front: configurable compaction model + live progress to the client

Auto-compaction can now summarize with a DIFFERENT model than the one
serving the request, with a global default (config.json `compaction`) and
a per-model override (models.json `auto_compact_model`). Empty = the
request's own model, as before.

- config: new CompactionConfig (enabled/pct/strategy/model) + round-trip
- text.py: resolve effective settings (per-model over global), resolve the
  summarizer LAZILY (only when actually over threshold, so a separate model
  isn't loaded on every request); map-reduce the dropped history into chunks
  sized to the CHOSEN summarizer's own context, reducing iteratively until
  it fits; stream status + live per-chunk progress to the client as content
  deltas (queue-bridged from the summarizer's callback)
- admin: global compaction card (settings) + per-model summarizer dropdown
  (models, shown only for the summarize strategy)

Raw two-pass path is skipped (prompt is built from system + last user turn).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

codai/admin/routes.py

@@ -2682,7 +2682,8 @@ async def api_model_configure(request: Request, username: str = Depends(require_
                 "balanced_gpu_percent", "acceleration",
                 "cache_type_k", "cache_type_v", "turboquant", "engine", "engine_fallback",
                 "quant_backend", "kv_cache_budget_mb", "kv_cache_slots", "mmproj",
-                "auto_compact", "auto_compact_pct", "auto_compact_strategy"):
+                "auto_compact", "auto_compact_pct", "auto_compact_strategy",
+                "auto_compact_model"):
         if key in data:
             entry[key] = data[key]
 
@@ -3432,6 +3433,12 @@ async def api_get_settings(username: str = Depends(require_admin)):
             "extra_env": c.ds4.extra_env,
             "auto_build": c.ds4.auto_build,
         },
+        "compaction": {
+            "enabled": c.compaction.enabled,
+            "pct": c.compaction.pct,
+            "strategy": c.compaction.strategy,
+            "model": c.compaction.model,
+        },
         "broker": {
             "enabled": c.broker.enabled,
             "base_url": c.broker.base_url,
@@ -3720,6 +3727,22 @@ async def api_save_settings(request: Request, username: str = Depends(require_ad
         if "auto_build" in d:
             c.ds4.auto_build = bool(d["auto_build"])
 
+    if "compaction" in data:
+        cp = data["compaction"] or {}
+        if "enabled" in cp:
+            c.compaction.enabled = bool(cp["enabled"])
+        if "pct" in cp:
+            try:
+                c.compaction.pct = max(50, min(99, int(cp["pct"])))
+            except (TypeError, ValueError):
+                pass
+        if "strategy" in cp:
+            _st = (cp.get("strategy") or "drop_oldest").strip()
+            if _st in ("drop_oldest", "keep_head_tail", "summarize"):
+                c.compaction.strategy = _st
+        if "model" in cp:
+            c.compaction.model = (cp.get("model") or "").strip()
+
     if "broker" in data:
         bro = data["broker"]
         c.broker.enabled = bool(bro.get("enabled", c.broker.enabled))

codai/admin/templates/models.html

@@ -707,12 +707,19 @@ window.__DEFAULT_WHISPER_SERVER_PATH__ = {{ default_whisper_server_path|tojson }
         </div>
         <div class="form-row" style="margin:0">
           <label class="form-label">Strategy</label>
-          <select id="cfg-autocompact-strategy" class="form-input">
+          <select id="cfg-autocompact-strategy" class="form-input" onchange="_toggleAutoCompact()">
             <option value="drop_oldest">Drop oldest — keep system + most recent turns</option>
             <option value="keep_head_tail">Keep head + tail — also keep the first turn, drop the middle</option>
             <option value="summarize">Summarize — replace the dropped middle with an LLM summary</option>
           </select>
         </div>
+        <div class="form-row" id="cfg-autocompact-model-row" style="margin:0;grid-column:1/-1;display:none">
+          <label class="form-label">Summarizer model <span class="muted">(summarize only)</span></label>
+          <select id="cfg-autocompact-model" class="form-input">
+            <option value="">Same as request model (default)</option>
+          </select>
+          <span class="form-hint" style="font-size:11px">A separate (e.g. smaller/faster) model can write the summary while the main model answers. The dropped history is chunked to fit the chosen model's own context. Empty = the request's own model summarizes. Leave empty to inherit the global default.</span>
+        </div>
       </div>
 
       <div style="display:flex;gap:1.5rem;flex-wrap:wrap;margin-top:.75rem">
@@ -2812,6 +2819,36 @@ async function freeDiskConfirm(idx){
 function _toggleAutoCompact(){
   const on = document.getElementById('cfg-autocompact').checked;
   document.getElementById('cfg-autocompact-opts').style.display = on ? 'grid' : 'none';
+  const strat = document.getElementById('cfg-autocompact-strategy').value;
+  const mrow = document.getElementById('cfg-autocompact-model-row');
+  if(mrow) mrow.style.display = (on && strat === 'summarize') ? 'block' : 'none';
+}
+
+// Populate the summarizer-model dropdown from the configured text/LLM models,
+// preserving the current selection.
+function _populateCompactModelSelect(selected){
+  const sel = document.getElementById('cfg-autocompact-model');
+  if(!sel) return;
+  const cur = selected != null ? selected : sel.value;
+  const seen = new Set();
+  let html = '<option value="">Same as request model (default)</option>';
+  (_localModels||[]).forEach(m=>{
+    const caps = m.capabilities||[];
+    const type = m.defaultType||'text_models';
+    const isText = type==='text_models' || type==='gguf_models' || type==='vision_models'
+      || caps.includes('chat') || caps.includes('text') || caps.includes('tools');
+    if(!isText) return;
+    const id = m.path || m.label;
+    if(!id || seen.has(id)) return;
+    seen.add(id);
+    html += `<option value="${esc(id)}">${esc(m.label||id)}</option>`;
+  });
+  // Keep a previously-saved value even if its model isn't in the local list.
+  if(cur && !seen.has(cur)){
+    html += `<option value="${esc(cur)}">${esc(cur)}</option>`;
+  }
+  sel.innerHTML = html;
+  sel.value = cur || '';
 }
 
 /* ── type checkbox helpers ─────────────────────────────── */
@@ -3197,6 +3234,7 @@ function openCfgModal(idx, cfgIdx){
   document.getElementById('cfg-autocompact').checked = !!s.auto_compact;
   document.getElementById('cfg-autocompact-pct').value = s.auto_compact_pct != null ? s.auto_compact_pct : 85;
   document.getElementById('cfg-autocompact-strategy').value = s.auto_compact_strategy || 'drop_oldest';
+  _populateCompactModelSelect(s.auto_compact_model || '');
   _toggleAutoCompact();
   document.getElementById('cfg-4bit').checked = !!s.load_in_4bit;
   document.getElementById('cfg-8bit').checked = !!s.load_in_8bit;
@@ -3586,6 +3624,7 @@ async function saveModelConfig(){
     auto_compact:      document.getElementById('cfg-autocompact').checked,
     auto_compact_pct:  parseInt(document.getElementById('cfg-autocompact-pct').value) || 85,
     auto_compact_strategy: document.getElementById('cfg-autocompact-strategy').value || 'drop_oldest',
+    auto_compact_model: document.getElementById('cfg-autocompact-model').value || '',
     max_gpu_percent:   isNaN(maxGpu) ? null : maxGpu,
     manual_ram_gb:     isNaN(ramGb) ? null : ramGb,
     load_in_4bit:      document.getElementById('cfg-4bit').checked,

codai/admin/templates/settings.html

@@ -493,6 +493,40 @@
     </div>
   </div>
 </div>
+
+<!-- Auto-compact context -->
+<div class="card">
+  <div class="card-title">Auto-compact context</div>
+  <p class="form-hint" style="margin-bottom:.6rem">Global <b>defaults</b> for shrinking an over-long chat history before generation so requests don't error out on context overflow. Per-model settings on the <b>Models</b> page override these. OFF by default.</p>
+  <div class="form-row">
+    <label style="display:flex;align-items:center;gap:.5rem;cursor:pointer">
+      <input type="checkbox" id="s-compact-enabled" onchange="toggleCompactFields()">
+      <span style="font-size:13px;font-weight:500">Enable auto-compaction by default</span>
+    </label>
+  </div>
+  <div id="compact-fields" style="display:none">
+    <div style="display:grid;grid-template-columns:1fr 1fr;gap:.75rem">
+      <div class="form-row">
+        <label class="form-label">Trigger at <span class="muted">(% of context)</span></label>
+        <input type="number" id="s-compact-pct" class="form-input" min="50" max="99" step="1" placeholder="85">
+        <span class="form-hint">Compacts to ~65% of the context window when the prompt reaches this %.</span>
+      </div>
+      <div class="form-row">
+        <label class="form-label">Strategy</label>
+        <select id="s-compact-strategy" class="form-input">
+          <option value="drop_oldest">Drop oldest — keep system + most recent turns</option>
+          <option value="keep_head_tail">Keep head + tail — also keep the first turn, drop the middle</option>
+          <option value="summarize">Summarize — replace the dropped middle with an LLM summary</option>
+        </select>
+      </div>
+    </div>
+    <div class="form-row">
+      <label class="form-label">Summarizer model <span class="muted">(summarize strategy)</span></label>
+      <input type="text" id="s-compact-model" class="form-input" placeholder="(empty = the model serving the request)">
+      <span class="form-hint">Model id/alias that writes the summary. Empty = the request's own model summarizes itself. Point it at a smaller/faster model to summarize old turns while the main model answers — the dropped history is chunked to fit the chosen model's own context window. The client receives a status message and live progress while compaction runs.</span>
+    </div>
+  </div>
+</div>
 {% endblock %}
 
 {% block scripts %}
@@ -561,6 +595,10 @@ function toggleDs4Fields(){
   document.getElementById('ds4-fields').style.display =
     document.getElementById('s-ds4-enabled').checked ? 'block' : 'none';
 }
+function toggleCompactFields(){
+  document.getElementById('compact-fields').style.display =
+    document.getElementById('s-compact-enabled').checked ? 'block' : 'none';
+}
 function toggleHttps(){
   document.getElementById('https-fields').style.display =
     document.getElementById('s-https').checked ? 'block' : 'none';
@@ -675,6 +713,13 @@ async function loadSettings(){
     document.getElementById('s-broker-reconnect-max').value = broker.reconnect_max_delay_seconds ?? 60;
     document.getElementById('s-broker-ws-ping').value = broker.websocket_ping_interval ?? 20;
     toggleBrokerFields();
+    // Auto-compact context
+    const compact = d.compaction || {};
+    document.getElementById('s-compact-enabled').checked = !!compact.enabled;
+    document.getElementById('s-compact-pct').value = compact.pct ?? 85;
+    document.getElementById('s-compact-strategy').value = compact.strategy || 'drop_oldest';
+    document.getElementById('s-compact-model').value = compact.model ?? '';
+    toggleCompactFields();
     // Thermal protection
     const therm = d.thermal || {};
     document.getElementById('s-therm-gpu-enabled').checked = therm.gpu_enabled !== false;
@@ -788,6 +833,12 @@ async function saveSettings(){
       expert_cache_reserve_gb: parseInt(document.getElementById('s-ds4-expert-cache-reserve').value) || 0,
       extra_env: document.getElementById('s-ds4-extra-env').value.trim(),
     },
+    compaction:{
+      enabled: document.getElementById('s-compact-enabled').checked,
+      pct: parseInt(document.getElementById('s-compact-pct').value) || 85,
+      strategy: document.getElementById('s-compact-strategy').value || 'drop_oldest',
+      model: document.getElementById('s-compact-model').value.trim(),
+    },
     broker:{
       enabled: document.getElementById('s-broker-enabled').checked,
       base_url: document.getElementById('s-broker-base-url').value.trim(),

codai/config.py

@@ -214,6 +214,24 @@ class EnhanceConfig:
     allow_rife_ncnn: bool = False     # allow the external rife-ncnn-vulkan binary instead of a torch model
 
 
+@dataclass
+class CompactionConfig:
+    """Global defaults for auto-compaction of an over-long chat history.
+
+    Per-model settings in a models.json entry (``auto_compact``,
+    ``auto_compact_pct``, ``auto_compact_strategy``, ``auto_compact_model``)
+    OVERRIDE the values here; when a model leaves one unset, the global default
+    below applies. ``model`` selects which model performs the summarization for
+    the ``summarize`` strategy — empty means use the same model that serves the
+    request. Pointing it at a smaller/faster model lets that model summarize the
+    old turns while the big model answers; the dropped history is chunked to fit
+    the chosen summarizer's own context window before it is summarized."""
+    enabled: bool = False
+    pct: int = 85                        # compact when the prompt reaches this % of n_ctx
+    strategy: str = "drop_oldest"        # drop_oldest | keep_head_tail | summarize
+    model: str = ""                      # model id/alias that summarizes; "" = same as request
+
+
 @dataclass
 class Ds4Config:
     """DeepSeek V4 via ds4 (antirez/DwarfStar) external-worker configuration.
@@ -273,6 +291,7 @@ class Config:
     jobs: JobsConfig = field(default_factory=JobsConfig)
     enhance: EnhanceConfig = field(default_factory=EnhanceConfig)
     ds4: Ds4Config = field(default_factory=Ds4Config)
+    compaction: CompactionConfig = field(default_factory=CompactionConfig)
     broker: BrokerConfig = field(default_factory=BrokerConfig)
     system_prompt: Optional[str] = None
     tools_closer_prompt: bool = False
@@ -457,6 +476,7 @@ class ConfigManager:
                 jobs=_dc(JobsConfig, config_data.get("jobs", {})),
                 enhance=_dc(EnhanceConfig, config_data.get("enhance", {})),
                 ds4=_dc(Ds4Config, config_data.get("ds4", {})),
+                compaction=_dc(CompactionConfig, config_data.get("compaction", {})),
                 broker=_dc(BrokerConfig, config_data.get("broker", {})),
                 system_prompt=config_data.get("system_prompt"),
                 tools_closer_prompt=config_data.get("tools_closer_prompt", False),
@@ -625,6 +645,12 @@ class ConfigManager:
                 "extra_env": self.config.ds4.extra_env,
                 "auto_build": self.config.ds4.auto_build,
             },
+            "compaction": {
+                "enabled": self.config.compaction.enabled,
+                "pct": self.config.compaction.pct,
+                "strategy": self.config.compaction.strategy,
+                "model": self.config.compaction.model,
+            },
             "broker": {
                 "enabled": self.config.broker.enabled,
                 "base_url": self.config.broker.base_url,