         "[deps]" (29 lines)
  manifest_toml        "# This file is machine-generated - editing it directly is n…" (1900 lines)
julia> get_parameters(iteration.id)1-element Vector{DearDiary.Parameter}: Parameter(max_depth = "7")
julia> get_metrics(iteration.id)1-element Vector{DearDiary.Metric}: Metric(accuracy = 0.96, step = 0)

A list prints one record per line, and a single record prints every field. Each list is also a Tables.jl table, so DataFrame(get_metrics(iteration.id)) or pretty_table(get_metrics(iteration.id)) work once DataFrames.jl or PrettyTables.jl is loaded.

See it in the browser

Start the server with DearDiary.run() and open http://127.0.0.1:9001. Every iteration gets its own page. Once a run has logged a few metrics per epoch, it looks like this:

An iteration page with notes, a tile and sparkline per metric, and step charts for train_loss and val_accuracy

Embedded web UI walks through the rest of the dashboard.