Mixing local AI models with Claude for practical app building
An IT systems administrator with no programming experience beyond HTML and CSS has built several working apps with . Two local AI workstations provide 48GB and 56GB of VRAM and run the same model through a with a 128K . The first setup used through Cline inside VS Code.
Connecting Claude to Cline produced useful results but consumed tokens very quickly. Moving to Claude CLI made sessions last much longer, but it removed easy access to the . The next experiment is a mixed setup based on Anthropic's publicly documented method for using a local model in place of its hosted model.
The related community submission repeats the same experience and does not provide independent confirmation.
Key points
- helped a non-programmer create several working apps.
- Claude used through Cline consumed tokens very quickly.
- Sessions lasted longer when the work moved to Claude CLI.
- Claude CLI did not provide easy access to the existing .
- The current goal is a mixed workflow combining Claude with local AI hardware.