Skip to content

Configuration

From lowest to highest priority:

  1. Built-in defaults.
  2. A TOML file: <data_dir>/dogubako.toml, or the file given with --config or DOGUBAKO_CONFIG. Start from deploy/dogubako.example.toml.
  3. Environment variables: DOGUBAKO_ plus the key in upper case, for example max_upload_mb is DOGUBAKO_MAX_UPLOAD_MB.
  4. Command-line flags: --bind, --data-dir, --config, --public-url, --no-auth.

dogubako server config prints the effective configuration as TOML, which is the quickest way to check what the server will use.

# <data_dir>/dogubako.toml
bind = "0.0.0.0:8420"
public_url = "https://tools.example.com"
instance_name = "Home toolbox"
secure_cookies = true
max_upload_mb = 4096
file_retention_hours = 48
sandbox = "bwrap"

Every setting is listed in the configuration reference.

A few trusted people, files kept for a couple of days:

Terminal window
DOGUBAKO_INSTANCE_NAME="Home toolbox"
DOGUBAKO_FILE_RETENTION_HOURS=72
DOGUBAKO_USER_QUOTA_MB=20480

Many agents, mostly short jobs, tight isolation:

Terminal window
DOGUBAKO_MCP_DISCOVERY=compact # agents search for tools instead of listing all 249
DOGUBAKO_PER_USER_CONCURRENCY=1
DOGUBAKO_USER_QUOTA_MB=2048
DOGUBAKO_FILE_RETENTION_HOURS=6
DOGUBAKO_SANDBOX=bwrap # native Linux installs

Few users, long CPU-heavy jobs:

Terminal window
DOGUBAKO_MAX_CONCURRENT_JOBS=2 # FFmpeg and whisper.cpp use several threads each
DOGUBAKO_MAX_JOB_SECONDS=14400 # four hours
DOGUBAKO_EXEC_CPU_SECS=14400
DOGUBAKO_EXEC_MEMORY_MB=8192
DOGUBAKO_MAX_UPLOAD_MB=8192

Keep the container or service memory limit above EXEC_MEMORY_MB × MAX_CONCURRENT_JOBS.

Let tools such as duplicate finders and folder size read your shares:

Terminal window
DOGUBAKO_FS_ROOTS=/srv/photos,/srv/documents

Then grant host_fs_read to the accounts that should use them. Paths are resolved (symlinks included) and must stay inside a root.

Point the server at a specific engine binary with DOGUBAKO_<ID>_PATH, for example DOGUBAKO_FFMPEG_PATH=/opt/ffmpeg/bin/ffmpeg. The ids are listed in Engines and models.