diff --git a/.claude/CLAUDE.md b/.claude/CLAUDE.md
index 118cf02..b66fb52 100644
--- a/.claude/CLAUDE.md
+++ b/.claude/CLAUDE.md
@@ -241,7 +241,7 @@ Vault structure: `infrastructure/`, `clients/`, `services/`, `projects/`, `msp-t
 
 ## Local AI (Ollama)
 
-Tier 0 — **Ollama is the documentation engine.** Route all prose generation through it: session log narratives, commit messages, ticket comments, client notes, code docs. Claude reviews output, owns credentials/facts/execution.
+Tier 0 — **Ollama is the documentation engine.** Route prose generation through it: commit messages, ticket comments, client notes, code docs. Claude reviews output, owns credentials/facts/execution. Session log narratives are written directly by Claude (Ollama too slow for /save).
 
 - **DESKTOP-0O8A1RL:** `http://localhost:11434`
 - **Other machines:** `http://100.92.127.64:11434` (Tailscale required)
diff --git a/.claude/commands/save.md b/.claude/commands/save.md
index 16b2c86..a9ebc69 100644
--- a/.claude/commands/save.md
+++ b/.claude/commands/save.md
@@ -6,68 +6,19 @@ Save a comprehensive session log to the appropriate `session-logs/` directory, t
 
 ## Phase 1 — Generate the narrative
 
-Ollama drafts the prose sections (qwen3:14b at `localhost:11434` or Tailscale `100.92.127.64:11434`). Claude owns the factual sections — never delegate facts to Ollama.
+Claude writes all sections directly. Be concise, factual, technical. No filler phrases. Past tense. No emojis.
 
 | Author | Sections |
 |---|---|
-| Ollama | Session Summary, Key Decisions, Problems Encountered |
-| Claude (verbatim) | User block, Configuration Changes, Credentials, Infrastructure, Commands & Outputs, Pending Tasks, References |
+| Claude | All sections |
 
-### Resolve Ollama endpoint
+### Narrative sections (Claude writes directly)
 
-```bash
-if curl -s -m 2 http://localhost:11434/api/tags >/dev/null 2>&1; then
-  OLLAMA="http://localhost:11434"
-elif curl -s -m 3 http://100.92.127.64:11434/api/tags >/dev/null 2>&1; then
-  OLLAMA="http://100.92.127.64:11434"
-else
-  OLLAMA=""  # Claude drafts narrative directly
-fi
-```
+**Session Summary** — 3-5 paragraphs: what was accomplished, in what order, why.
 
-### Draft via Ollama
+**Key Decisions** — bullet list of non-obvious decisions and their rationale.
 
-**Always delete the prompt file first** — prior `/save` runs leave a stale `save_narrative_prompt.txt` in `%LOCALAPPDATA%/Temp`, and the Write tool refuses to overwrite without a Read. The silent fallback is `py` reading the leftover file and Ollama producing a perfectly-coherent narrative about *the previous session's work*. Documented in `.claude/memory/feedback_tmp_path_windows.md`.
-
-```bash
-PROMPT_FILE="$LOCALAPPDATA/Temp/save_narrative_prompt.txt"
-rm -f "$PROMPT_FILE"  # critical — see above
-# ... then Write the prompt with this session's WORK DONE bullets ...
-```
-
-Prompt template:
-
-```
-You are a technical session log writer for an MSP (Arizona Computer Guru). Write three sections in markdown. Be concise, factual, technical. No filler phrases. Use past tense. No emojis.
-
-WORK DONE THIS SESSION:
-<paste bullet list of what happened>
-
-Write these three sections only:
-
-## Session Summary
-<3-5 paragraph narrative: what was accomplished, in what order, why>
-
-## Key Decisions
-<bullet list of non-obvious decisions and their rationale>
-
-## Problems Encountered
-<bullet list of problems hit and how each was resolved; omit if none>
-```
-
-Run via Python (PyJWT-free path):
-
-```bash
-py -c "
-import urllib.request, json
-prompt = open('$PROMPT_FILE', encoding='utf-8').read()
-body = json.dumps({'model':'qwen3:14b','messages':[{'role':'user','content':prompt}],'stream':False,'think':False}).encode()
-res = json.loads(urllib.request.urlopen(urllib.request.Request('$OLLAMA/api/chat', body), timeout=300).read())
-print(res['message']['content'])
-"
-```
-
-Claude reviews the output before assembling. Common Ollama issues to fix on review: invented compliance/risk language, hallucinated reasoning, off-task content if the prompt file was stale.
+**Problems Encountered** — bullet list of problems hit and how each was resolved. Omit section if none.
 
 ---
 
diff --git a/.claude/memory/MEMORY.md b/.claude/memory/MEMORY.md
index e3dd35f..e5bb31e 100644
--- a/.claude/memory/MEMORY.md
+++ b/.claude/memory/MEMORY.md
@@ -26,6 +26,7 @@
 - [Bypass Permissions Setting](feedback_bypass_permissions_setting.md) - Set permissions.defaultMode to bypassPermissions in settings.json on all machines
 - [365 Remediation Tool](feedback_365_remediation_tool.md) - Always means Graph API app fabb3421, not CIPP
 - [Ollama Tier-0 Routing](feedback_ollama_tier0_routing.md) - Route drafts/summaries/classifications through Ollama (qwen3:14b). Mike designed ClaudeTools this way — not optional.
+- [/save writes narrative directly](feedback_save_no_ollama.md) — No Ollama for /save; write all sections inline — too slow
 - [Syncro Emergency Billing](feedback_syncro_emergency_billing.md) — Emergency = 1.5× multiplier, not additive. Branch by `customer.prepay_hours`: no-prepaid → `26184` at actual hrs; prepaid → `26118` at hrs×1.5. Never stack. Always set `price_retail`.
 - [Identity precedence](feedback_identity_precedence.md) — Trust `.claude/identity.json` over the system-reminder `userEmail` hint when they disagree (shared-login machines).
 - [1Password — always use service token](feedback_1password_service_token.md) — Source OP_SERVICE_ACCOUNT_TOKEN from SOPS for every `op` call. Desktop-app integration prompts are unacceptable in agent flows.
diff --git a/.claude/memory/feedback_save_no_ollama.md b/.claude/memory/feedback_save_no_ollama.md
new file mode 100644
index 0000000..af36033
--- /dev/null
+++ b/.claude/memory/feedback_save_no_ollama.md
@@ -0,0 +1,11 @@
+---
+name: /save writes narrative directly — no Ollama
+description: Claude writes all /save session log sections directly; Ollama removed from save workflow
+type: feedback
+---
+
+Do NOT use Ollama for /save session log narrative sections (Session Summary, Key Decisions, Problems Encountered). Write them directly. Ollama takes too long.
+
+**Why:** User explicitly removed Ollama from the /save documentation role on 2026-05-12 — the Ollama round-trip added unacceptable latency to every /save.
+
+**How to apply:** On every /save invocation, write all narrative sections yourself inline. Do not curl Ollama, do not write a prompt file, do not spawn a py subprocess. Ollama is still available for other prose tasks (commit messages, ticket comments, client notes) but not /save.