fu7ur3pr00f

Career intelligence agent that gathers professional data, searches job boards, analyzes career trajectories, and generates ATS-optimized CVs through conversational chat.

Quick Start

# Install
pipx install fu7ur3pr00f

# Run
fu7ur3pr00f

In the chat:

• /setup — Configure your LLM provider
• /gather — Import LinkedIn, GitHub, portfolio, CliftonStrengths
• /analyze — Get skill gap analysis
• /search — Query 7 job boards + Hacker News
• /generate — Create ATS-optimized CV (Markdown + PDF)

Installation

Debian/Ubuntu (amd64)

curl -fsSL https://juanmanueldaza.github.io/fu7ur3pr00f/fu7ur3pr00f-archive-keyring.gpg | \
  sudo tee /usr/share/keyrings/fu7ur3pr00f-archive-keyring.gpg >/dev/null

echo "deb [arch=amd64 signed-by=/usr/share/keyrings/fu7ur3pr00f-archive-keyring.gpg] \
https://juanmanueldaza.github.io/fu7ur3pr00f stable main" | \
  sudo tee /etc/apt/sources.list.d/fu7ur3pr00f.list

sudo apt update && sudo apt install fu7ur3pr00f

Development

git clone https://github.com/juanmanueldaza/fu7ur3pr00f.git
cd fu7ur3pr00f
pip install -e .
pip install -r requirements-dev.txt

# Run locally
fu7ur3pr00f

# Run with debug logs
fu7ur3pr00f --debug

Configuration

Run /setup in the chat, or manually edit ~/.fu7ur3pr00f/.env:

# Pick ONE provider (auto-detected if empty)
FUTUREPROOF_PROXY_KEY=fp-...   # Default, zero config
OPENAI_API_KEY=sk-...
ANTHROPIC_API_KEY=sk-ant-...
GOOGLE_API_KEY=...
OLLAMA_BASE_URL=http://localhost:11434  # Local, offline

See .env.example for all options.

Chat Commands

Architecture

Multi-Agent with Blackboard Pattern

graph TB
    User --> Chat[Chat Client]
    Chat --> Engine[Engine<br/>invoke_turn]
    Engine --> Outer[Outer Graph<br/>SessionState persistent]
    Outer --> Classify[classify_turn<br/>factual / follow_up / new_query]
    Classify --> Route[route_turn<br/>LLM routing + keyword fallback]
    Route --> Inner[Inner Blackboard Graph<br/>per-turn execution]
    Inner --> S1[Coach]
    Inner --> S2[Jobs]
    Inner --> S3[Learning]
    Inner --> S4[Code / Founder]
    S1 & S2 & S3 & S4 --> KB[(ChromaDB<br/>career knowledge)]
    Inner --> Synth[Synthesis<br/>LLM]
    Outer --> Accum[accumulate findings<br/>cross-turn context]
    Accum --> Suggest[suggest_next<br/>proactive follow-ups]

Routing Architecture:

• LLM-based semantic routing: Understands query intent, selects 1-4 specialists
• Keyword fallback: Deterministic fallback if LLM unavailable (rate limits, network errors)
• Fast paths: Factual queries → coach only; follow-ups → reuse previous specialists
• Structured output: RoutingDecision model guarantees valid specialist names
• Specialist guidance: All instructions load from prompts/md/specialist_guidance.md (no hardcoded fallbacks)
• Direct model selection: Purpose-specific models are selected from configured provider settings; invocation errors surface directly instead of retrying across models

Design decisions:

Tools

41 tools organized by domain:

MCP Clients

12 MCP clients for real-time data access:

Development

# Install dev tools
pip install pyright pytest ruff

# Test
pytest tests/ -q
pyright src/fu7ur3pr00f
ruff check .
ruff check . --fix

Scripts

Testing

# Unit tests
pytest tests/ -q

# Benchmarks
pytest tests/benchmarks/ -v

# Fresh install check
scripts/fresh_install_check.sh --source local --config-from .env

# Vagrant apt repo testing
scripts/vagrant.sh test-apt

# Multi-agent system testing
scripts/vagrant.sh multi

Offline behavior:

• Specialist routing falls back to deterministic keyword scoring in CI or other offline environments.
• CV parsing falls back to local heading extraction for Markdown and plain-text resumes when LLM section extraction is unavailable.
• Sensitive LinkedIn/social sections such as conversations and sponsored message threads are excluded before knowledge indexing, not just hidden at search time.

System Dependencies (Optional)

Tech Stack

Python 3.13 · LangChain + LangGraph · ChromaDB · Typer + Rich · WeasyPrint · MCP

Documentation

Documentation is embedded in the codebase:

• Prompts: src/fu7ur3pr00f/prompts/md/ — All system and specialist prompts
• Tools: src/fu7ur3pr00f/agents/tools/ — Tool implementations with docstrings
• MCP Clients: src/fu7ur3pr00f/mcp/ — MCP client implementations
• Memory: src/fu7ur3pr00f/memory/ — ChromaDB, RAG, episodic memory
• Gatherers: src/fu7ur3pr00f/gatherers/ — Data collection modules
• Specialists: src/fu7ur3pr00f/agents/specialists/ — Multi-agent specialists and orchestrator
• Blackboard: src/fu7ur3pr00f/agents/blackboard/ — Blackboard pattern implementation

Key documentation files:

• QWEN.md — Project context for AI assistants
• GEMINI.md — Additional project documentation
• .env.example — Configuration reference

Licensed under GPL-2.0.