InsightRelay

Discover, analyze, deliver.

InsightRelay is a modular Python research pipeline that runs three LLM-driven agents in sequence to turn a raw topic into a structured, decision-ready report:

1. Researcher gathers source material using simulated or live web search.
2. Analyst turns the raw notes into ranked findings and conclusions.
3. Writer produces a final report and saves it to a timestamped text file.

The code is split into a small CLI entry point and a reusable swarm/ package so the workflow is easier to maintain, extend, and publish.

Features

• Multi-agent workflow with a clear research-to-report handoff.
• Supports ollama for local inference.
• Supports cerebras for cloud inference.
• Uses simulated search by default for offline runs.
• Restricts web search to the Researcher only.
• Writes UTF-8 reports with timestamped filenames.
• Keeps runtime configuration in .env.

How It Works

InsightRelay runs a three-stage pipeline:

1. Researcher collects source material from simulated or live search.
2. Analyst ranks findings, surfaces patterns, and flags gaps.
3. Writer turns the analysis into a polished final report.

This separation keeps each role narrow and makes the outputs easier to reason about and improve.

Project Structure

.
├── agentsSwarm.py
├── swarm/
│   ├── __init__.py
│   ├── agents.py
│   ├── config.py
│   ├── llm.py
│   ├── orchestrator.py
│   ├── runner.py
│   └── tools.py
├── .env.example
├── .gitignore
└── requirements.txt

Requirements

• Python 3.11+
• An LLM provider:
  • Ollama running locally, or
  • a Cerebras API key
• Tavily API key only if you want live web search

Setup

python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
cp .env.example .env

Then edit .env with the provider and search settings you want.

Environment Variables

Variable	Purpose	Default
LLM_PROVIDER	ollama or cerebras	ollama
MODEL	Ollama model name	llama3.2
CEREBRAS_MODEL	Cerebras model name	llama3.1-8b
CEREBRAS_API_KEY	Required for Cerebras runs	empty
OLLAMA_URL	Ollama chat endpoint	http://localhost:11434/api/chat
SEARCH_MODE	simulated or live	simulated
TAVILY_API_KEY	Required for live search	empty
REQUEST_TIMEOUT_SECONDS	Per-request timeout	120
MAX_AGENT_STEPS	Tool-loop step cap per agent	8
MAX_OUTPUT_CHARS	Output truncation limit	20000

Running

Run interactively:

python3 agentsSwarm.py

Run non-interactively:

python3 agentsSwarm.py --topic "AI in healthcare"
python3 agentsSwarm.py --topic "Semiconductor supply chains" --max-steps 10

Each run writes a report like report_YYYYMMDD_HHMMSS.txt in the project root.

Provider Notes

Ollama

ollama serve
ollama pull llama3.2

Set in .env:

LLM_PROVIDER=ollama
MODEL=llama3.2

Cerebras

Set in .env:

LLM_PROVIDER=cerebras
CEREBRAS_API_KEY=your_api_key_here
CEREBRAS_MODEL=llama3.1-8b

GitHub Push Checklist

• Keep .env out of version control.
• Commit .env.example instead of real secrets.
• Review generated report_*.txt files before committing anything.
• Validate the code before pushing:

python3 -m py_compile swarm/config.py swarm/llm.py swarm/tools.py swarm/runner.py swarm/agents.py swarm/orchestrator.py swarm/__init__.py agentsSwarm.py

Optional Next Steps

• Add tests for config loading and tool-call parsing.
• Add a GitHub Actions workflow for linting and compile checks.
• Add structured logging if you want to run this non-interactively in automation.