Local Zero-Cost AI Agent Stack
ollama + llama_index + chroma
Read the stack →The Verified Local AI Stack
The only local AI stack where every component has been independently verified.
Most “local AI stack” guides recommend tools the author never actually tested.
The setup guide looks clean. The README is polished. The star count is impressive. But nobody followed those instructions on a clean machine and recorded what actually happened. You find out the tool is broken three hours into debugging — not in the review.
Every component on this page was tested differently. Clone the repo. Follow the README exactly. Test the core claim on a clean machine. Record what actually happens. No shortcuts. No assumed fixes.
Twelve tools tested. Twelve SOLID.
| Tool | Verdict | Role in Stack |
|---|---|---|
| ollama | SOLID | Local LLM inference |
| llama_index | SOLID | Document indexing + RAG |
| chroma | SOLID | Local vector store |
| lobe-chat | SOLID | Local chat UI |
| fastapi | SOLID | API layer for Python AI apps |
| ScrapeGraphAI | SOLID | AI-guided web scraping |
| crewAI | SOLID | Multi-agent orchestration |
| litellm | SOLID | Unified LLM interface |
| crawl4ai | SOLID | Web content extraction |
Every verdict links to the full test evidence. Every error is documented verbatim.
Three more tools verified. Different category.
These aren’t components of the local AI stack — they’re tools for developers building with Claude Code.
| Tool | Verdict | Role |
|---|---|---|
| everything-claude-code | SOLID | Claude Code agent harness |
| andrej-karpathy-skills | SOLID | Claude Code skill library |
| gstack | SOLID | Claude Code workflow framework |
Four stacks. All verified components.
Web Intelligence Agent Stack
ScrapeGraphAI + llama_index + chroma
Read the stack →Multi-Agent Orchestration Stack
crewAI + ollama
Read the stack →Local ChatGPT Alternative Stack
lobe-chat + ollama + llama_index + chroma
Read the stack →Three use cases shipped with this stack.
Local AI File Reader
Ask questions about your own files. Runs entirely offline.
See the use case →Vibe Coding Price Compare
Scrapes and compares AI coding tool pricing in plain English. No API key. Free to run.
See the use case →Personal Notes AI
Ask your Obsidian vault questions. Answers grounded in your own notes. No subscription.
See the use case →Every review follows RepoVerifier Standard v1.0.
Clone it. Follow the README. Test the core claim. Publish the evidence.