DLEAPP parses artifacts left behind by desktop applications — the logs, events, and stored data of Electron/Chromium-based desktop apps: IndexedDB/LevelDB stores (including protobuf-encoded values), Local Storage, service-worker and HTTP caches, cookies, and application logs. It is a member of the LEAPP family, built on the RLEAPP framework.
DLEAPP is also meant to be a home for parsers that don't fit neatly into any of the other LEAPPs — a place for desktop-application and other odds-and-ends artifacts to live rather than being forced into iLEAPP, ALEAPP, RLEAPP, and the like.
| Application | What is parsed |
|---|---|
| Wire (desktop) | Accounts, devices, conversations, messages, calls, attachments, cookies, service-worker cache, and media recovered by decrypting cached asset blobs. |
| Discord (desktop) | Messages, attachments and recovered media, servers, channels, users, searches, reactions, message drafts, client activity, channel navigation, gateway sessions, account and application details, and a full cache index. |
| Signal (desktop) | Messages, attachments decrypted from disk, conversations and groups, calls, reactions, protocol sessions and identity keys, and account details. Requires the database credential — see below. |
Discord Desktop keeps no message database of its own: the client renders from REST API responses, and those responses stay in the Chromium HTTP cache. The Discord artifacts read that cache directly, so messages, attachments and the images themselves are recoverable after they were deleted server-side. The approach follows Alex Caithness's work on treating a web app's browser artifacts as an application in their own right (browser-forensics-presentation-2025, mister-skinnylegs).
The Chromium container formats these parsers rely on live in scripts/chromium/
(Simple Cache reader, Local Storage LevelDB reader) and are reusable by any
future Electron application parser. More desktop-application parsers will be
added over time.
Signal encrypts its message database with SQLCipher, and encrypts every file in
attachments.noindex with a key held inside that database. Recent versions wrap
the database key with the OS credential store, so it is not in the profile:
- macOS — login Keychain, service
Signal Safe Storage - Windows — Credential Manager
Capture it from the host and pass it in. DLEAPP accepts the credential, the 64 character database key itself, or a file holding either:
python3 dleapp.py -t fs -i <profile> -o <output> --signal-key
Given the flag with no value it prompts without echo, which keeps the secret out
of shell history and the process list. The GUI has an equivalent Signal key
field. A file named signal_password.txt beside the extraction is also picked
up, which suits batch runs. Older profiles that still hold a plaintext key in
config.json need nothing at all.
On a dead-box macOS image there is no host to read the credential from, but the
login.keychain-db is in the extraction. Supply the account's login
password in the same place, and if a login.keychain-db is present DLEAPP
recovers the Signal Safe Storage credential from it offline
(scripts/macos_keychain.py) and unwraps the key with no external tooling. The
same field therefore accepts either the safeStorage credential or the login
password — DLEAPP tries the credential interpretation first, then the keychain.
If the database was already decrypted, with DB Browser for SQLCipher or another tool, DLEAPP detects that and parses it as it is — no credential needed.
Without a credential, and with the database still encrypted, the Signal
artifacts report why and produce no rows rather than failing silently. scripts/sqlcipher_decrypt.py is the same pure-python
reader ALEAPP and iLEAPP use, so it needs no native SQLCipher build.
If you want to contribute hit me up on twitter: https://twitter.com/AlexisBrignoni
Download a pre-built release, no Python installation required, from DLEAPP GitHub Releases or LEAPPs Releases.
| Platform | Download |
|---|---|
| Windows (Intel/AMD) | DLEAPP-*-windows-x64-portable.zip |
| Windows (ARM) | DLEAPP-*-windows-arm64-portable.zip |
| macOS (Apple Silicon) | DLEAPP-*-macos-arm64.dmg |
| macOS (Intel) | DLEAPP-*-macos-x64.dmg |
| Linux (Intel/AMD) | DLEAPP-*-linux-x64.AppImage |
| Linux (ARM) | DLEAPP-*-linux-arm64.AppImage |
Each download holds one program, dleapp. SHA256SUMS.txt in each release lets you check a download.
GUI: open DLEAPP the usual way: by double-clicking dleapp.exe from the zip on
Windows, DLEAPP in Applications on macOS, or the AppImage on Linux. Started without
arguments, it opens the window.
CLI: give dleapp arguments in a terminal and it runs as a command line instead. The
output folder must already exist. On Windows, use dleapp.exe from the zip.
dleapp.exe -t fs -i C:\path\to\extraction -o C:\path\to\output\
On Linux, run the AppImage with the same arguments. On macOS it is inside the app; to
type just dleapp in a terminal, link it onto your PATH once:
sudo ln -s /Applications/DLEAPP.app/Contents/MacOS/dleapp /usr/local/bin/dleapp
Python 3.9 or above (older versions of 3.x will also work with the exception of one or two modules)
Dependencies for your python environment are listed in requirements.txt. Install them using the below command. Ensure the py part is correct for your environment, eg py, python, or python3, etc.
py -m pip install -r requirements.txt
or
pip3 install -r requirements.txt
To run on Linux, you will also need to install tkinter separately like so:
sudo apt-get install python3-tk
To install dependencies offline Troy Schnack has a neat process here: https://twitter.com/TroySchnack/status/1266085323651444736?s=19
To read Apple Unified Logs (the tracev3 data in a Mac extraction) when running from
source, DLEAPP needs Mandiant's unifiedlog_iterator.
python admin/scripts/fetch_unifiedlog_iterator.py downloads the pinned release for your
platform into bin/ after checking its SHA-256; see bin/PROVENANCE.md.
$ python dleapp.py -t <zip | tar | fs | gz | raw> -i <path_to_extraction> -o <path_for_report_output>
raw reads a disk image (.img, .dd, .bin, or any numbered .001 segment of
a split set), or an acquisition and the segments or files beside it (EnCase/EWF
.E01, SMART .s01, EWF2 .Ex01, AFF .aff, AFM .afm, any .aff in an AFD folder,
AFF4 .aff4, an Apple .dmg, with any .dmgpart files beside it, .sparseimage or
.sparsebundle folder, or a virtual machine disk, .vhd, .vhdx, .vmdk or .qcow2), in
place: no mounting and no administrator rights. Its NTFS, FAT32, FAT16, FAT12, exFAT, ext2/3/4,
F2FS, HFS+, APFS, QNX6, QNX4, ETFS, EFS, SquashFS, JFFS2, UBI/UBIFS, YAFFS and QNX IFS
volumes are searched directly (and a U-Boot environment or Belkin NVRM store is read as one
file), and
only the files an artifact asks for are read out of the image. Logical evidence, an
EnCase .L01 or an FTK Imager .ad1, is read as the files it holds. The GUI picks
raw on its own for those extensions, and for a sparse bundle or AFD folder chosen with
its folder button. See admin/docs/raw_image_input.md.
An encrypted image opens with its password (--image_password_file or
--image_password_env), or, when it is sealed to a certificate, with that certificate's
RSA private key (--image_private_key). A BitLocker volume in an image opens with its
password or recovery password, given the same way, or its startup key
(--bitlocker_key, repeatable), an APFS volume macOS encrypted in software with its
password or personal recovery key, and a LUKS container with its passphrase, each given
the same way. At a terminal whatever is missing is asked for, and the GUI asks in dialogs;
a BitLocker, APFS or LUKS volume nothing opens is reported and not searched.
The disk images of a Linux software RAID set (md, or LVM2 across disks) are read
together: give them all to -i with -t raw, or pick them together in the GUI's file
dialog. Given one image that holds a member, the other members are looked for in its
folder by what each file records about itself, not by its name; the command line lists
what it found and what it left out and why before reading the set (--no-find-members
reads only the image given), and the GUI asks before reading files it found. The report
names every image read and marks the ones found automatically. One image that holds an
md member is read with its arrays too: a lone RAID1 member with metadata 1.1 or 1.2
now reads its filesystem through the array, and its volume names gain the image's or
the array's name in front. One disk given twice is read once. An image in which no
RAID metadata is found is read as before; see admin/docs/raw_image_input.md for what
qnxprobe 1.71 does with DDF, IMSM and md member images read alone.
tar also reads an xz-compressed tar (.tar.xz), and the GUI picks tar for that
extension. A compressed tar, .tar.gz included, is decompressed once into the report folder
before any file is read, so the run needs free space there for the uncompressed tar. The
copy is deleted when the run ends, and the run log says how long the step took.
-m <profile> runs only the artifacts listed in a DLEAPP profile file, and -c <folder>
creates one. Profiles are saved as .dlprofile. Profiles saved as .rlprofile by
v2026.4.1 and earlier still load, in the CLI and in the GUI.
$ python dleappGUI.py
$ python dleapp.py --help
packaging/build.py builds dleapp with PyInstaller for the machine it runs on, from the
same virtual environment. Fetch the Unified Log parser first, or the build ships without
native Unified Log support:
python admin/scripts/fetch_unifiedlog_iterator.py # bin/unifiedlog_iterator, digest-verified
python packaging/build.py exe # dist/DLEAPP/, and dist/DLEAPP.app on macOS
python packaging/build.py smoke # run what it built, without opening a window
python packaging/build.py installer # Windows: Inno Setup installer; macOS: .dmg; Linux: AppImage
exe --onefile makes dist/dleapp (dist\dleapp.exe on Windows) as a single file
instead. The Windows installer needs Inno Setup; on
Linux, smoke needs a display, which xvfb-run provides. python packaging/build.py --help
has the rest.
Artifact modules live in scripts/artifacts/ and are loaded each time DLEAPP runs.
New modules: start with How to Write an iLEAPP Module. It uses iLEAPP as its reference, and DLEAPP modules use the same __artifacts_v2__ block and @artifact_processor decorator. The helpers are not identical across the tools, so check that one exists in DLEAPP's scripts/ilapfuncs.py before you import it.
The helper functions a module calls to find files, read SQLite, check in media and convert timestamps are described in Helper Functions for Artifact Modules.
A module has two parts: an __artifacts_v2__ dictionary at the top of the file, and one function per artifact decorated with @artifact_processor. For example:
__artifacts_v2__ = {
"get_cool_messages": {
"name": "Cool App Messages",
"description": "Messages from the Cool App database",
"author": "@username",
"creation_date": "2026-10-14",
"last_update_date": "2026-10-14",
"requirements": "none",
"category": "Cool App",
"notes": "",
"paths": ('*/CoolApp/cool.db*',),
"output_types": "standard",
"artifact_icon": "message",
}
}
from scripts.ilapfuncs import (
artifact_processor, convert_unix_ts_to_utc, get_file_path, get_sqlite_db_records)
@artifact_processor
def get_cool_messages(context):
source_path = get_file_path(context.get_files_found(), 'cool.db')
if not source_path:
return (), [], ''
query = 'SELECT sent_at, sender, body FROM messages'
data_list = []
for row in get_sqlite_db_records(source_path, query):
data_list.append((convert_unix_ts_to_utc(row[0]), row[1], row[2]))
data_headers = (('Sent', 'datetime'), 'Sender', 'Message')
return data_headers, data_list, source_pathThe key of each entry (get_cool_messages above) must match the name of its decorated function. That is how the loader and the decorator find each other. The entry holds:
name: the artifact name shown in the report. It must be unique within DLEAPP; the loader stops on a duplicate.description,author,notes: text about the artifact.notescan be empty.creation_date,last_update_date: dates inYYYY-MM-DDform. Both are written to the LAVA output. Bumplast_update_datewhen you change the module.requirements: a note on anything the artifact needs, or"none".category: groups artifacts in the report.paths: a tuple of patterns for the files the artifact reads. They are matched with Python'sfnmatch, so*can cross/.output_types:"standard"(every output except KML),"all", or a list taken fromhtml,tsv,timeline,lavaandkml.artifact_icon: a Tabler icon name that appears inscripts/_elements/tabler-icons.css.
Optional keys:
html_columns: columns written to the HTML report without escaping. Pass any value that comes from the evidence through the helpers inscripts/html_safe.py.data_views: alternate LAVA views, such as a conversation view for message artifacts. Column order for those is in Conversation Column Order.sample_data: a mapping of test corpus name to a short note about what that corpus produced, for example{"discord_macos": "Discord 0.0.402 macOS | 12940 rows"}. The artifact processor ignores it; it records where the artifact has been run. See Test corpora andsample_data.
The function takes one argument, context, which gives it the files its paths matched (context.get_files_found()) and the rest of the run. It returns three things: the column headers, a list of row tuples, and the path of the file the rows came from. The decorator writes the HTML, TSV, timeline, KML and LAVA output from them, as output_types asks. An empty list writes no table.
A header can be a tuple that names a column type: ('Sent', 'datetime'), ('Day', 'date'), ('Number', 'phonenumber') or ('Photo', 'media'). A plain string is a text column.
Corpora live outside this repository, because sample images are usually private. A corpus directory carries a samples.json registry:
{
"version": 1,
"samples": {
"corpus_name": {
"match": { "zip": "relative/path.zip", "sha256": "..." },
"platform": "macos",
"os_version": "macOS 26.5.2 (build 25F84)",
"app_versions": { "discord": "0.0.402" },
"notes": "how the capture was made and what it is good for"
}
}
}The keys in that registry are what artifacts cite in sample_data. admin/scripts/validate_sample_data.py keeps the two in step:
python3 admin/scripts/validate_sample_data.py # structure only
python3 admin/scripts/validate_sample_data.py --registry <path>/samples.json # + keys resolve, corpora present
python3 admin/scripts/validate_sample_data.py --registry <path> --verify-hashes
python3 admin/scripts/validate_sample_data.py --registry <path> --run <corpus> # + re-parse and diff row counts
The structural check needs no test data and runs in CI on every pull request. The registry and row-count checks need the images, so run those locally before changing a parser's output.
sample_data records how many rows an artifact produced. To catch a change that
keeps the count and alters the values, record a fingerprint of a corpus and
compare against it later:
python3 admin/test/scripts/make_test_data.py --registry <path> --corpus <key>
python3 admin/test/scripts/test_module_output.py --registry <path> --corpus <key>
python3 admin/test/scripts/test_module_output.py --registry <path> --all
Encrypted corpora take their secret the same way a normal run does, and
--secret signal with no value prompts without echo:
python3 admin/test/scripts/test_module_output.py --registry <path> \
--corpus signal_macos_needed --secret signal
Baselines live in admin/test/results/<corpus>.json and are committed. They
hold row counts, the column list, per-column digests and how many values were
populated — never rows, because the corpora are private application
profiles. A digest still changes when any value does, so the regression is
caught without the baseline carrying anyone's messages.
These need the corpora, so they run locally rather than in CI.
For example:
__artifacts_v2__ = {
"cool_artifact_1": {
"name": "Cool Artifact 1",
"description": "Extracts cool data from database files",
"author": "@username",
"version": "0.1",
"date": "2022-10-25",
"requirements": "none",
"category": "Really cool artifacts",
"notes": "",
"paths": ('*/com.android.cooldata/databases/database*.db',),
"function": "get_cool_data1"
},
"cool_artifact_2": {
"name": "Cool Artifact 2",
"description": "Extracts cool data from XML files",
"author": "@username",
"version": "0.1",
"date": "2022-10-25",
"requirements": "none",
"category": "Really cool artifacts",
"notes": "",
"paths": ('*/com.android.cooldata/files/cool.xml',),
"function": "get_cool_data2"
}
}The functions referenced as entry points in the __artifacts__ dictionary must take the following arguments:
- An iterable of the files found which are to be processed (as strings)
- The path of DLEAPP's output folder(as a string)
- The seeker (of type FileSeekerBase) which found the files
- A Boolean value indicating whether or not the plugin is expected to wrap text
For example:
def get_cool_data1(files_found, report_folder, seeker, wrap_text):
pass # do processing herePlugins are generally expected to provide output in DLEAPP's HTML output format, TSV, and optionally submit records to
the timeline. Functions for generating this output can be found in the artifact_report and ilapfuncs modules.
At a high level, an example might resemble:
__artifacts_v2__ = {
"cool_artifact_1": {
"name": "Cool Artifact 1",
"description": "Extracts cool data from database files",
"author": "@username", # Replace with the actual author's username or name
"version": "0.1", # Version number
"date": "2022-10-25", # Date of the latest version
"requirements": "none",
"category": "Really cool artifacts",
"notes": "",
"paths": ('*/com.android.cooldata/databases/database*.db',),
"function": "get_cool_data1"
}
}
import datetime
from scripts.artifact_report import ArtifactHtmlReport
import scripts.ilapfuncs
def get_cool_data1(files_found, report_folder, seeker, wrap_text):
# let's pretend we actually got this data from somewhere:
rows = [
(datetime.datetime.now(), "Cool data col 1, value 1", "Cool data col 1, value 2", "Cool data col 1, value 3"),
(datetime.datetime.now(), "Cool data col 2, value 1", "Cool data col 2, value 2", "Cool data col 2, value 3"),
]
headers = ["Timestamp", "Data 1", "Data 2", "Data 3"]
# HTML output:
report = ArtifactHtmlReport("Cool stuff")
report_name = "Cool DFIR Data"
report.start_artifact_report(report_folder, report_name)
report.add_script()
report.write_artifact_data_table(headers, rows, files_found[0]) # assuming only the first file was processed
report.end_artifact_report()
# TSV output:
scripts.ilapfuncs.tsv(report_folder, headers, rows, report_name, files_found[0]) # assuming first file only
# Timeline:
scripts.ilapfuncs.timeline(report_folder, report_name, rows, headers)A PR that adds or changes an artifact is easiest to review and merge when it arrives with
two things: a small test fixture cut from a real extraction, and sample_data values that
record what the module produced. Scripts generate both. Here is the whole flow.
One rule before anything else: whatever you commit here becomes public. Only use data you are allowed to share, like a test device you populated yourself, a public research image, or a file you sanitized by hand. Never casework.
1. Cut a fixture from your extraction
python admin/test/scripts/make_test_data.py <module> --case 1 --input <extraction.zip>
This pulls the files your module's paths patterns match out of the extraction and writes
the case file admin/test/cases/testdata.<module>.json plus one small zip per artifact
under admin/test/cases/data/<module>/.
Size rules: under 10 MB per zip, commit it with the PR. Between 10 and 25 MB, commit the case file and attach the zip to a PR comment. Bigger than that, say so in the PR and a maintainer will arrange a handoff.
2. Record the expected output
TZ=UTC python admin/test/scripts/test_module.py <module> -a all -c all
This runs the module against the fixture and writes a snapshot of the output under
admin/test/results/<module>/. Commit the snapshot too. It becomes the baseline that
guards the module after merge. Keep the TZ=UTC part: the committed snapshots are UTC
and CI runs UTC.
3. Run the same comparison CI will run
python admin/test/scripts/run_test_cases.py --module <module>
4. Generate the sample_data values
python admin/scripts/validate_sample_data.py --emit <extraction.zip> --key <image_name>
This runs DLEAPP end to end on your extraction and prints ready-to-paste sample_data
blocks for the modules changed on your branch. Paste them into your module's
__artifacts_v2__ and add the app name and version you saw on the image. If a count is
zero, check the source file really is empty before recording it.
5. Commit it all and open the PR
Commit the module, the case file, the fixture zips, and the recorded snapshot together. More detail lives in admin/docs/testing/create_module_test_cases.md.
If your extraction cannot be shared, open the PR anyway and say so. A fixture can often be cut from a public research image instead, or the real file can be sanitized by hand. The review does not stop while we work that out.
Free code signing provided by SignPath.io, certificate by SignPath Foundation.
The Windows builds are not signed yet. DLEAPP is applying to SignPath Foundation's free code signing program for open source projects, and this policy applies to every Windows release signed under it.
The Windows executables attached to
GitHub Releases are built from this
repository by GitHub Actions, on GitHub-hosted runners
(release.yml). Under this policy, SignPath signs only
what that workflow built. The macOS disk images are signed separately, with an Apple
Developer ID, and notarised by Apple.
SignPath signs dleapp.exe, the single-file program in the portable zip. Third-party files
packed inside it, such as Mandiant's Unified Log parser (unifiedlog_iterator.exe), are not
signed on their own.
- Committers and reviewers: @abrignoni, @JamesHabben, @Johann-PLW, @stark4n6
- Approvers: @abrignoni, @Johann-PLW
Changes proposed by people who are not committers are reviewed by a committer before they are merged. Every release signing request is approved by an approver.
DLEAPP processes extractions locally, on the machine it runs on. This program will not transfer any information to other networked systems unless specifically requested by the user or the person installing or operating it.
The same policy covers the other LEAPPs: leapps.org/releases#code-signing-policy.
This tool is the result of a collaborative effort of many people in the DFIR community.
DLEAPP logo courtesy of Kevin Pagano.
The earlier DLEAPP logo artwork was by Johann Polewczyk, with the per-OS window controls (Linux, Windows, macOS) suggested by James Habben.
DLEAPP is built on the RLEAPP framework by Alexis Brignoni and contributors.
Apple Unified Log .tracev3 data is read with the unifiedlog_iterator binary from
Mandiant's macos-UnifiedLogs, licensed under
Apache-2.0. Builds that include it carry its license and the notices for the Rust crates
it links; see bin/PROVENANCE.md.