Can a local model condense a Claude Code session? I built an eval to find out
I tested five local Ollama models against Haiku, Sonnet, Codex and a no-model script on a job that looked made for them: distilling a long Claude Code session into a brief a…