Database forensics · Version 1.22.2
TecInt-DBlite
Secure, read and prepare databases from apps and browsers forensically, SQLite and 36 further formats. TecInt-DBlite detects the format itself, secures the files without changing them and makes the content readable. It shows how the data relates and brings back what an app overwrote or deleted, as far as it is still in the files. Every step is logged and can be verified later.
Free of charge for private individuals for private use. Organisations please ask before using it.
This page describes version 1.22.2 (24 October 2026). New releases appear in the news and in the feed for this product.
On this page 30 sections
More than a table viewer
Messages, calls, contacts, histories, settings: every app keeps its data in a database. Mostly that is SQLite, but often enough it is LevelDB, IndexedDB, Realm, Hive, MMKV or one of the other formats used by mobile apps and browsers.
Column names are cryptic, timestamps are numbers, content sits packed inside blobs, and the link between two tables is written down nowhere. TecInt-DBlite takes that work off your hands and documents every step along the way.
Nobody needs to know a file’s format beforehand. Whether a single file or a whole app folder, the program detects the format by its content and picks the matching reader. If an SQLite file belongs to an app format such as Couchbase Lite, it is evaluated that way on request, attachments included.
What TecInt-DBlite reads and what comes out
The original stays untouched
When opening a database, TecInt-DBlite secures it together with its WAL and journal files into a case folder and computes SHA-256 and MD5. All work happens on a write-protected copy.
- A case log records every step, its entries chained to each other by hash value
- On closing the program checks the original again
- Your own databases can be edited directly, evidence only ever as a copy
Thirty-six formats besides SQLite
Besides SQLite, TecInt-DBlite reads Android SharedPreferences, in binary form too, along with DataStore, Protocol Buffers and FlatBuffers, LevelDB with Local and Session Storage, IndexedDB from Chrome, Edge, Safari and Firefox, Hive, MMKV, LMDB, ObjectBox, Realm, RocksDB, Berkeley DB and Couchbase Lite. Added to those are the Bluetooth configuration of Android and Linux, CAN bus captures, flight logs, GPS tracks, FIT files, OBD-II captures, tachograph downloads and systemd journals.
The source is secured with hash values like any database. Its content is added to the case as a new SQLite database, with the origin recorded for every table.
- Old states stay readable: many of these formats overwrite nothing in place. A change is appended, a deletion merely noted. Every entry is marked as current, superseded or deleted.
- Couchbase Lite: deleted and purged documents come back from free areas of the file
- Berkeley DB: the remains of freed pages are read along with the rest
- Encrypted Hive and MMKV data opens with a known key, and the key is not stored
- Every reader was checked against files written by the original library or by a real browser; the exceptions so far are the Bluetooth reader and the readers for vehicle, flight and sports data, which were checked against real, publicly available recordings
- Randomly corrupted files are run against every reader regularly. A row that does not fit its table is kept separately in
_fehlerhafte_zeileninstead of being written in shifted.
Looking inside the files
For every format the program shows the layout of the source files: every area with its position, size and state, as a list and as a coloured strip from the start to the end of the file. In use, superseded, free, remnant or damaged is visible at a glance, and so are files that are missing according to the management file.
- The version history puts every readable change in order, by sequence, transaction or version numbers or by position in the file, together with the previous value
- From a table, one click leads to the history of a single key
- An order the source does not provide is not invented
- For LMDB it also reads the complete state of the previous transaction, for LevelDB and RocksDB a write that broke off at the end of the log
Berkeley DB down to the transaction log
If the whole folder of a Berkeley DB environment is available, TecInt-DBlite reads its transaction log as well. Every change is in there with its content.
- For each deletion the deleted value, for each replacement the old value
- For each confirmed transaction the time
- Changes that were never confirmed are marked and do not count towards the state
- This was checked against logs written by Berkeley DB itself
Relationships, even without foreign keys
Most apps declare no foreign keys. TecInt-DBlite finds the links anyway: through column names, by matching values and through the structures of Apple Core Data.
- A diagram of the relationships found, with a confidence score and the evidence for every link
- Empty references: where an entry points to a record that no longer exists, the spot is marked. Often that is a sign of deleted data.
- IDs become names: on request, a contact ID shows the contact's name.
Readable, not cryptic
Timestamps in Apple, Unix, Chrome or Windows format appear as date and time, and the program knows 19 formats in all. Even identifiers give away when they were created, such as UUIDs, ULIDs or the message IDs of Discord and X.
Where several readings are arithmetically possible, the program shows them side by side. Only an unambiguous one is applied.
- Packed content such as Base64, gzip, zstd, LZ4, LZString, Brotli, Snappy, MessagePack, CBOR, property lists, NSKeyedArchiver or protobuf is unpacked right in the table and the cell is marked. This also works for values packed a hundred times over and for JSON whose fields hold further encoded content, at any depth.
- URL encoding, HTML entities and quoted-printable are resolved as well
- A layer analysis with hexdump covers the rest. It recognises rare encodings, certificates and the content of ZIP archives.
- Values that were merely obscured: if an app only used XOR or a similarly simple key, the program finds the key itself, from the format’s signature with its checksum, from the statistics of the value or from the whole column, even when the same key stream was used over and over.
- Broken umlauts from old character sets are repaired
- File header and WAL with explanations, along with the frames of the WAL file
JSON, XML and HTML laid out for reading
Many apps put whole records into a single text cell as JSON or XML, and mail programs store HTML. TecInt-DBlite indents JSON and XML and also takes apart JSON that is stored once more as text inside a field.
HTML becomes readable text with paragraphs and links, and an address becomes the list of its parameters with the search term decoded. This applies in the cell panel and in the prepared output, where the stored value is kept alongside.
- XML is checked beforehand for entity attacks
- The original value stays visible, so the prepared form can be held against what was stored at any time
Opening encrypted values
Some apps store individual fields encrypted. If you know the key from another source, from a keychain, a configuration file or the app's own source code, TecInt-DBlite opens the value.
- AES and ChaCha20, with the nonce either at the start of the value or given separately
- The checksum of the algorithm shows whether the key was right
- The key itself is not stored. The case log holds only its fingerprint.
- Whole databases are classified from the file header alone: SQLCipher with its version, SQLite3 Multiple Ciphers, SEE or mere checksums
- With a known key a decrypted evidence copy is created in the case folder, while the encrypted backup stays untouched
Recovering deleted data in SQLite
What the other formats give up is described further above. In SQLite it looks like this: deleted rows are freed but often not overwritten. TecInt-DBlite reads the file byte by byte and reconstructs records from freeblocks, free pages, the WAL file and the rollback journal. Tables without rowid, tables that were dropped altogether and long content spread over several overflow pages come back as well.
- Every find is compared with the current state: deleted, changed or unchanged
- Where a find came from is shown too, so freeblock, freelist, WAL or journal
- Dropped tables are rebuilt from remnants of the schema, and the rows found are matched to their columns
- The finds can be written out as an SQLite database of their own, with the location and the confidence for every row
- What comes out is a pointer to an earlier version of the data. Why an entry disappeared is not something it can tell you.
Measured on public test data
How much a tool really recovers can be measured on public test data. TecInt-DBlite ships the measuring programs, so anyone can reproduce the figures.
- SQLite Forensic Corpus of the University of Erlangen-Nuremberg: TecInt-DBlite finds all 84 deleted rows of the comparison databases without inventing a single one
- CFReDS of the US standards institute NIST: all 100 deleted rows show up, with a rollback journal as well as with WAL
- Test data and measuring programs come with the source code. The figures hold for these data sets; a real database may look different.
The path of a row in SQLite
The version history puts the states of a row side by side: the journal, the main file, every transaction in the WAL, the remains of deleted versions and the current state. This shows, for instance, that a message was first rewritten and then deleted.
- For every state the place it was found, meaning page, offset and frame, with the changed values highlighted
- Identical states are folded together so the list stays readable
- A separate check looks for traces of cleaning and manipulation: areas overwritten with zeros, contradictions in the file header, gaps in consecutive keys or timestamps later than the time of securing. The evidence is shown with every hint.
- Clock against order: the check compares the time with an order that does not depend on the clock, such as sequence numbers or the position in a transaction log. Where time runs backwards there, the program names the spot. Often that is the trace of a device clock that was changed.
- A hint is a reason to look, not a proof. Whether a gap comes from ordinary use or from deliberate deletion is settled only by comparing it with the recovered records.
- Passing results on to other forensic tools in the CASE/UCO exchange format
What vanished when
Deleted rows leave traces in places one rarely thinks of. Indexes often keep the indexed values and the row number longer than the table itself. TecInt-DBlite reads them and gives a find whose row number was lost the right one back.
Apps with a search function keep a full-text index. It often holds the words of deleted messages longer than the table does. The program puts them back together in their order, for FTS3, FTS4 and FTS5.
- For every transaction in the WAL the program shows which rows vanished together. That makes visible how a whole chat went in one step and a single message later.
- For each group it names the likely reason: a whole chat or contact, a retention rule of the app that removes everything before a cut-off, or single rows by hand
- Where a transaction was never completed, the journal shows what it had already deleted or changed. A journal whose header SQLite zeroed after the commit is read anyway, together with older transactions behind it.
- Remains of
sqlite_masterreveal dropped tables, indexes and triggers and earlier column lists; the-shmfile serves as a second record of the WAL’s state - Pages of the database sitting in a page file, a memory image or in free space on a disk are found there and evaluated with the structure of the open database
- On top of that, samples from the index statistics, the time window in which a deleted row was created and, for Apple Core Data, the number of times a record was saved
What the values mean
Many apps store codes instead of plain text: a 1 for read, a 3 for a missed call, a 4097 for several features at once. At the push of a button TecInt-DBlite describes the whole database and tells for every column what kind it is, which values occur and what they probably stand for. Bit fields are split into single bits.
The description is produced entirely inside the program, without network access, and can go into the examination report.
- Every statement names its evidence, such as the column name, a relation in the data (“with value 3 the duration is always 0”), an expression in the schema or a value that changed since an earlier state
- For known databases such as the Android call log, messages, contacts, calendar and downloads, or the history of Chromium-based browsers, the meaning comes from a catalogue with source and version
- What you have reproduced yourself on a test device goes into a verification catalogue of your own, with your name, the date and the app version checked, and can be passed on with a checksum
- Documented codes then also appear right next to the raw value in the table, such as “3 · missed”
- When a status changes over time, the program follows each row through journal, main file and WAL. If a value jumps differently from usual, say from “done” back to “open”, the spot is listed among the anomalies.
Everything on one timeline
Calls, messages, events, downloads and page visits become a sequence of readable sentences. “Call · missed · +49 170 …”, followed by “SMS · inbox · ‘On my way.’”
- Deleted entries from recovery appear in red in between
- On request all databases of a case go in, each with its name
- The timeline can be filtered, saved as CSV or HTML and appears in the examination report
Contacts and identifiers at a glance
TecInt-DBlite collects phone numbers, e-mail addresses, web and IP addresses, IBANs, coordinates, Bitcoin and Ethereum addresses and card numbers from all tables, from BLOB content and, if wanted, from deleted remains.
- Different spellings are folded together: “0170 5550103” and “+49 170 5550103” count as the same number
- What occurs in the most tables comes first, so the number that turns up in contacts, messages and calls at once
- Checksums keep chance hits out
- Protobuf columns get a derived schema that names time fields, booleans and possible coordinates
A full SQLite editor as well
- Browse tables, filter and sort them
- SQL editor with highlighting and completion
- Structure of the database with columns, keys and row counts
- Tables and indexes can be created and rebuilt
- Import and export as CSV, SQL and JSON
- PRAGMAs with an explanation of every setting
- Charts drawn from the data
Many databases at once
An extraction with hundreds of databases? TecInt-DBlite finds every SQLite file in a folder, even without an extension, and every data store of the other formats. It secures them with hash values, converts what needs converting and ranks them by yield.
Nobody needs to know beforehand how a folder is organised. Drag an app folder with several databases in its subfolders onto the start page and the program lists them, three LMDB environments, one LevelDB and two SQLite files for instance. Each can be secured on its own, and where they all share one format they can be examined together.
- An overview with time ranges, time columns and hints about the content of each table
- One search for a name, phone number or keyword covers all tables of all databases at once
- Very large databases stay manageable too: the program makes no second copy, pages to the end of a long table without a wait and spreads recovery across several processor cores.
- An HTML report with case details, hash values, file header, relationships, recovered records and the verified case log. The file has no external content.
Every table in its own tab
An examination means jumping between messages, contacts and attachments all the time. In TecInt-DBlite every table you open gets a tab, and each tab keeps its state: filters, search, sort order, column widths and the spot where you left off.
- A project remembers the open tabs for next time, with the filter, sorting and search of every tab
- Every function found quickly: Ctrl+K opens a search across every menu entry, every area and the actions of the selected column. A keyword such as “time”, “csv” or “deleted” is enough.
- The “Column” button in the toolbar opens the column menu without a right-click
Even when someone has tampered with the schema
Anyone covering their tracks can edit a table's entry in the schema by hand until SQLite rejects the file or the table disappears. In the forensic view TecInt-DBlite opens such files anyway.
- SQLite then skips the faulty entries, nothing in the file changes, and the search for deleted records reads the pages by itself in any case
- Among the anomalies you find what SQLite reported and which schema entries can never have come from SQLite, such as a table name stored as a BLOB
Images and attachments
Photos, voice messages, videos and documents often sit in the database as BLOBs. TecInt-DBlite finds them in all tables, shows thumbnails and reads the capture data: device, capture time with time zone and the location with altitude, from JPEG, PNG, WebP, TIFF and HEIC.
- Every find has MD5, SHA-1 and SHA-256
- Your own hash lists hide known files or highlight wanted ones
- Saved files are hashed again after writing, and everything goes into the case log
Infotainment in vehicles
When someone pairs a phone with a car, phone book, call history and messages also end up in the head unit. On Android Automotive, TecInt-DBlite recognises this data, names the phone's Bluetooth address and name and points out that the system deletes it on disconnect. That is exactly when recovery matters.
- Paired devices come from the Bluetooth configuration of Android and of Linux systems. Link keys appear only as a fingerprint.
- CAN bus captures, as diagnostic tools and data loggers write them, are read as candump logs, in the Vector ASC format and in the binary formats BLF and MF4. Every frame appears with its time, channel, identifier and data in one table, while a second summarises per identifier when and how often it occurred.
- Where the manufacturer's DBC file is present, the bytes become signals with a name, a value and a unit: engine speed, road speed, gear, brake. Both byte orders, signed signals, value tables and multiplexing are supported. This reader has so far been checked against captures produced in-house.
- Connected USB devices and the activity of single apps per day sit in the same sources
Drones, tachographs and connected devices
Flight logs from PX4, ArduPilot and DJI, together with the MAVLink telemetry of ground stations, land in shared tables: flight path, altitudes, speed, flight modes, messages from the autopilot and parameters. A world time appears only where the log supports it, and the program says where it comes from.
With DJI, most of the content stays encrypted from format version 13 on. Serial numbers, start time and start location are still readable, and anyone holding the DJI key store simply supplies it. The program opens no connection to DJI.
These readers were checked against real recordings published as examples by open-source projects and compared with independent readers: millions of values from flight logs, CAN captures, data loggers and sports watches agree.
- Digital tachograph: card holder, vehicles used, the activities of each day and, on newer units, GNSS positions and border crossings
- OBD-II captures from an ELM327 adapter become measurements with units, fault codes and the vehicle identification number
- GPS tracks from NMEA, GPX, KML, TCX, IGC, AIS and the subtitles of DJI videos, plus FIT files from sports watches
- systemd journals from routers, cameras or head units
- Photos and video from the memory card bring drone details from the DJI XMP and GoPro telemetry with them
Prepared for passing on
One click writes out all tables the way a person wants to read them: as CSV for a spreadsheet, or as a single HTML page.
Next to every identifier stands what it refers to. sender_id = 2 becomes the name of the sender, chat_id = 5 the name of the group. Where the record is missing, it says so.
- Times as dates, packed content unpacked, with the raw value kept alongside
- In the HTML page every reference leads to its record with one click
- And backwards too: every contact shows which messages and calls refer to it.
- Both outputs are recorded in the case log with their hash value
Screenshots
The screenshots show invented sample data. Phone numbers and e-mail addresses are redacted.
Technical details
The program works completely offline. It opens no connection to the network.
| Version | 1.22.2 |
|---|---|
| Operating system | Windows 10 and 11 (64 bit), from source also on Linux |
| Installation | none, unpack the folder and start |
| Portable use | settings and cases stay in the program folder, no registry entries of its own |
| Network | no connections, no telemetry |
| Robustness | version 1.22 reviewed the whole code base: crafted files neither halt the program nor fill its memory, a search pattern that would run away is refused before it runs, and many readers got noticeably faster along the way |
| Time formats | 19, including the time inside UUIDs, ULIDs and Snowflake IDs |
| Encryption | detected from the file header (SQLCipher, SQLite3 Multiple Ciphers, SEE); individual values with AES and ChaCha20 where the key is known, and for XOR the program finds the key itself; whole databases through SQLCipher with a known key |
| Hash values | SHA-256 and MD5 per file, case log as a hash chain |
| Computing core | recovery computes in Rust; without that core the program computes in Python with the same results |
| Evidence of integrity | a bill of materials of all libraries as CycloneDX and the SHA-256 of every program file are included |
| Further formats | 36 besides SQLite: SharedPreferences, DataStore, protobuf, FlatBuffers, LevelDB, IndexedDB, Hive, MMKV, LMDB, ObjectBox, Realm, RocksDB, Berkeley DB, Couchbase Lite, the Bluetooth configuration of Android and Linux, CAN bus as candump, Vector ASC, BLF and MF4, flight logs from PX4, ArduPilot, DJI and MAVLink, GPS tracks as NMEA, GPX, KML, TCX, IGC and AIS, FIT, OBD-II, tachograph downloads, systemd journals and Home Assistant; the format is detected from the content when opening |
| Images and attachments | thumbnails, capture data from JPEG, PNG, WebP, TIFF and HEIC, MD5, SHA-1 and SHA-256 per find, your own hash lists |
| Output | examination report as HTML, prepared output as CSV or HTML, export as CSV, SQL, JSON, CASE/UCO and as an SQLite database of the finds |
| Languages | interface and manuals in German and English, light and dark theme |
| Availability | download on this page |
License
TecInt-DBlite is free of charge for private individuals for private use. Companies, authorities, associations, educational institutions and other organisations can request permission from the developer.
The source code can be inspected. Commercial use and redistribution are not permitted. The license terms supplied with the program apply. Enquiries to mail@tecint.de.
DBlite questions
Which formats does DBlite read?
SQLite and 36 more, among them LevelDB, IndexedDB, Realm, RocksDB, LMDB, ObjectBox, Hive, MMKV, Berkeley DB, Couchbase Lite, the Bluetooth configuration of Android and Linux, and CAN bus captures. The program works out the format from the content.
Is the original file changed?
No. On opening, the database with its WAL and journal is secured into a case folder with SHA-256 and MD5, and the work happens on a write-protected copy. On closing, the original is checked again.
Does it find deleted records?
From freeblocks, freelist, WAL and rollback journal, including tables without rowid and dropped tables. In the other formats superseded and deleted states come back as far as they still sit in the file.
What does DBlite cost?
For private individuals nothing, for private use. Organisations ask the author first.
Can it run from a stick?
Yes, the program is portable and needs no installation.
Download
The file comes as a ZIP, unpack it and start. Two downloads per address per day are possible.