DogiCat DMS

Don't just browse your database. Follow the evidence.

A Python desktop workbench for exploring databases, tracing records,
comparing changes, and investigating with your own AI model.

简体中文 · Quick start · Try the demo · AI setup

The payment succeeded. Why is the order still pending?

Start with one record. Follow its relationships, compare what changed, and share the evidence—not access to your entire database.

DogiCat brings everyday database management and a read-only investigation workspace into the same app. SQLite and the built-in demo work without extra Python packages. AI is optional.

Five ways to go beyond a query result

A record is the starting point

The included fictional dataset has an order with a successful payment but a stale status:

flowchart LR
    O["orders · 1001<br/>status = pending"]
    C["customers · 1"]
    I["order_items<br/>Keyboard × 1"]
    P["payments<br/>status = success"]
    O --- C
    O --- I
    O --- P

Illustrative relationships from the demo dataset—not an application screenshot or proof of why the inconsistency occurred.

Follow the records → capture a baseline → make a controlled change → compare → review and share.

Start in one command

Download or clone this repository, then run:

git clone https://github.com/HeartingU/Dogicat_DMS.git
cd Dogicat_DMS
python main.py

You need Python 3.9+ with Tk. On Windows, use a Python installation that includes Tcl/Tk. No database server, model, or API key is needed for the demo.

Install only the optional drivers you need:

python -m pip install mysql-connector-python psycopg2-binary

The app starts in Chinese by default. Choose 菜单 → 语言 → English to switch. Language changes are remembered and preserve open tabs, SQL and input; some older administration messages are still Chinese.

Try the investigation in two minutes

1. Open Menu → Tools → Try demo. DogiCat creates a fresh, fictional SQLite database.

2. Select orders, choose id, enter 1001, and click Trace records. Inspect its customer, line item and successful payment.

3. Click Mark as before to keep the baseline.

4. In the SQL editor for that demo connection only, run:

   UPDATE orders SET status = 'paid' WHERE id = 1001;
   

5. Return to the investigation tab and click Capture and compare. You should see status change from pending to paid.

6. Export a case, review the masking preview, then reopen it with Open case.

Want an AI explanation? Click Ask AI, configure your service, review the exact context and send a question. Nothing is sent simply by opening the assistant.

Your model. Your context. Your decision.

Use an online API or a local model server. The implemented protocols cover OpenAI Responses, Chat Completions-compatible APIs, and Ollama. LM Studio can use the compatible API option.

• Five built-in tasks: investigate data, draft read-only SQL, explain errors, interpret diffs, and analyze relationships.
• Inspect before sending: start with schema-only context; add records/diffs explicitly, then review and edit the text.
• Choose the language: prompts and answers follow the interface by default, with a separate override.
• Keep execution in your hands: answers are suggestions, never automatically executed database commands.
• No model required: browsing, editing, tracing, diffs and offline cases work without AI.

Models and servers are not bundled or started by DogiCat. Actual endpoint/model combinations need testing in your environment.

Configure APIs and local models →

Still a database workbench

Alongside investigation tools, you get:

• Multi-database navigation and remembered connections.
• SQL editing, query history, data editing and table-structure tools.
• Search, filtering, sorting, and current-connection search across non-system databases.
• ER diagrams and SQL/CSV import/export workflows.
• Light/dark themes, switchable Chinese/English UI, and a resizable explorer.
• Font-sized tab close targets and batched table rendering to keep interaction responsive.

Recent local checks: a 10,001-row, two-column table rendered in 0.50–0.77 seconds; 25 repeated AI-tab open/close cycles left no extra pages or accumulating timers. These are local Windows measurements, not database-query benchmarks or guarantees for all workloads.

See the methodology and full performance notes →

Clear boundaries, fewer surprises

DogiCat is actively developed. It is an independent project—not affiliated with Navicat and not a claim of feature parity.

• Evidence snapshots are not backups. Captures are sequential reads, not atomic cross-table snapshots. Comparing two observations does not prove which action caused a change.
• Capture limits are explicit: 10,000 rows per table and 50,000 total. Oversized captures are rejected without replacing the previous snapshot. These limits do not reintroduce a cap on ordinary table browsing.
• Offline SQL is a supported-subset preview. It is not a complete restore engine; unsupported or simplified statements are reported.
• Masking needs review. Pseudonymization is not guaranteed anonymization. Raw snapshots are unredacted; case files are not encrypted.
• Local-first does not mean every AI request stays local. Sending to a remote model transmits the reviewed context to that service under its policies. Do not commit .dogicat/, keys, dumps, logs, snapshots or case files.
• Validation has a scope. Windows desktop and offline/mock-service tests are covered; production MySQL/PostgreSQL workflows, every model provider, and other desktop platforms still need target-environment verification.

Build confidence, then build features

python -m compileall -q dogicat tests tools
python -m unittest discover -s tests -v
python tools/smoke_investigation.py
python tools/smoke_responsiveness.py

The GUI checks require a desktop session. Tests use fictional SQLite data and mock services, not production credentials.

Found a bug? Open an issue with reproduction steps and a minimal fictional example—not a real customer dump.

Contribution guide · Changelog · Release checklist

License: The Unlicense, preserved from this repository's initial commit.

If tracing a record is part of your daily work, star the project to follow its progress. Useful bug reports and reproducible examples help shape what comes next.