Open-source. Self-hostable. No API keys required.
Find the right open-source AI tool.
Honest reviews and comparisons of open-source AI tools — LLM runners, image generators, coding assistants, and self-hosted stacks. No hype, just what works.
What we cover
Four categories of open-source AI tooling.
-
LLM Runners
Ollama, llama.cpp, LM Studio, vLLM, Tabby — what each does best and which one fits your setup.
-
Image Generation
ComfyUI, Automatic1111, InvokeAI, Fooocus — feature comparisons, workflow depth, and ease-of-use tradeoffs.
-
Coding Assistants
Continue.dev, Cline, Aider, Tabby — open-source alternatives to Copilot that run on your own models.
-
Self-Hosted Stack
Open WebUI, Flowise, n8n, Langfuse — the glue layer that turns individual tools into a working AI stack.
Latest posts
Fresh from the bench.
-
Gemma 4 12B Unified Multimodal Self-Hosting 2026
Self-host Gemma 4 12B — Apache 2.0 text, image, audio, and video in one model. Ollama setup on 8GB VRAM, how to pass image and audio inputs, and honest limits.
Read guide → -
Codestral 2 Review 2026: Apache 2.0 Coding on 16GB VRAM
Codestral 2 is Mistral's 22B coding model, relicensed Apache 2.0 in April 2026. Runs on 16GB VRAM via Ollama with fill-in-the-middle. Full self-host guide.
Read guide → -
GLM-5.2 Review 2026: MIT 744B Coding Model, Real VRAM Math
GLM-5.2 tops SWE-bench Pro at 62.1% under a clean MIT license. Honest VRAM math, Unsloth GGUF sizes, vLLM setup, and when the Z.ai API beats self-hosting.
Read guide → -
Browser-Use Review 2026: MIT Browser Automation for Any Local LLM
Browser-Use review 2026: MIT-licensed, 103K GitHub stars, YC W25. How it works, self-hosting with Ollama, the local-model gotchas, and where it falls short.
Read guide → -
OpenCode + Ollama Setup Guide 2026: Local Coding Agent
Wire OpenCode, the MIT-licensed terminal coding agent, to a local Ollama model. Full opencode.json config, the num_ctx trap, and a zero-cost setup.
Read guide →