I built ResPilot, a Python repository on my GitHub, as an autonomous competitive-intelligence agent for researching complex, multi-step goals with live web data. The repository metadata says it started and was last updated in 2026-10.
I describe ResPilot as an autonomous, self-healing competitive-intelligence agent that plans and executes complex multi-step research goals using live web data.
I designed ResPilot to go beyond an LLM wrapper by breaking down user goals, executing deterministic tools safely, recovering from failures, and synthesizing structured reports.
ResPilot builds and displays a structured JSON plan before executing any steps.
ResPilot executes web_search, webpage_fetcher, and calculator safely, using Python AST for math rather than arbitrary eval() exploits.
ResPilot gathers up-to-date information dynamically from the internet instead of relying only on static LLM training data.
ResPilot detects malformed tool outputs and automatically recovers using an exponential backoff retry mechanism.
ResPilot produces comprehensive Markdown reports and JSON execution traces.
ResPilot uses a Plan-and-Execute pattern: an LLM determines sequential steps, an execution engine feeds intermediate states back to the LLM to construct arguments for deterministic Python tools, and strict Python logic validates outputs before proceeding.
This separation of planning, execution, and validation is intended to make multi-step research more reliable. The README points to docs/architecture.md for a deeper architectural description.
I list Python 3.10 or higher as a ResPilot prerequisite.
I list an active LLM API key, such as OpenAI, as a ResPilot prerequisite.
git clone <repository_url>
cd respilot
python -m venv venv
.\venv\Scripts\activate
source venv/bin/activate
pip install -r requirements.txt
I configure ResPilot by copying .env.example to .env; the example sets LLM_PROVIDER=openai and OPENAI_API_KEY=sk-xxxxxx.
I list MODEL_NAME=gpt-4o-mini as an optional model configuration.
I provide both a command-line interface and a Streamlit web interface for running ResPilot.
python main.py --goal "Analyze NVIDIA's competitive position in the AI accelerator market and give me a concise competitive landscape report."
python main.py --goal "Compare OpenAI and Google DeepMind" --output report.json
python main.py --goal "Analyze NVIDIA's competitive position" --demo-failure
ResPilot's --demo-failure mode injects a failure_simulator tool that intentionally fails on the first attempt to demonstrate recovery.
streamlit run frontend/streamlit_app.py
The local Streamlit interface is usually available at http://localhost:8501.
I use pytest tests to validate LLM outputs, schemas, and failure recovery with mocks and no network required.
pytest tests/ -v
My README's example goal compares OpenAI and Google DeepMind using their latest publicly available AI developments and asks for the percentage difference between 175B and 1.76T parameters.
The example execution has 4 steps: two web_search steps, one calculator step, and one synthesis step.
In the example execution, the first three steps show TOOL_SUCCESS; synthesis shows SYNTHESIS_START | IN_PROGRESS, followed by AGENT_COMPLETED | SUCCESS.
The repository is organized around core agent logic, deterministic tools, output validators, documentation, sample runs, a Streamlit frontend, and tests. The README directs readers to docs/architecture.md and the examples/ directory for deeper design details and full execution transcripts.
My ResPilot repository is Python; it started in 2026-10 and was last updated in 2026-10.
My ResPilot Streamlit frontend provides live execution tracing and report downloading.
My ResPilot app/ directory contains the core application logic, including the agent, planner, and executor.
My ResPilot app/tools/ directory contains deterministic tools, including web search and calculator.
My ResPilot examples/ directory contains sample execution transcripts.
My ResPilot frontend/ directory contains the Streamlit web interface application.
My ResPilot tests/ directory contains end-to-end and recovery test suites.
My ResPilot CLI entry point is main.py.
My ResPilot root directory includes requirements.txt for Python dependencies.
My ResPilot docs/ directory contains architecture and design documentation.