Forge

Forge is a research-first AI workspace built to turn papers, benchmarks, and technical signals into decisions a product or engineering team can act on.

The current app lives in chat-app/ and is structured around two research modes:

• Deep Research: follow a focused question through methods, evidence, tradeoffs, and the strongest available answer.
• Wide Research: scan the landscape quickly across papers, labs, and solution families before deciding where to go deeper.

Product Shape

The public landing page frames Forge as a research agent for engineering and product teams.

The app workspace is built around:

• a dashboard for starting new research threads
• a threaded chat interface for continuing investigation
• file upload and model selection inside the composer
• deep/wide mode switching directly inside the input workflow

Repository Layout

• chat-app/: Nuxt application for the Forge landing page, chat interface, and proxy API
• research_agent/: Python/Agno backend running the agentic workflow on Amazon Nova (AWS Bedrock)
• LICENSE: repository license

Local Development

Forge runs as a two-part system: a Nuxt frontend and a Python/Agno backend.

1. Start the Agno Research Backend

The backend requires AWS credentials to access Amazon Nova models via Bedrock.

cd research_agent

# Install dependencies using uv
uv sync

# Configure your environment variables
cp .env.example .env
# Edit .env and add your AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY, and AWS_REGION

# Create the sqlite database directory
mkdir -p tmp

# Start the AgentOS server (runs on port 7777)
uv run python main.py

2. Start the Nuxt Frontend

In a new terminal window:

cd chat-app
bun install
bun run dev

The application will be available at http://localhost:3000.

Tooling Commands (Frontend)

cd chat-app
bun run typecheck
bun run lint
bun run build

Positioning

Forge is designed for questions like:

• Which planning architecture is holding up best for browser agents?
• What do recent papers say about hallucination reduction strategies?
• Which benchmark results actually matter for this implementation decision?
• What is the strongest recommendation once evidence and tradeoffs are compared?

The goal is not just to summarize research. The goal is to turn research into clear direction.