devobin 1.1.0

AI Engineering Context Compiler and Prompt Architect CLI — reads your code, builds the perfect prompt

DevObin AI

AI Engineering Context Compiler & Prompt Architect

DevObin is a CLI-based prompt engineering system that autonomously reads your project code and produces customized engineering prompts for AI coding agents. It does NOT write code — it crafts the perfect prompt so Codex, Claude, Cursor, or any AI agent builds exactly what you need.

---

What DevObin Does

You: "Add caching to my provider calls"
   │
   ▼
DevObin scans your project → reads your actual code → understands your architecture
   │
   ▼
DevObin writes: caching_feature_prompt.md  (in your directory)
   │
   ▼
Give the file to Codex/Claude/Cursor → they implement it perfectly

Key difference: DevObin reads YOUR code and produces a prompt based on YOUR actual architecture — not a generic template.

---

Installation

pip install devobin

Or from source:

git clone https://github.com/mobinhasanghasemi/devobin.git
cd devobin
pip install -e .

---

Quick Start

1. Launch DevObin

cd your-project/
devobin

2. Connect an AI Provider

> /connect

Provider	Key Format	Example Model
OpenAI	sk-...	gpt-4o
Anthropic	sk-ant-...	claude-sonnet-4-20250514
Google	AIza...	gemini-2.0-flash
Ollama	(no key)	llama3

3. Describe What You Want

> Add a caching layer to the provider calls

4. DevObin Does Everything

1. Scans your workspace automatically
2. Asks permission to read your files
3. Reads every important source file
4. Understands your real architecture
5. Writes the prompt file to your directory
6. Tells you "Created: caching_prompt.md"

You never type /export. DevObin creates the file itself.

5. Give the File to a Coding Agent

# The file is in your current directory
cat caching_prompt.md

# Give it to Codex, Claude, Cursor, or any AI agent

---

How It Works

Agent System

DevObin has an autonomous agent that explores your codebase using tools:

Tool	What It Does
[TOOL:glob:**/*.py]	Find files by pattern
[TOOL:grep:regex]	Search file contents
[TOOL:read:path]	Read a file completely
[TOOL:ls:path]	List directory contents
[TOOL:write:name.md]	Create the prompt file

The agent runs up to 10 rounds of tool calls to thoroughly understand your project before producing the prompt.

Permission System

Before reading your files, DevObin asks permission:

┌─ File Access Permission ─────────────┐
│ DevObin found 44 files.             │
│ Allow file access?                  │
│                                    │
│ [Yes — Read All Files]             │
│ [Ask Before Each File]             │
│ [No — Skip File Reading]           │
└────────────────────────────────────┘

Grounding Check

After generating the prompt, DevObin verifies it doesn't hallucinate technologies:

• If your project has NO database → prompt won't mention PostgreSQL
• If your project has NO Dockerfile → prompt won't mention Docker
• If your project has NO frontend → prompt won't mention React

If hallucinations are detected, DevObin automatically rewrites the prompt.

Session-Based History

Each chat conversation is stored separately:

.devobin/projects/myproject/sessions/
├── 2026-07-14T10-30-00.json
├── 2026-07-14T11-15-00.json
└── 2026-07-14T14-00-00.json

Use /history to browse past sessions.

---

Commands

Chat Commands

Command	Description
/connect	Set up an AI provider
/model <name>	Switch model
/project create <name>	Create project
/project switch <name>	Switch project
/project list	List projects
/scan	Re-scan workspace
/history	View past sessions
/memory	View saved memories
/export [name]	Manual export (rarely needed)
/settings	Show settings
/clear	Clear current chat
/help	Command palette

Shell Commands

Command	Description
/run <cmd>	Execute shell command
/read <file>	Read file
/ls [path]	List directory
/git <args>	Git commands
/pip <args>	Pip commands
/python <args>	Run Python

Keyboard Shortcuts

Shortcut	Action
Ctrl+K	Command palette
Ctrl+L	Clear chat
Ctrl+Q	Quit
/ alone	Command palette

---

Example Session

$ cd my-python-project
$ devobin

> /connect
  [Select: openai, enter API key, select gpt-4o]

> بچین کدها رو و یه پرامپت مهندسی شده بساز

  [Scanning workspace... 44 files found]
  [Permission: Read All Files]
  [Reading files... (round 1)]
  [Reading files... (round 2)]
  [Reading files... (round 3)]
  [Creating prompt file... (round 4)]
  Created: my_feature_prompt.md

> یه caching اضافه کن

  [Reading files... (round 1)]
  [Reading existing prompt... (round 2)]
  [Updating prompt file... (round 3)]
  Updated: my_feature_prompt.md

---

Architecture

devobin/
├── main.py              # Typer CLI entry
├── app.py               # Textual TUI app
├── core/
│   ├── scanner.py       # Workspace scanner
│   ├── tools.py         # Agent tools (glob/grep/read/ls/write)
│   ├── validator.py     # Output + grounding validation
│   ├── prompt_builder.py # Template-based prompts (fallback)
│   ├── analyzer.py      # Tech detection
│   ├── researcher.py    # Knowledge base
│   ├── memory.py        # Cross-session memory
│   ├── optimizer.py     # Token compression
│   └── executor.py      # Shell & file ops
├── providers/
│   ├── __init__.py      # Provider registry
│   ├── openai_provider.py
│   ├── anthropic_provider.py
│   ├── google_provider.py
│   └── ollama_provider.py
├── ui/
│   ├── chat_screen.py   # Main screen + agent loop
│   ├── export_modal.py  # Export dialog
│   ├── ask_modal.py     # Question dialog
│   ├── permission_modal.py # Permission dialog
│   ├── history_modal.py # Session history
│   └── command_palette.py
├── config/
│   └── settings.py      # Config + local_data_dir()
└── storage/
    └── database.py      # .devobin/ + sessions

---

Configuration

Platform Config (providers, settings)

Platform	Location
Windows	%APPDATA%/devobin/
Linux	~/.config/devobin/
macOS	~/Library/Application Support/devobin/

Project Data (history, memory)

Stored in .devobin/ inside your project directory:

your-project/
├── .devobin/
│   └── projects/
│       └── myproject/
│           ├── memory.json
│           ├── meta.json
│           └── sessions/
│               ├── 2026-07-14T10-30-00.json
│               └── 2026-07-14T11-15-00.json
├── src/
│   └── ...
└── caching_prompt.md   ← DevObin created this

---

Security

• API keys stored locally, never sent anywhere except the provider
• Tool write only allows .md files (no code execution)
• Directory traversal blocked
• File access requires explicit user permission
• Grounding check prevents hallucinated technology suggestions

---

Dependencies

• Python ≥ 3.12
• Typer (CLI)
• Textual (TUI)
• Rich (rendering)
• openai / anthropic / google-genai (AI providers)
• httpx (HTTP client)
• platformdirs (config paths)
• arabic-reshaper + python-bidi (RTL support)

---

License

MIT License

Author

Mobin Hasanghasemi

• Website: devobin.ir
• Instagram: @devobin
• GitHub: github.com/mobinhasanghasemi
• Email: [e-mail]