From f7ef090d7c61b56453524e6df7c79db7b615f6db Mon Sep 17 00:00:00 2001
From: Claude <noreply@anthropic.com>
Date: Sun, 22 Mar 2026 18:30:43 +0000
Subject: [PATCH] Add context awareness: token tracker + compaction + 6GB GPU
 tier
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Context management for local models with limited windows:
- Context Tracker (community function): shows tokens used vs available,
  progress bar, percentage remaining — so you see the cliff coming
- Checkpoint Summarization Filter: auto-summarizes old messages when
  context fills up, like Claude's auto-compaction
- Both added as recommended post-install links in setup output
  (Open WebUI Functions install with one click from the UI)

6GB GPU tier (Quadro P3300, GTX 1060, etc.):
- Qwen 3.5 4B at Q4_K_M = ~2.5GB weights, leaves 3.5GB for KV cache
- With Q8 KV cache: ~32K usable context on 6GB
- Better than squeezing 9B into nothing — more context > slightly smarter

https://claude.ai/code/session_01PtYTPherSJaxDEVPgF6Nxu
---
 laptop_full_setup.sh   | 11 +++++++++++
 local-ai-setup.sh      | 13 +++++++++++--
 ubuntu-post-install.sh |  3 +++
 3 files changed, 25 insertions(+), 2 deletions(-)

diff --git a/laptop_full_setup.sh b/laptop_full_setup.sh
index 4fa43f9..dbe2709 100755
--- a/laptop_full_setup.sh
+++ b/laptop_full_setup.sh
@@ -66,6 +66,9 @@ elif [[ "$TOTAL_VRAM" -ge 14 ]]; then
 elif [[ "$TOTAL_VRAM" -ge 8 ]]; then
     CHAT_MODEL="qwen3.5:9b";  CODE_MODEL="qwen3.5:9b"
     CTX=32768;  GPU_TIER="${TOTAL_VRAM}GB VRAM — 9B dense, 32K context"
+elif [[ "$TOTAL_VRAM" -ge 6 ]]; then
+    CHAT_MODEL="qwen3.5:4b";  CODE_MODEL="qwen3.5:4b"
+    CTX=32768;  GPU_TIER="${TOTAL_VRAM}GB VRAM — 4B + KV quant, 32K context (3.5GB free for cache)"
 elif [[ "$TOTAL_VRAM" -ge 4 ]]; then
     CHAT_MODEL="qwen3.5:4b";  CODE_MODEL="qwen3.5:4b"
     CTX=16384;  GPU_TIER="${TOTAL_VRAM}GB VRAM — 4B models, 16K context"
@@ -1391,6 +1394,14 @@ if $INSTALL_AI; then
         echo "    Via MCP:      gitea_github_sync(mode='all')"
     }
 fi
+if $INSTALL_AI; then
+    echo ""
+    echo -e "  ${YELLOW}Recommended Open WebUI Functions (install from Admin → Functions → ＋):${NC}"
+    echo "    Context tracker:    https://openwebui.com/f/centrisic/context_tracker"
+    echo "      → Shows tokens used vs available, progress bar, context % remaining"
+    echo "    Context compaction: https://openwebui.com/f/projectmoon/checkpoint_summarization_filter"
+    echo "      → Auto-summarizes old messages when context fills up (like Claude)"
+fi
 if $SVC_KIWIX && [[ "$ZIM_CHOICE" == "3" ]]; then
     echo -e "  ${YELLOW}ZIM downloads:${NC}      ./kiwix_download.sh  (not started)"
 elif $SVC_KIWIX; then
diff --git a/local-ai-setup.sh b/local-ai-setup.sh
index ceff246..675952c 100755
--- a/local-ai-setup.sh
+++ b/local-ai-setup.sh
@@ -42,6 +42,9 @@ elif [[ "$TOTAL_VRAM" -ge 14 ]]; then
 elif [[ "$TOTAL_VRAM" -ge 8 ]]; then
     CHAT_MODEL="qwen3.5:9b";  CODE_MODEL="qwen3.5:9b"
     CTX=32768;  TIER="${TOTAL_VRAM}GB — 9B dense, 32K context"
+elif [[ "$TOTAL_VRAM" -ge 6 ]]; then
+    CHAT_MODEL="qwen3.5:4b";  CODE_MODEL="qwen3.5:4b"
+    CTX=32768;  TIER="${TOTAL_VRAM}GB — 4B + KV quant, 32K context"
 elif [[ "$TOTAL_VRAM" -ge 4 ]]; then
     CHAT_MODEL="qwen3.5:4b";  CODE_MODEL="qwen3.5:4b"
     CTX=16384;  TIER="${TOTAL_VRAM}GB — 4B models, 16K context"
@@ -898,7 +901,13 @@ echo ""
 echo -e "  ${YELLOW}Aider CLI:${NC}  cd your-repo && $BASE/aider.sh <files>"
 echo    "  (or open browser UI above — both use your local code model)"
 echo ""
-echo -e "  ${YELLOW}Save Claude usage:${NC} use local 14B for boilerplate, docs,"
-echo    "  simple fixes. Use Claude Sonnet 4.6 for hard bugs,"
+echo -e "  ${YELLOW}Recommended Open WebUI Functions${NC} (install from Admin → Functions → ＋):"
+echo    "    Context tracker:    https://openwebui.com/f/centrisic/context_tracker"
+echo    "      → Shows tokens used vs available, progress bar, context % remaining"
+echo    "    Context compaction: https://openwebui.com/f/projectmoon/checkpoint_summarization_filter"
+echo    "      → Auto-summarizes old messages when context fills up (like Claude)"
+echo ""
+echo -e "  ${YELLOW}Save Claude usage:${NC} use local models for boilerplate, docs,"
+echo    "  simple fixes. Use Claude for hard bugs,"
 echo    "  multi-file refactoring, architecture decisions."
 echo ""
diff --git a/ubuntu-post-install.sh b/ubuntu-post-install.sh
index eeda5ad..80ca43e 100644
--- a/ubuntu-post-install.sh
+++ b/ubuntu-post-install.sh
@@ -1125,6 +1125,9 @@ elif [[ "$TOTAL_VRAM" -ge 14 ]]; then
 elif [[ "$TOTAL_VRAM" -ge 8 ]]; then
     CHAT_MODEL="qwen3.5:9b";  CODE_MODEL="qwen3.5:9b"
     CTX=32768;  TIER="${TOTAL_VRAM}GB — 9B dense, 32K context"
+elif [[ "$TOTAL_VRAM" -ge 6 ]]; then
+    CHAT_MODEL="qwen3.5:4b";  CODE_MODEL="qwen3.5:4b"
+    CTX=32768;  TIER="${TOTAL_VRAM}GB — 4B + KV quant, 32K context"
 elif [[ "$TOTAL_VRAM" -ge 4 ]]; then
     CHAT_MODEL="qwen3.5:4b";  CODE_MODEL="qwen3.5:4b"
     CTX=16384;  TIER="${TOTAL_VRAM}GB — 4B models, 16K context"