Binarius AI Python Code Studio — Product documentation

Version 1.0.0 · Binarius Systems Private Limited · Support: support@binarius.co.in

License

Binarius AI Python Code Studio is licensed under the Eclipse Public License 2.0 together with the Binarius end user license terms; it includes PyDev (Eclipse Public License 1.0). Every installation includes a 28-day free evaluation; after that a license key for Binarius AI Python Code Studio is required (keys of the other studios do not apply). Python is downloaded on first use from the python-build-standalone project; packages come from PyPI under their own licenses.

• Full terms: License terms. Get a key: Request a license (mention "Python Code Studio").

System requirements

Download and installation

Download from the download section:

• Windows installer: accept the license, choose the folder, and optionally tick Install local AI (Ollama). It installs next to the other Binarius studios; each is a separate product.
• Zip: unzip to a short path such as C:\BAPCS and start BAPCS.exe.
• macOS: drag the app to Applications; on the first start use right-click > Open.
• Linux: sudo apt install ./binarius-ai-python-code-studio-1.0.0-linux-amd64.deb, or unzip and run ./BAPCS.

First start

1. Choose a workspace folder. Python is downloaded in the background (about 50 MB, once).
2. File > New > Python Project: pick a template (console program, FastAPI, Flask or Django) and press Run.
3. Sign in to your AI agent (welcome page > Sign in to an agent) or set up local AI, and send a request from the AI Assistant tab.

Features and usage

Binarius AI Python Code Studio is a complete Python IDE (editor, code analysis, Console, debugger, pytest and Git) plus an AI Assistant tab that connects to coding agents: Claude Code, Codex, Gemini CLI, GitHub Copilot CLI, or a local model that runs on your computer. You never set up Python: the studio downloads it once, gives each project its own .venv and installs the packages of requirements.txt before your program runs.

Install

• Windows: run binarius-ai-python-code-studio-1.0.0-setup.exe. Keep Install local AI (Ollama) ticked to work offline. Python is downloaded for you on the first start; nothing else is needed.
• macOS / Linux: unpack the archive and start BAPCS. Local AI is installed automatically when you pick the Local agent.

First run in three steps

1. Choose a workspace folder (where your projects live).
2. On the welcome page, click Set up local AI (offline) or Sign in to an agent.
3. Create a project (File > New > Python Project) and press Run, or in the AI Assistant tab type what you want, e.g. Create a project that demonstrates OOP concepts in Python with tests, and press Ctrl+Enter.

Welcome page

Shown when the IDE starts. Cards open the AI Assistant, set up local AI, sign in to an agent, create or import projects and open settings. Clicking an example prompt puts it into the AI tab.

Workspaces and projects

A workspace is a folder with your projects. The Project box in the AI tab chooses where the agent works: one project, or (workspace folder) to let the agent create new projects. Projects the agent creates are imported and selected automatically; your next message continues on them.

Projects are shown in the Project Explorer on the left. Add files with File > New > Python Module (a .py file) or Python Package (a folder with __init__.py), or right-click a folder > New. Python editor, code analysis and interpreter settings are under Window > Preferences > Python.

Run and debug

Every project has requirements.txt (its packages, like pom.xml in Java), .python-version (default 3.13) and its own .venv, which the Project Explorer hides. Press Run on the toolbar: the studio downloads Python if needed, creates the .venv, installs missing packages (shown in the Python Environment console) and starts the program in the Console, where errors are links to the line.

• Run starts the open or selected .py file, else the project's src/main.py, main.py, app.py… A saved run configuration is created (Run > Run Configurations for arguments and environment).
• Debug does the same with breakpoints: double-click the left margin of a line, press Debug, and the Debug perspective shows the stack, Variables and Expressions.
• Tests: right-click a test file > Run As > Python unit-test, or ask the agent; the IDE runs pytest itself after agent changes.

Packages (Manage Packages)

Right-click a project (or requirements.txt) > Manage Packages..., like NuGet in Visual Studio. Every change runs pip in the project's .venv at once, keeps requirements.txt in step, shows pip's output live, and can be undone with Undo last change.

Real time: saving requirements.txt installs its packages immediately. A line pip cannot install (a typo, a version that does not exist) is marked as an error on that line, in the editor and the Problems view, until you fix it. A failed install never changes requirements.txt. Opening a project from Git installs all its packages on the first Run.

Web apps and the preview

File > New > Python Project offers templates: Console program, FastAPI web API, Flask web app and Django web app, each with pytest tests that pass. Press Run: the server starts and the page opens in the IDE's own Preview browser tab as soon as the server prints its address (any program that prints e.g. Running on http://127.0.0.1:5000 gets the preview). FastAPI's interactive API docs are at /docs.

AWS Lambda and Azure Functions

The templates AWS Lambda function and Azure Functions app make a function you can run, debug and test in the IDE without any cloud account or extra tools:

• AWS Lambda: src/lambda_function.py (handler lambda_function.lambda_handler), events/event.json, and src/main.py, a local runner that calls the handler with the event and a context object (request id, remaining time) like Lambda does.
• Azure Functions (Python v2 model): src/function_app.py with an HTTP-triggered function, src/host.json, events/hello.json and events/hello-post.json (function, method, url, params, headers, body), and src/main.py, which builds the HTTP request and calls the function. For the full host use Azure Functions Core Tools (func start).

Press Run (or Debug, to stop at breakpoints in the function) to call it with the default event. Right-click any events/*.json file > Run Function with This Event (or Debug): each event file gets its own launch configuration. The pytest tests of the template pass from the start.

Right-click the project > Build > Build AWS Lambda Zip makes dist/<name>-lambda.zip: your code from src/ (without the local runner and tests) and the packages of requirements.txt for Lambda's Linux, ready to upload. Azure apps are published from src/ with func azure functionapp publish or VS Code.

Notebooks

Open any .ipynb file (or create one, e.g. analysis.ipynb) in the studio's notebook editor. Cells run in the project's .venv and keep their variables; the last expression of a cell is shown (a pandas DataFrame as a table), errors show the traceback of your cell.

Data viewer

While debugging and stopped at a breakpoint, right-click a variable in the Variables view > View Data (or Run > View Data... for any expression). Lists, lists of dicts or rows, dicts, pandas DataFrames and Series, numpy arrays and objects open as a table in the Data Viewer view, up to 1000 rows.

Security check

Manage Packages > Security checks the installed packages against known vulnerabilities (osv.dev, the open database pip-audit uses; only package names and versions are sent). Each advisory shows its severity and the version that fixes it: Update selected to the fixed version or Fix all. Agents can run the same check (bapcs_check_security).

Wheel and single-file app

Right-click a project > Build Python Package:

• Build Wheel... makes dist/<name>-<version>-py3-none-any.whl to install with pip or publish on PyPI. Its dependencies come from requirements.txt (development tools such as pytest are left out); main() in src/main.py becomes a command.
• Build Single-File App makes dist/<name> (.exe on Windows) with PyInstaller: it runs on computers without Python. It is built for the system the studio runs on; build on Windows, macOS and Linux for each.

The AI Assistant tab

The transcript shows your messages, the agent's replies, each tool call (e.g. Write books.py [done]) and notices from the IDE (ℹ information, ⚠ problems, with buttons to fix them). The plan pane below the transcript shows the agent's plan when it sends one.

Agents and sign-in

Node.js is set up automatically: if the computer has no Node.js 22+, the IDE downloads the latest Node.js LTS from nodejs.org into ~/.bapcs/node (per user, no administrator rights, the system is not changed), so you only sign in to your agent. Online agents are downloaded on first use (via npx) and run under their publishers' own terms. Your prompts and the files they read are sent to that provider.

Models and options

The Model box lists the models the selected agent offers; Options… holds its other settings. The list is read when you select the agent (it starts briefly in the background) and your choice is remembered per agent.

Temperature (local models)

How much randomness the model uses: 0 exact and repeatable (fixing errors), 0.2 recommended for code, 0.5 more varied ideas, 0.8–1 creative but more mistakes (not for code).

Local AI (offline)

The local agent runs a model on your computer with Ollama: code never leaves the machine and no account is needed.

• Default model: qwen2.5-coder:14b (about 9 GB; 16 GB RAM, a GPU with 8 GB+ recommended). Smaller PCs: choose qwen2.5-coder:7b in the Model box.
• Company model server: set the endpoint in Preferences > Local agent, or by admin policy.
• Settings: Preferences > Local agent (endpoint, model, temperature, max tool calls, max tokens, request timeout, Test connection).

Writing requests

• Say what you want and, if it matters, how: Create a library management system in Python with pytest tests; Add a REST endpoint to delete a book, with a test.
• Name a framework if you want one: … with FastAPI, … with Django, … with pandas.
• Follow-ups work on the same project: add an interface Payable and use it in Main.
• Ask questions too: Explain this project, Why does this test fail?

The first message on a project with code makes the agent read it first and summarise it before working, so it keeps your modules, naming and libraries.

Framework and Python version

The agent reads the request first, then decides:

• You name a framework or library (FastAPI, Django, Flask, SQLAlchemy, pandas…) → it uses exactly that.
• Plain programs, algorithms, OOP and data-structure demos → the standard library only (and pytest for tests).
• The request needs a framework (REST API, web app, database) but names none → if one clearly fits it says which and starts; if several fit it asks you which one before creating files.

Packages the agent needs are installed with the package manager (bapcs_install_packages, recorded in requirements.txt); it never runs pip in your system Python.

Check, test, run and fix

After every reply that changed files, the IDE checks the work itself:

1. Environment: the project's .venv and the packages of requirements.txt (a line pip cannot install is reported with its line).
2. Code check: every .py file is parsed (syntax errors with file and line) and every import resolved. A package the code imports but requirements.txt lacks is added and installed; imports used only by tests are reported, never installed.
3. Tests: if there are none, the agent is asked to write pytest tests; then pytest runs and each failure is reported.
4. No code written: if the project still holds only the new-project template, the agent is told that describing code does not create it.
5. Run: src/main.py is run and its output read (a server that keeps running counts as started); if there is none, the agent is asked to add one. A run configuration is created for the Run button.
6. Anything failing is sent back to the agent with the errors and the traceback, up to 5 fix rounds.

The transcript ends with ✔ Done (the code checks clean, tests pass, the program ran) or with what is still failing.

MCP server for your project

Make any Python project an MCP server, so AI assistants (Claude Desktop, Claude Code, VS Code, Cursor and other MCP clients), chatbots and other applications can call your functions as tools.

1. Right-click the project > MCP Server > Add MCP Server (also in the Project menu).
2. Every top-level function of src/ that is public, not decorated and has parameters typed int, float, bool or str becomes a tool named <module>_<function>; its docstring is the description.
3. The IDE writes src/mcp_server.py on the official MCP Python SDK, its pytest test, .mcp.json, .vscode/mcp.json and MCP.md, installs the SDK into the .venv, and then starts the server and calls it like a real client.

Connect it: Claude Code and VS Code read .mcp.json and .vscode/mcp.json; Claude Desktop takes the same entry in its config file. On the network: MCP_TOKEN=<a long secret> python src/mcp_server.py --http 8090 --host 0.0.0.0 serves Streamable HTTP at /mcp; calls without Authorization: Bearer <secret> get 401, and it refuses to listen on the network without a token. FastAPI apps serve the same tools inside the app at /mcp, next to the API. MCP.md has a ready-to-use Python client.

Test MCP Server starts it over stdio and over HTTP (free port, random token), lists the tools and checks that calls without the token are rejected. With the AI tab, ask "Make this project an MCP server": the agent uses the same generator (bapcs_add_mcp_server) and adds custom tools in add_custom_tools().

Approvals and reviewing changes

• Auto-approve on (header checkbox): the agent works without asking, like an autonomous mode. Every step is still recorded in the audit log.
• Auto-approve off: each tool call asks Allow once / Always allow for this session / Deny, and each file change opens a Compare editor. The pending changes list in the AI tab has Review, Apply, Reject and Apply all. You can edit the change before applying it.
• Allow read-only tool calls without asking (Preferences) skips the question for reading and searching.
• An admin can force review of every change and block shell commands by policy.

Switching agents and follow-ups

Change the Agent box at any time. The new agent starts on its next message, first analyses the existing code and is told which agent worked on it before and what it was asked. Use Use my last message with … in the notice to resend your request to the new agent.

Git and GitHub

Chat history

Every chat is saved per project: your messages, the agent's replies, tool calls and IDE notices, with the time of each entry and secrets removed.

Audit log

A record of AI activity for teams and compliance: sessions, prompts (secrets removed; optional), tool calls, approvals, file changes (paths, hashes and line counts, never file contents), verification results, model changes and Git actions.

Admin policy

Administrators control the IDE for a team with a policy file bapcs-policy.json: allowed agents, default agent, enforced Git mode, local model server, whether shell commands are allowed, forced review, audit and chat history settings, updates. Locked settings appear greyed out with Set by your organization's policy.

Evaluation and license keys

Every installation can be used free for 28 days from its first start (the AI tab and the welcome page show the days left). After that a license key is required: at startup the IDE asks for it (Activate or Exit). Reinstalling does not restart the evaluation.

A key (BAPCS1.…) names the licensee, the number of seats and the expiry date. It is checked on your computer; no internet connection is needed. Request one at the Binarius AI Python Code Studio website (License page).

Preferences

Agents table

The coding agents available in the AI tab's Agent box.

Claude Code, Gemini CLI, Codex and Copilot CLI need Node.js; the IDE installs it automatically when it is missing. Sign in once with each agent's own login (⋮ > Sign in to Agent in the AI tab) or put its API key into the agent's environment.

Local agent

Settings of the built-in local agent, which works with Ollama, LM Studio, vLLM or a company server with an OpenAI-compatible API.

Environment for all agents

Variables given to every agent, one NAME=value per line, for example HTTPS_PROXY=http://proxy.company:8080 or NODE_EXTRA_CA_CERTS=C:\certs\company.pem. Kept in secure storage. An agent's own environment (Edit…) overrides a variable with the same name.

Approvals

IDE tools

Git and GitHub

The Git mode itself (Manual, Branch or Pull request) is chosen per project in the AI tab.

Audit log and chat history

Updates

License

Where things are stored

Add .bapcs/ to .gitignore if chats should not be committed with the project.

All features

Menus and shortcuts

Updates

If your company or Binarius provides an update site, the IDE checks it at startup and offers updates (Preferences > Updates: on/off, stable or beta). Help > Check for Updates checks manually.

Troubleshooting

© 2026 Binarius Systems Private Limited · Based on Eclipse · Eclipse Public License 2.0