Claude Code + Gemma 4 --> local and free

[Antonio Linares]

1. download Ollama from:

https://ollama.com/

1. From Ollama app download gemma4:e4b

2. Create a Modelfile file with these contents: (we increase the context up to 32K)

   Code: Select all | Expand

   FROM gemma4:e4b
   PARAMETER num_ctx 32768

3. ollama create gemma4-e4b-32k -f Modelfile

4. ollama launch claude --model gemma4-e4b-32k

5. From Claude Code do /permissions and select 1. Add a new rule... then write: Bash(ls:*)

[Antonio Linares]

If you have enough harddisk space for it then use:

gemma4:26b

Modelfile

Code: Select all | Expand

FROM gemma4:26b
PARAMETER num_ctx 32768

ollama launch claude --model gemma4-26b-32k

same process as above

[Antonio Linares]

[Antonio Linares]

Very far away from the real Claude Code experience...

Muy lejos de la verdadera experiencia con Claude Code...

[Antonio Linares]

https://aitoolanalysis.com/claude-code-router-review/

You type in Claude Code → CCR intercepts → Routes to DeepSeek/Gemini/Ollama/etc. → Response transformed → Appears in Claude Code

[Antonio Linares]

Claude Code + DeepSeek:

Code: Select all | Expand

export ANTHROPIC_BASE_URL="https://api.deepseek.com/anthropic"
export ANTHROPIC_AUTH_TOKEN="tu_clave_api_aquí"
export ANTHROPIC_MODEL="deepseek-chat"
# Opcional: define el modelo secundario para tareas ligeras
export ANTHROPIC_DEFAULT_HAIKU_MODEL="deepseek-chat"

claude

Code: Select all | Expand

export ANTHROPIC_BASE_URL="https://api.deepseek.com/anthropic"
export ANTHROPIC_AUTH_TOKEN="tu_clave_api_aquí"
export ANTHROPIC_MODEL="deepseek-reasoner"
# Importante: Aumenta el tiempo de espera porque R1 tarda en "pensar"
export API_TIMEOUT_MS=600000 

claude