Pathfind AI

A local-first desktop agent that can search, summarize, and orchestrate tools -- built for the Hack with DC hackathon. It keeps a privacy-friendly memory (SQLite) of conversations, artifacts, and facts for better follow-ups.

---

Getting Started

Follow these steps to get the app running on your computer.

1. Clone the repo

Open a terminal and run:

git clone https://github.com/your-username/desktop-agent-agi.git
cd desktop-agent-agi

2. Create a virtual environment

This keeps your dependencies isolated so nothing breaks on your system.

python3 -m venv venv
source venv/bin/activate

On Windows use:

python -m venv venv
venv\Scripts\activate

3. Install dependencies

pip install -r requirements.txt

4. Set up environment variables

The app needs two API keys to work. Create a .env file in the project folder:

cp .env.example .env

Then open .env and fill in your keys:

GOOGLE_API_KEY=your-google-api-key-here
HF_TOKEN=your-huggingface-token-here

• GOOGLE_API_KEY -- get one from Google AI Studio (Gemini API key)
• HF_TOKEN -- get one from Hugging Face

You can also add your resume path so the agent knows where to find it:

RESUME_PATH=/absolute/path/to/your/resume.pdf

5. Add your resume (optional)

If you want to tailor your own resume, drop your resume file into the outputs/ folder using these exact names:

outputs/tailored_resume.docx
outputs/tailored_resume_ats.pdf

Put both a .docx and a .pdf version with the same naming. The Resume Tailor Agent will read from here and write the tailored version back to the same spot.

6. Run the app

streamlit run streamlit_app.py

The app will open in your browser automatically (usually at http://localhost:8501). That's it!

---

Project Overview

Memory

• Every conversation turn, artifact, and fact is saved to a local SQLite database (memory.db).
• Data is scoped by session so different runs don't collide.
• Call memory.init_db() once to create the tables. The Streamlit app does this for you.

Agents

The app is powered by three specialized agents that work together:

• Role Search Agent -- Searches for open job postings across the web using Browser Use. You give it job titles, location, and keywords and it returns a deduplicated list of matching roles with company, URL, and requirements.

• Job Description Agent -- Takes a selected job posting and fetches the full, clean job description text (responsibilities, requirements, skills). It can pull directly from a URL or search for the listing. Returns up to 3 variants so you can pick the best one.

• Resume Tailor Agent -- The core of the app. Give it your resume (PDF) and a job description (text or URL) and it will tailor your resume to match the role. It extracts keywords from the JD, rewrites your bullet points to highlight relevant experience, and exports the result as both DOCX and ATS-friendly PDF.

Email / Recruiter Outreach

• Finds recruiter contacts (name, email, LinkedIn) for a company and role using Browser Use.
• Filters out junk emails and prefers same-company contacts.
• Can draft personalized outreach emails using an LLM.
• Optional Gmail integration for reading/summarizing threads (requires Google OAuth credentials).

Key Files

File	What it does
streamlit_app.py	Chat UI (Streamlit)
app.py	Core agent logic
agents/role_search_agent.py	Searches for open job postings
agents/job_description_agent.py	Fetches and cleans job descriptions
agents/resume_tailor_agent.py	Tailors your resume to a job description
memory.py	SQLite memory layer
browser_use_client.py	Browser Use API wrapper
email_handler.py	Recruiter lookup + email drafting
requirements.txt	Python dependencies

Architecture Diagram