Spektiq
Can Windsurf use Ollama? Wiring a local model
Windsurf is now Devin Desktop; its docs show no native Ollama option on standard plans. Two routes that work as of November 2026, and what stays in the cloud.
Local AI, measured
Run AI on your own hardware, with numbers you can check
Practical guides to running language models on your own machine: which tools to install, what your graphics card can really hold, and what it costs compared with the cloud. Every figure comes from a dated, sourced table.
Start here
Get a first model running, check what your hardware can hold, then pick the right model: the recommended reading order.
What we cover
Four questions come back every time someone moves AI onto their own machine.
Hardware
Graphics cards, video memory, laptops: what actually limits the models you can run.
Tools and setup
Ollama, LM Studio, llama.cpp, local servers and editor integrations, set up step by step.
Choosing a model
Matching a model and its quantization to your use case and your memory budget.
Real costs
Self-hosting against paying per token: hardware, electricity and the costs nobody lists.
How we work
What we do so that you can rely on what you read.
Sourced, dated figures
Memory sizes, prices and benchmarks come from a registry where each row has its own source and reading date. Technical figures are injected from that registry, not typed by the writer; the rare illustrative value is declared and justified.
Reviewed before publication
Articles are written and fact-checked by AI agents, then critically reviewed by a separate AI reviewer. As long as a blocking issue is open, the article is not published.
Figures that expire
AI moves fast, so every table has an expiry date: prices go stale sooner than hardware specs. An article whose figures are being re-verified says so at the top.
Affiliate links disclosed
Some links to products are affiliate links. They are labeled where they appear, and the site-wide disclosure sits in the footer.
All articles
Best Local LLM for Coding (October 2026)
Three open-weight local LLMs for coding compared from their model cards, why vendor SWE-bench scores mislead, and the Ollama context trap that breaks agents.
Best local LLM: choose a model for your hardware
Skip the ranking. Pick a local LLM based on your use case, context and memory budget, then verify the fit with LM Studio and ollama ps before trusting it.
Self-Hosting AI: Work Out What a Local LLM Costs
Install a local runtime, fit a quantized model in memory, then measure your own power, speed and tariff to see whether self-hosting beats an API on cost.
Is Your GPU's VRAM Enough for a Local LLM?
Is a small-VRAM GPU enough for a local LLM? Sourced weight sizes, why context length changes the answer, and how to check where your model actually runs.
Jev: Independent Benchmarks and Real Costs
Independent Jev benchmarks reveal strong speed and cost results, plus important limits around accuracy, confidence, throughput, and local alternatives.