Open-source AI coding agents that actually work in 2026

Marker (datalab-to/marker) converts PDF, DOCX, EPUB and more to Markdown, JSON, HTML and chunks quickly and accurately. 35,694 GitHub stars, GPL-3.0 code license. Covers installation, the CLI and Python API, real code examples, LLM mode, and an honest comparison with alternatives.

GraphRAG is Microsoft''s modular, knowledge-graph-based RAG system (33,403 GitHub stars, MIT license). This guide covers installation, the init/index/query workflow, real CLI examples, and an honest comparison with LangChain and Haystack.

GPT Researcher is an open deep-research agent that runs web and local research on any task and writes cited reports. 27,473 GitHub stars, Apache-2.0. Covers installation, the async Python API, Docker, and real code examples.

Firecrawl is the open-source web data API that scrapes, crawls, maps, and searches the web into clean, LLM-ready markdown or structured JSON. 127,747 GitHub stars, AGPL-3.0. Covers install, the official SDKs, real code, self-hosting, and an honest comparison with Puppeteer, Scrapy, and Axios.

exo turns your Macs, PCs and phones into a single cluster that runs frontier AI models locally. 45,088 GitHub stars, Apache-2.0. Covers installation, the dashboard, the OpenAI/Claude/Ollama-compatible API, real commands, and an honest comparison.

Side-by-side breakdown of Claude Code and Cline — terminal autonomy vs VS Code step-by-step approval, model support, pricing, and when to pick each. The control-vs-autonomy decision for agentic coding. Updated 2026.

You already orchestrate subagents inside Claude Code. Do you actually need LangGraph, CrewAI, or AutoGen? A 2026 decision guide with real benchmarks, GitHub-star reality, and the honest line between "built-in is enough" and "time to graduate."

The complete learning + tooling stack for mastering Claude Code multi-agent workflows: 5 subagent patterns + custom agent authoring + the skill/subagent/MCP decision framework + orchestration failure modes + skill authoring. The full path from single-threaded coding to a reliable agent pipeline.

Side-by-side breakdown of the two leading agent SDKs — architecture (hooks+subagents vs handoffs+guardrails), built-in tools, OS access, voice, lock-in, and when to pick each. Updated 2026.

The complete free toolkit for SEO in the AI era: llms.txt generator + AI-crawler robots.txt + meta tags + Schema.org JSON-LD + hreflang + OG card preview. Covers both classic search (Google/Bing) and generative engines (ChatGPT, Claude, Perplexity). All browser-based, no signup.

Claude Code has three extension points — skills, subagents, and MCP servers — and they solve different problems. A decision framework for choosing the right one, with worked scenarios and the anti-patterns that waste your time.

Five real failure modes of Claude Code multi-agent pipelines — trusting unverified reports, context bleed, runaway fan-out, silent truncation, and orphaned worktrees — each with the symptom, the root cause, and the fix.