Unlock Framework‑Less Deep Research AI with GoSkills Agent in Pure Go

Background and Goals

In the AI 2.0 era, Deep Research agents must decompose a goal, gather information, analyze it, and produce a comprehensive report. Existing Python solutions are abundant, but the Go ecosystem lacked a high‑performance, easy‑to‑deploy, and logically transparent implementation.

Key Capabilities of GoSkills Agent

Autonomous Planning : The agent parses a user query and automatically breaks it into sub‑tasks such as search, analysis, and report generation.

Deep Search : Integrated with the Tavily API, it fetches up‑to‑date web content.

Intelligent Analysis : Filters, extracts, and reasons over the retrieved data.

Professional Reporting : Emits a structured Markdown research report.

Native TUI : Provides a Gemini‑style terminal UI for a smooth interactive experience.

Compilation

The project builds with the standard Go toolchain. After cloning the repository, run either

# Clone the repository
git clone https://github.com/smallnest/goskills.git
cd goskills
# One‑click build with Makefile
make agent
# Or build directly
go build -o agent-cli ./cmd/agent-cli

which produces the agent-cli binary.

Installation

Because the binary is statically linked, no Python dependencies or virtual environments are required. Move the binary into a directory on your PATH (e.g., mv agent-cli /usr/local/bin/) and you can invoke agent-cli from anywhere.

Usage

Start the agent with agent-cli to see the TUI. Commands include:

Enter a research topic, e.g., “Analyze the latest breakthroughs in quantum computing 2024”. \help – display help. \clear – clear context history. \exit or \quit – quit the program.

Example Session

Set environment variables for the LLM and search service:

# Example using Baidu DeepSeek
export OPENAI_API_KEY=YOUR_KEY
export OPENAI_API_BASE=https://qianfan.baidubce.com/v2
export OPENAI_MODEL=deepseek-v3
# Tavily API key for web search
export TAVILY_API_KEY=tvly-dev-xxxxxxxxxxxxxxxx

Run the agent with debug output: agent-cli -v. The agent then:

Perceives the user input.

Planning Agent uses the LLM to generate a JSON Plan containing ordered Task objects.

Execution Agent schedules each Subagent (Search, Analysis, Report, Render) according to the plan.

All Subagent outputs are stored in a shared context for later steps.

Render Subagent converts the Markdown report into a color‑rich terminal view.

Images in the original article illustrate each stage (search, analysis, rendering) and are retained for reference.

Implementation Details

The framework‑less design relies on a Planner‑Executor‑SubAgents pattern:

Planning Agent (agent.go) : Interprets intent, produces a JSON plan, and updates strategy based on execution feedback.

Search Subagent (subagents.go) : Calls Tavily, handles pagination, and previews results.

Analysis Subagent (subagents.go) : Reads raw search results, extracts key points, detects contradictions, and performs logical reasoning.

Report Subagent (subagents.go) : Generates a well‑structured Markdown report with citations.

Render Subagent (subagents.go) : Uses go-term-markdown to render the Markdown with colors, tables, and syntax highlighting.

TUI (cmd/agent-cli/tui.go) : Built with the bubbletea framework, offering a modern, resizable, color‑highlighted CLI similar to Claude Code or Gemini CLI.

The system also supports a human‑in‑the‑loop mode, allowing users to edit the plan or intervene during sub‑agent execution.

Future Roadmap

Long‑term memory via a vector database for Retrieval‑Augmented Generation.

Checkpoints to pause and resume agent execution.

Multimodal output such as podcasts or PPT generation.

Middleware ecosystem to extend functionality.

GoSkills Agent demonstrates that sophisticated Agentic AI can be built with pure Go, minimal dependencies, and a clear, extensible architecture.