Ollama, 9router and OpenHands: self-hosted stack overview

The speaker outlines a developer stack built around five open source projects, with options for running models on your own hardware and using commercial model providers. This is an overview of how the tools fit together, rather than a step-by-step installation guide.

Ollama supplies a command-line interface and API to download and run models locally. The speaker presents local inference as a way to keep prompts private, but notes that hardware limits which models you can use. Smaller models are more practical on ordinary machines; frontier-sized models require much more infrastructure.

9router provides an OpenAI-compatible proxy through a local endpoint. Its fallback tiers can move requests between an existing subscription, a paid API and other providers. A localhost endpoint does not mean those remote models run on your machine. Headroom addresses token consumption by compressing tool output and logs before they reach a model, with a local cache for retrieving content when needed.

Dify adds a visual workflow builder whose workflows can become APIs. OpenHands handles autonomous coding tasks and can use OpenAI or Anthropic models, or connect to Ollama. The speaker also promotes sponsor Hostinger's VPS and Docker catalog as a deployment option. Claims about savings and productivity reflect the speaker's account, without a measured comparison in the video.