One engine, many tools
Compatible servers
Ollama and LM Studio run local servers that speak the same 'language' (API) as OpenAI's and Anthropic's services. Many tools let you change the address they talk to. Point them at your own machine and they use your local model instead. Ollama added support for Claude Desktop as a gateway in 2026, and LM Studio has a guide for Claude Code.
Coding with local models
Local models are decent at explaining code, spotting obvious bugs and writing small functions, and they keep proprietary code on your machine. They are weaker on large, multi-file changes than the best cloud models. A common pattern is local for review and routine work, cloud for the hardest problems.
Agents on your own hardware
Open-source agents such as Hermes Agent can run with a local model, so tasks don't run up a per-token bill. Agent work is demanding: it uses long contexts and many steps, so it needs a capable model and plenty of memory. The safety rules are the same as for any agent: limited access, approval before sending or deleting, and an easy off switch.
The real costs
Local isn't free: there's the hardware, electricity for long runs, your set-up time, and slower results. For occasional use, a cloud subscription can be cheaper. For heavy private use, or when data can't leave the building, local wins. Our Hermes Agent cost guide compares local and cloud running costs with real numbers.
Which AI tool do you already use that you'd like to run more privately or more cheaply?