Modelfile

# The agent's context window has to be set on the SERVER, not the request.
#
# Ollama's OpenAI-compatible /v1 endpoint silently drops the `options` block, so
# `extra_body={"options": {"num_ctx": 16384}}` in llm/client.py has no effect there
# (measured: `ollama ps` still reports CONTEXT 4096). Exporting OLLAMA_CONTEXT_LENGTH
# in your own shell does nothing either — it has to be in the server process's
# environment, which on macOS means launchctl and a restart.
#
# Deriving the model with the parameter baked in avoids all of that: it is one
# command, identical on every platform, and it survives whichever endpoint the
# client happens to use.
#
#   ollama pull hf.co/JetBrains/Mellum2-12B-A2.5B-Instruct-GGUF-Q4_K_M
#   ollama create agentfix-mellum2 -f Modelfile
#
# Verify with `ollama ps` after one request: the CONTEXT column must read 16384.
# `agentfix doctor` checks this for you.

FROM hf.co/JetBrains/Mellum2-12B-A2.5B-Instruct-GGUF-Q4_K_M

PARAMETER num_ctx 16384