sapper-iq 1.4.8 → 1.4.10

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "sapper-iq",
-  "version": "1.4.8",
+  "version": "1.4.10",
   "description": "AI-powered development assistant that executes commands and builds projects",
   "main": "sapper.mjs",
   "bin": {

package/sapper-ui.mjs

@@ -2752,10 +2752,17 @@ function setEditorMode(path) {
 
 // ─── Config tab ──────────────────────────────────────────────
 window.reloadConfig = function() {
-  fetch('/api/config').then(function(r){return r.json();}).then(function(d){
-    document.getElementById('cfgJson').value = JSON.stringify(d.config || {}, null, 2);
-    renderQuickConfig(d.config || {});
-  }).catch(function(e){ showToast('Config read failed: ' + e.message, 'err'); });
+  // Fetch local Ollama models first so the model dropdowns can populate, then load config.
+  var modelsPromise = fetch('/api/models')
+    .then(function(r){ return r.json(); })
+    .then(function(d){ window._sapperModels = (d && d.models) || []; })
+    .catch(function(){ window._sapperModels = window._sapperModels || []; });
+  modelsPromise.then(function(){
+    fetch('/api/config').then(function(r){return r.json();}).then(function(d){
+      document.getElementById('cfgJson').value = JSON.stringify(d.config || {}, null, 2);
+      renderQuickConfig(d.config || {});
+    }).catch(function(e){ showToast('Config read failed: ' + e.message, 'err'); });
+  });
 };
 window.saveConfig = function() {
   var raw = document.getElementById('cfgJson').value;
@@ -2777,15 +2784,36 @@ function renderQuickConfig(cfg) {
   host.innerHTML = '';
   function add(html) { host.insertAdjacentHTML('beforeend', html); }
   var T = function(tip, below) { return '<span class="tip' + (below ? ' below' : '') + '" data-tip="' + tip.replace(/"/g, '&quot;') + '">!</span>'; };
-  add('<label>Default model ' + T('The AI model used when no model is specified at startup. E.g. gpt-4o, claude-3-5-sonnet, qwen3.5. Leave blank to be prompted on launch.') + '</label><input type="text" id="qDefMod" placeholder="auto" value="' + esc(cfg.defaultModel || '') + '">');
+
+  // Build a <select> populated from local Ollama models, with a free-text fallback
+  // for models that are not (yet) installed locally.
+  function modelSelect(id, currentVal, allowEmpty) {
+    var opts = '';
+    if (allowEmpty) opts += '<option value="">(auto / prompt at launch)</option>';
+    var seen = false;
+    (window._sapperModels || []).forEach(function(m){
+      var sel = (m.name === currentVal) ? ' selected' : '';
+      if (m.name === currentVal) seen = true;
+      opts += '<option value="' + esc(m.name) + '"' + sel + '>' + esc(m.name) + (m.size ? ' \u2014 ' + esc(m.size) : '') + '</option>';
+    });
+    if (currentVal && !seen) opts += '<option value="' + esc(currentVal) + '" selected>' + esc(currentVal) + ' (not installed locally)</option>';
+    return '<select id="' + id + '">' + opts + '</select>';
+  }
+
+  var cConsult = (cfg.consultant && cfg.consultant.model) || '';
+  var cThink = (cfg.deepthink && cfg.deepthink.model) || '';
+
+  add('<label>Default model ' + T('The AI model used when no model is specified at startup. Pick one from your local Ollama models.') + '</label>' + modelSelect('qDefMod', cfg.defaultModel || '', true));
   add('<label>Default agent ' + T('Agent role (.sapper/agents/*.md) to activate automatically when Sapper starts. Leave blank for the default general-purpose assistant.') + '</label><input type="text" id="qDefAgent" placeholder="(none)" value="' + esc(cfg.defaultAgent || '') + '">');
+  add('<label>Consultant model ' + T('Heavyweight model used by the consult_expert tool. Pick something stronger than your main model \u2014 the consultant is supposed to be a deeper reasoner.') + '</label>' + modelSelect('qConsultMod', cConsult, false));
+  add('<label>Deep-think model ' + T('Model used by the deep_think tool for multi-turn back-and-forth reasoning. Pick a strong reasoning model.') + '</label>' + modelSelect('qThinkMod', cThink, false));
   add('<label>Context limit ' + T('Hard cap on tokens sent to the model per request. Leave blank to use the model\\'s full context window. Useful to reduce cost or avoid slow responses.') + '</label><input type="number" id="qCtxLim" value="' + esc(cfg.contextLimit == null ? '' : cfg.contextLimit) + '">');
   add('<label>Tool round limit ' + T('Maximum number of tool calls (file reads, shell commands, patches…) Sapper may make in a single response turn. Default is 40. Lower to prevent runaway loops.') + '</label><input type="number" id="qToolRnd" value="' + esc(cfg.toolRoundLimit != null ? cfg.toolRoundLimit : 40) + '">');
   add('<div class="toggle-row"><span>Summary phases ' + T('When ON, Sapper displays a step-by-step progress bar while it compresses long conversations. Turn OFF for a quieter experience.') + '</span><div class="switch ' + (cfg.summaryPhases ? 'on' : '') + '" id="qSumPh"></div></div>');
   add('<label>Summary trigger % ' + T('When the conversation reaches this percentage of the context window, Sapper automatically summarizes older messages to keep the window from overflowing. Default 65%.') + '</label><input type="number" id="qSumTr" value="' + esc(cfg.summarizeTriggerPercent != null ? cfg.summarizeTriggerPercent : 65) + '">');
   add('<div class="toggle-row"><span>Debug mode ' + T('Enables verbose output — shows raw tool call details, API request sizes, and internal errors. Useful for troubleshooting but noisy during normal use.') + '</span><div class="switch ' + (cfg.debug ? 'on' : '') + '" id="qDebug"></div></div>');
   add('<div class="toggle-row"><span>Auto-attach files ' + T('When ON, files you open in the sidebar are automatically referenced in the AI context so Sapper knows what you are looking at without you typing @filename.') + '</span><div class="switch ' + (cfg.autoAttach !== false ? 'on' : '') + '" id="qAutoAtt"></div></div>');
-  add('<div class="row-btns"><button class="primary" onclick="saveQuickConfig()">Apply quick changes</button></div>');
+  add('<div class="row-btns"><button onclick="refreshModelList()" title="Re-fetch local Ollama models">\u21bb Refresh models</button><button class="primary" onclick="saveQuickConfig()">Apply quick changes</button></div>');
 
   function bindSwitch(id) {
     var el = document.getElementById(id);
@@ -2794,6 +2822,18 @@ function renderQuickConfig(cfg) {
   bindSwitch('qSumPh'); bindSwitch('qDebug'); bindSwitch('qAutoAtt');
 }
 
+window.refreshModelList = function() {
+  fetch('/api/models').then(function(r){return r.json();}).then(function(d){
+    window._sapperModels = (d && d.models) || [];
+    // Re-render with the current config so the dropdowns repopulate
+    try {
+      var cfg = JSON.parse(document.getElementById('cfgJson').value || '{}');
+      renderQuickConfig(cfg);
+      showToast('Loaded ' + window._sapperModels.length + ' local model' + (window._sapperModels.length === 1 ? '' : 's'));
+    } catch (e) {}
+  }).catch(function(e){ showToast('Could not list models: ' + e.message, 'err'); });
+};
+
 window.saveQuickConfig = function() {
   var current;
   try { current = JSON.parse(document.getElementById('cfgJson').value || '{}'); }
@@ -2803,6 +2843,10 @@ window.saveQuickConfig = function() {
   var v;
   v = val('qDefMod').trim();  current.defaultModel = v || null;
   v = val('qDefAgent').trim(); current.defaultAgent = v || null;
+  v = val('qConsultMod').trim();
+  if (v) { current.consultant = Object.assign({}, current.consultant || {}, { model: v }); }
+  v = val('qThinkMod').trim();
+  if (v) { current.deepthink = Object.assign({}, current.deepthink || {}, { model: v }); }
   v = val('qCtxLim').trim();   current.contextLimit = v === '' ? null : parseInt(v, 10);
   v = val('qToolRnd').trim();  current.toolRoundLimit = v === '' ? 40 : parseInt(v, 10);
   v = val('qSumTr').trim();    current.summarizeTriggerPercent = v === '' ? 65 : parseInt(v, 10);
@@ -3393,7 +3437,23 @@ const server = http.createServer(async (req, res) => {
       const body = await readReqJSON(req);
       try {
         ensureDir(SAPPER_DIR);
-        fs.writeFileSync(CONFIG_FILE, JSON.stringify(body.config || {}, null, 2));
+        // Deep-merge incoming over existing on-disk config so that fields the
+        // Web UI doesn't know about (e.g. deepthink, consultant.verbose) are
+        // preserved instead of being silently wiped by a stale textarea save.
+        const existing = readJSON(CONFIG_FILE, {}) || {};
+        const incoming = body.config || {};
+        const isPlainObj = (v) => v && typeof v === 'object' && !Array.isArray(v);
+        const deepMerge = (base, over) => {
+          if (!isPlainObj(base)) return over;
+          if (!isPlainObj(over)) return over;
+          const out = { ...base };
+          for (const k of Object.keys(over)) {
+            out[k] = isPlainObj(base[k]) && isPlainObj(over[k]) ? deepMerge(base[k], over[k]) : over[k];
+          }
+          return out;
+        };
+        const merged = deepMerge(existing, incoming);
+        fs.writeFileSync(CONFIG_FILE, JSON.stringify(merged, null, 2));
         return json(res, { ok: true });
       } catch (e) { return json(res, { error: e.message }, 500); }
     }
@@ -3402,6 +3462,28 @@ const server = http.createServer(async (req, res) => {
     if (req.method === 'GET' && path === '/api/agents') return json(res, { agents: listMdDir(AGENTS_DIR) });
     if (req.method === 'GET' && path === '/api/skills') return json(res, { skills: listMdDir(SKILLS_DIR) });
 
+    // ── Local Ollama models (for model-picker dropdowns in the UI)
+    if (req.method === 'GET' && path === '/api/models') {
+      try {
+        const out = await new Promise((resolve) => {
+          let buf = '';
+          const p = spawn('ollama', ['list']);
+          const t = setTimeout(() => { try { p.kill(); } catch {} resolve(''); }, 4000);
+          p.stdout.on('data', (d) => { buf += d.toString(); });
+          p.on('error', () => { clearTimeout(t); resolve(''); });
+          p.on('close', () => { clearTimeout(t); resolve(buf); });
+        });
+        const models = out.split('\n').slice(1) // drop header
+          .map(line => line.trim()).filter(Boolean)
+          .map(line => {
+            const parts = line.split(/\s+/);
+            return { name: parts[0], size: parts.slice(1, 3).join(' '), modified: parts.slice(3).join(' ') };
+          })
+          .filter(m => m.name && !m.name.startsWith('NAME'));
+        return json(res, { models });
+      } catch (e) { return json(res, { error: e.message, models: [] }, 500); }
+    }
+
     res.writeHead(404, { 'Content-Type': 'text/plain' });
     res.end('Not found');
   } catch (e) {

package/sapper.mjs

@@ -1164,7 +1164,7 @@ const DEFAULT_CONFIG = Object.freeze({
     // every invocation requires a written summary + a specific question + supporting files.
     // Use it when the agent is stuck or about to make an important / hard-to-reverse change.
     enabled: true,
-    model: 'juilpark/Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-heretic:q4_k_m', // Default consultant model — change to any model you have via `ollama list`.
+    model: 'qwen3.5:9b-q8_0', // Default consultant model — change to any model you have via `ollama list`.
     contextLimit: null,            // Tokens to give the consultant (null = use model default).
     temperature: 0.2,              // Lower = more deliberate.
     thinking: 'on',                // 'on' | 'off' | 'auto' — reasoning models should stay on.
@@ -1189,7 +1189,7 @@ const DEFAULT_CONFIG = Object.freeze({
     // main agent can have a real back-and-forth: ask, get a thoughtful answer, ask follow-up,
     // refine, etc. Use when the main agent hits something unexpected or needs deep reasoning.
     enabled: true,
-    model: 'juilpark/Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled-heretic:q4_k_m', // Pick a strong reasoning model.
+    model: 'qwen3.5:9b-q8_0', // Pick a strong reasoning model.
     contextLimit: null,            // Tokens to give the thinker (null = use model default).
     temperature: 0.7,              // Higher than consultant — reasoning benefits from a little exploration.
     thinking: 'on',                // 'on' | 'off' | 'auto' — should usually stay on for reasoning models.