Configuration

Sensible defaults, and the settings for when they don't fit.

Settings come from four places, each overriding the one before: built-in defaults, config.json in the state directory, environment variables, command-line options. The daemon reads them when it starts, so stop it and start it again after a change.

The state directory

--state, else GHIDRA_SESSION_HOME, else per user:

OS	Path
Windows	`%LOCALAPPDATA%\GhidraSession`
macOS	`~/Library/Application Support/GhidraSession`
Linux	`$XDG_STATE_HOME/ghidra-session` or `~/.local/state/ghidra-session`

It holds config.json, the endpoint and lock files, logs, and the class, AOT and upgrade caches. On POSIX it's created owner-only.

An example

config.json

{  "maxHeap": "16g",  "tmpdir": "D:/ghidra-tmp",  "idleTimeout": 7200,  "maxImports": 2,  "scriptPath": ["C:/re/scripts"]}

Check what the daemon will actually use with python $GS config.

Daemon

Key	Environment	Default	Meaning
`idleTimeout`	`GHIDRA_SESSION_IDLE_TIMEOUT`	`0`	Seconds without clients before the daemon exits; `0` stays up.
`keepWarm`	`GHIDRA_SESSION_KEEP_WARM`	`3600`	Seconds an unused read-only program or project stays open.
`maxQuick`		CPU count	Concurrent quick requests.
`maxLong`		CPU / 4, at least 2	Concurrent long requests.
`maxImports`		CPU / 4, 1 to 4	Concurrent imports, across all projects.
`scriptPath`		`[]`	Extra GhidraScript folders.
`log`		`<state>/<instance>.log`	The daemon's log.

JVM

Key	Environment	Default	Meaning
`java`	`GHIDRA_SESSION_JAVA_HOME`	discovered	JDK home.
`maxHeap`	`GHIDRA_SESSION_MAXMEM`	a quarter of RAM, 2-31 GiB	`-Xmx`. G1 takes only what it needs and returns idle memory.
`minHeap`		`512m`	`-Xms`.
`gc`	`GHIDRA_SESSION_GC`	`g1`	`g1`, `zgc`, `parallel` or `default`.
`vmargs`	`GHIDRA_SESSION_VMARGS`	`[]`	Extra JVM options, added last.
`tmpdir`	`GHIDRA_SESSION_TMPDIR`	the JVM's	Where Ghidra keeps unsaved database buffers. A fast disk with room.
`mode`	`GHIDRA_SESSION_MODE`	`fast`	`fast` or `standard` class loading.
`aot`	`GHIDRA_SESSION_AOT`	`auto`	Use the AOT cache when one exists, or `off`.
`classCache`		`true`	Cache Ghidra's extension-point scan.

Projects and caches

Key	Environment	Default	Meaning
`upgradeCache`	`GHIDRA_SESSION_UPGRADE_CACHE`	`true`	Serve older-format programs from an upgraded copy.
`upgradeCacheDir`	`GHIDRA_SESSION_UPGRADE_CACHE_DIR`	`<state>/upgrade-cache`	Where those copies live; best on the projects' volume.
`forkCopyLimitMb`	`GHIDRA_SESSION_FORK_COPY_LIMIT_MB`	`2048`	Largest full copy `fork_project` makes in `auto` mode.

Sizing memory

A large analysed program, such as a 1.5 GB project with 240,000 functions, needs a few GiB of heap once open. Several projects add up; status reports memory.usedMb and maxMb.
Each running import holds its whole program until it's saved: about 0.7 GB of heap per 25 MB PE.
Heaps of 32 GiB or more turn off compressed pointers, which costs memory; the default stops at 31.
On a Windows Dev Drive, put tmpdir on NTFS if you can: ReFS writes Ghidra's buffer files back more slowly.