Run Claude Code locally on Mac.

Step 1 · Launch via Ollama

One command to start.

$ ollama launch claude

Step 2 · Pre-flight memory check

Will your Mac fit it?

Claude Code pairs best with Llama 3.3 70B at Q4_K_M (~40.6 GB). Add ~4–8 GB on top for the agent's working set, plus headroom for the rest of your dev stack. Recommended: 64 GB Mac (minimum 32 GB).

$ devpulse ai --before-load 41574 --auto-clean
before: Won't fit — 6.4 GB short
  - unloaded idle ollama model: qwen2.5:7b (4.2 GB)
  - killed 4 zombie procs (612 MB reclaimed)
after:  Fits comfortably — 3.1 GB headroom

# safe to launch
$ ollama launch claude

Why this matters: Fully agentic + large model = real OOM risk. Pre-flight every launch.

Step 3 · Babysit long sessions

Don't lose progress to OOM mid-task.

Agent runs that span hours hit memory pressure as context grows. devpulse babysit watches and auto-cleans without crashing the session.

# in another tmux pane
$ devpulse babysit --target-free-mb 8192 --json > claude-code.log &

# emits NDJSON tick / cleanup events you can checkpoint on:
{"event":"tick","tickNum":47,"availableForAIMB":7200,"pressure":"free<8192MB",...}
{"event":"cleanup","reasons":"free<8192MB","reclaimedMB":5400,...}

Tips for Claude Code

What we've learned running it.

• Subagents fan out — context budget multiplies fast. Pre-flight with `devpulse ai --before-load <MB>` before kicking off long sessions
• If pairing with Llama 3.3 70B locally, count on ~42 GB just for the model — leave 16 GB+ for the agent's working set
• Babysit long runs: `devpulse babysit --target-free-mb 8192 --json` reclaims when context expansion squeezes free VRAM

Recommended models

Best local pairings for Claude Code.

Related agents