ablate-llm 0.4.0__tar.gz → 0.4.1__tar.gz

{ablate_llm-0.4.0/src/ablate_llm.egg-info → ablate_llm-0.4.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: ablate-llm
-Version: 0.4.0
+Version: 0.4.1
 Summary: Open-source, local chat with unrestricted small language models (ablate chat), plus a directional ablation (abliteration) toolkit to remove refusal from open LLMs.
 Author: AI Anytime
 License: Apache-2.0
@@ -117,8 +117,8 @@ ablate chat            # opens http://localhost:7861
 </p>
 
 A private chat app for the AI Anytime **unrestricted small language models**. Models download once from
-HuggingFace, then everything runs offline on your CPU or GPU. Chats, memories and your profile are JSON
-files in `~/.ablate/chat`. No telemetry, no account, no server of ours.
+HuggingFace, then everything runs offline on your CPU or GPU. Chats, memories, settings and your profile
+(with avatar) live in one local SQLite file, `~/.ablate/chat/ablate.db`. No telemetry, no account, no server of ours.
 
 | Model | Params | Refusals before → after | Download | RAM | Licence |
 |---|---|---|---|---|---|
@@ -128,12 +128,12 @@ files in `~/.ablate/chat`. No telemetry, no account, no server of ours.
 
 Refusal rates are on held-out harmful prompts with a keyword judge, from each model's card.
 
-- **Chat history**: saved locally, with search, pin, rename, delete and Markdown export.
+- **Chat history**: saved locally, with ChatGPT-style search (⌘K) across titles and messages, pin, rename, delete and Markdown export.
 - **Context window you set**, with a usage meter; the oldest turns are dropped first. Follow-ups see the whole conversation.
 - **Memory**: saved facts added to every chat.
 - **Agentic tools**: calculator, date and time, search past chats. Web fetch exists but is off by default.
 - **Copy** answers and code blocks, **edit** and resend, **regenerate**, **stop**, attach text and code files.
-- **Local sign-in** with an optional passcode and an 18+ terms gate. `ablate chat --share` lets devices on your Wi-Fi sign in with that passcode; model management stays on the host.
+- **Local profile** with a photo, an optional passcode (`ablate chat --reset-passcode` if you forget it) and an 18+ terms gate. `ablate chat --share` lets devices on your Wi-Fi sign in with that passcode; model management stays on the host.
 - **OpenAI-compatible API** with streaming at `http://localhost:7861/v1`:
 
 ```python

{ablate_llm-0.4.0 → ablate_llm-0.4.1}/README.md

@@ -74,8 +74,8 @@ ablate chat            # opens http://localhost:7861
 </p>
 
 A private chat app for the AI Anytime **unrestricted small language models**. Models download once from
-HuggingFace, then everything runs offline on your CPU or GPU. Chats, memories and your profile are JSON
-files in `~/.ablate/chat`. No telemetry, no account, no server of ours.
+HuggingFace, then everything runs offline on your CPU or GPU. Chats, memories, settings and your profile
+(with avatar) live in one local SQLite file, `~/.ablate/chat/ablate.db`. No telemetry, no account, no server of ours.
 
 | Model | Params | Refusals before → after | Download | RAM | Licence |
 |---|---|---|---|---|---|
@@ -85,12 +85,12 @@ files in `~/.ablate/chat`. No telemetry, no account, no server of ours.
 
 Refusal rates are on held-out harmful prompts with a keyword judge, from each model's card.
 
-- **Chat history**: saved locally, with search, pin, rename, delete and Markdown export.
+- **Chat history**: saved locally, with ChatGPT-style search (⌘K) across titles and messages, pin, rename, delete and Markdown export.
 - **Context window you set**, with a usage meter; the oldest turns are dropped first. Follow-ups see the whole conversation.
 - **Memory**: saved facts added to every chat.
 - **Agentic tools**: calculator, date and time, search past chats. Web fetch exists but is off by default.
 - **Copy** answers and code blocks, **edit** and resend, **regenerate**, **stop**, attach text and code files.
-- **Local sign-in** with an optional passcode and an 18+ terms gate. `ablate chat --share` lets devices on your Wi-Fi sign in with that passcode; model management stays on the host.
+- **Local profile** with a photo, an optional passcode (`ablate chat --reset-passcode` if you forget it) and an 18+ terms gate. `ablate chat --share` lets devices on your Wi-Fi sign in with that passcode; model management stays on the host.
 - **OpenAI-compatible API** with streaming at `http://localhost:7861/v1`:
 
 ```python

{ablate_llm-0.4.0 → ablate_llm-0.4.1}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "ablate-llm"
-version = "0.4.0"
+version = "0.4.1"
 description = "Open-source, local chat with unrestricted small language models (ablate chat), plus a directional ablation (abliteration) toolkit to remove refusal from open LLMs."
 readme = "README.md"
 requires-python = ">=3.9"

{ablate_llm-0.4.0 → ablate_llm-0.4.1}/src/ablate/__init__.py

@@ -23,7 +23,7 @@ from .pipeline import Ablator, run
 from .weights import bake_direction, bake_subspace
 from .hub import build_model_card, push_to_hub
 
-__version__ = "0.4.0"
+__version__ = "0.4.1"
 
 __all__ = [
     "Ablator",

{ablate_llm-0.4.0 → ablate_llm-0.4.1}/src/ablate/chat/server.py

@@ -39,7 +39,8 @@ MODELS = [
      "ram": "16 GB RAM"},
     {"id": "ai-anytime/Llama-3.2-1B-Instruct-abliterated", "name": "Ablate Llama 1B", "params": "1.2B",
      "size_gb": 2.5, "license": "Llama 3.2 Community", "context": 131072, "base": "meta-llama/Llama-3.2-1B-Instruct",
-     "blurb": "Fastest. Runs on almost any laptop. Refusals cut from 75% to 34%.", "ram": "8 GB RAM"},
+     "blurb": "Fastest. Runs on almost any laptop. Refusals cut from 75% to 34%. Too small for tools.",
+     "ram": "8 GB RAM", "tools": False},
 ]
 ALLOW = ["*.json", "*.safetensors", "*.jinja", "*.txt", "*.model", "*.md", "tokenizer*"]
 
@@ -244,7 +245,10 @@ SESSIONS: Dict[str, float] = {}
 # Chat turn (with tools)
 # --------------------------------------------------------------------------- #
 def _system_prompt(opts: dict, with_tools: bool) -> str:
-    parts = [opts.get("system") or ""]
+    import datetime as _dt
+    now = _dt.datetime.now().astimezone()
+    parts = [opts.get("system") or "",
+             f"Current local date and time: {now:%A, %d %B %Y, %H:%M} ({now.tzname()})."]
     if opts.get("use_memory", True):
         mem = [m["text"] for m in store.memories() if m.get("on")]
         if mem:
@@ -277,16 +281,20 @@ def chat_turn(body: dict, emit: Emit) -> None:
     store.save_chat(chat)
     emit({"type": "meta", "chat_id": chat["id"], "title": chat["title"], "index": len(msgs) - 1})
 
-    with_tools = bool(opts.get("tools"))
-    schemas = tools.schemas(bool(opts.get("network_tools"))) if with_tools else None
+    # Small models over-use tools, so only offer the tools this message plausibly needs.
+    tools_ok = next((m.get("tools", True) for m in MODELS if m["id"] == E.model_id), True)
+    schemas = (tools.pick(str(body.get("message") or ""), bool(opts.get("network_tools")))
+               if opts.get("tools") and tools_ok else [])
+    with_tools = bool(schemas)
     limit = min(int(opts.get("context_tokens", 4096)), E.max_context())
     reserve = int(opts.get("max_new_tokens", 768))
     history = [{"role": m["role"], "content": m["content"]} for m in msgs]
-    convo, ids, trimmed = E.fit(_system_prompt(opts, with_tools), history, limit, reserve, schemas)
+    convo, ids, trimmed = E.fit(_system_prompt(opts, with_tools), history, limit, reserve, schemas or None)
     emit({"type": "context", "used": int(ids.shape[1]), "limit": limit, "trimmed": trimmed})
 
-    steps, stats, answer = [], {}, ""
-    for _ in range(4):  # at most 3 tool calls, then a final answer
+    steps, stats, answer, seen = [], {}, "", set()
+    for rnd in range(4):  # up to 3 tool calls; the last round never offers tools, so it must answer
+        offer = with_tools and rnd < 3
         buf, mode = "", None
 
         def on_text(piece: str) -> None:
@@ -294,25 +302,37 @@ def chat_turn(body: dict, emit: Emit) -> None:
             buf += piece
             if mode is None:
                 head = buf.lstrip()
-                if with_tools and tools.looks_like_call(head) and len(head) < 12:
+                if offer and tools.looks_like_call(head) and len(head) < 12:
                     return
-                mode = "tool" if with_tools and tools.looks_like_call(head) else "text"
+                mode = "tool" if offer and tools.looks_like_call(head) else "text"
                 if mode == "text":
                     emit({"type": "token", "t": buf})
             elif mode == "text":
                 emit({"type": "token", "t": piece})
 
+        if rnd and not offer:  # final round: same conversation, tools removed
+            ids = E.encode(convo)
         stats = E.stream(ids, opts, on_text)
-        call = tools.parse_call(stats["text"]) if with_tools and mode != "text" else None
+        call = tools.parse_call(stats["text"]) if offer and mode != "text" else None
+        key = call and json.dumps([call["name"], call["arguments"]], sort_keys=True)
+        if call and key in seen:  # repeating itself: stop calling tools and answer
+            with_tools = False
+            continue
         if not call:
             answer = stats["text"].strip()
             if mode != "text" and answer:  # held back as a possible tool call, but it was plain text
                 emit({"type": "token", "t": stats["text"]})
-            break
+            if answer or stats.get("stopped") or not steps:
+                break
+            with_tools = False  # empty reply after tool use: ask once more without tools
+            continue
+        seen.add(key)
         emit({"type": "tool_call", "name": call["name"], "args": call["arguments"]})
         result = tools.run(call["name"], call["arguments"], bool(opts.get("network_tools")))
         steps.append({"name": call["name"], "args": call["arguments"], "result": result})
         emit({"type": "tool_result", "name": call["name"], "result": result})
+        if isinstance(result, dict) and result.get("error"):
+            with_tools = False  # a failed call: let the model answer with what it has instead of retrying
         convo = convo + [
             {"role": "assistant", "content": "", "tool_calls": [
                 {"type": "function", "function": {"name": call["name"], "arguments": call["arguments"]}}]},
@@ -487,7 +507,7 @@ class Handler(BaseHTTPRequestHandler):
                 return self._static(url.path[len("/static/"):])
             if url.path == "/api/me":
                 p = store.public_profile()
-                return self._json({"profile": p and {k: p[k] for k in ("name", "has_passcode", "accepted_terms")},
+                return self._json({"profile": p and {k: p[k] for k in ("name", "has_passcode", "accepted_terms", "avatar")},
                                    "authed": self._authed(), "local": self._local()})
             if url.path == "/v1/models":
                 if not self._api_authed():
@@ -503,11 +523,15 @@ class Handler(BaseHTTPRequestHandler):
                     "models": models_state(), "loaded": E.model_id,
                     "device": E.device or E.pick()[0], "max_context": E.max_context() if E.model else None,
                     "chats": store.list_chats(), "busy": E.lock.locked(), "local": self._local(),
+                    "tools_ok": next((m.get("tools", True) for m in MODELS if m["id"] == E.model_id), True),
                     "port": self.server.server_address[1]})
             if url.path == "/api/models":
                 return self._json({"models": models_state(), "loaded": E.model_id})
+            if url.path == "/api/search":
+                q = (urllib.parse.parse_qs(url.query).get("q") or [""])[0]
+                return self._json({"results": store.search_chats(q) if q.strip() else []})
             if url.path.startswith("/api/chats/"):
-                c = store.get_chat(url.path.rsplit("/", 1)[-1])
+                c = store.get_chat(url.path.rsplit("/", 1)[-1], mark_opened=True)
                 return self._json(c or {"error": "not found"}, 200 if c else 404)
             return self._json({"error": "not found"}, 404)
         except Exception as e:
@@ -568,25 +592,25 @@ class Handler(BaseHTTPRequestHandler):
                     else:
                         E.unload()
                 return self._json({"models": models_state(), "loaded": E.model_id,
-                                   "device": E.device, "max_context": E.max_context() if E.model else None})
+                                   "device": E.device, "max_context": E.max_context() if E.model else None,
+                                   "tools_ok": next((m.get("tools", True) for m in MODELS if m["id"] == E.model_id), True)})
             if path == "/api/settings":
                 return self._json({"settings": store.save_settings(body)})
             if path == "/api/memories":
                 return self._json({"memories": store.save_memories(body.get("memories") or [])})
             if path == "/api/chats/update":
-                c = store.get_chat(str(body.get("id")))
-                if not c:
-                    return self._json({"error": "not found"}, 404)
-                for k in ("title", "pinned"):
-                    if k in body:
-                        c[k] = body[k]
-                store.save_chat(c)
+                store.update_chat(str(body.get("id")), **{k: body[k] for k in ("title", "pinned") if k in body})
                 return self._json({"chats": store.list_chats()})
             if path == "/api/chats/delete":
                 store.delete_chat(str(body.get("id")))
                 return self._json({"chats": store.list_chats()})
             if path == "/api/profile":
-                store.update_profile(body.get("name"), body.get("passcode"))
+                avatar = body.get("avatar")
+                if avatar and not (isinstance(avatar, str) and len(avatar) < 400_000 and
+                                   avatar.startswith(("data:image/png;base64,", "data:image/jpeg;base64,",
+                                                      "data:image/webp;base64,"))):
+                    return self._json({"error": "Avatar must be a PNG, JPEG or WebP image under 300 KB."}, 400)
+                store.update_profile(body.get("name"), body.get("passcode"), avatar)
                 return self._json({"profile": store.public_profile()})
             return self._json({"error": "not found"}, 404)
         except (BrokenPipeError, ConnectionResetError):

{ablate_llm-0.4.0 → ablate_llm-0.4.1}/src/ablate/chat/static/index.html

@@ -24,7 +24,7 @@ code,pre,.mono{font-family:ui-monospace,SFMono-Regular,Menlo,Consolas,monospace}
 /* ---------- sign-in (split) ---------- */
 #auth{position:fixed;inset:0;display:grid;grid-template-columns:minmax(0,41%) minmax(0,1fr);z-index:30;background:var(--bg)}
 .auth-l{background:#E9E7E1;color:#14121A;display:flex;flex-direction:column;justify-content:space-between;padding:64px 72px}
-.auth-l img{height:64px;width:auto;align-self:flex-start;mix-blend-mode:multiply}
+.auth-l img{height:64px;width:auto;align-self:flex-start}
 .auth-l h1{font-size:clamp(2.2rem,4.2vw,3.7rem);line-height:1.02;letter-spacing:-.035em;font-weight:500;margin:0}
 .auth-l p{margin:18px 0 0;color:#4A4758;max-width:36ch}
 .auth-r{background:#07080C;color:#ECEBF5;display:grid;place-items:center;padding:32px}
@@ -65,10 +65,9 @@ code,pre,.mono{font-family:ui-monospace,SFMono-Regular,Menlo,Consolas,monospace}
 #app.collapsed{grid-template-columns:0 minmax(0,1fr)}
 #app.collapsed aside{visibility:hidden}
 aside{background:var(--side);border-right:1px solid var(--line);display:flex;flex-direction:column;min-height:0;overflow:hidden}
-.side-top{display:flex;align-items:center;justify-content:space-between;padding:16px 14px 10px 18px}
-.side-top img{height:34px;width:auto}
-.logo-d{mix-blend-mode:lighten}
-.logo-l{mix-blend-mode:multiply}
+.side-top{display:flex;align-items:center;justify-content:space-between;padding:14px 10px 10px 16px}
+.side-top img{height:36px;width:auto}
+.side-top .acts{display:flex;gap:2px}
 .logo-l{display:none}
 [data-theme="light"] .logo-l{display:block}
 [data-theme="light"] .logo-d{display:none}
@@ -97,7 +96,11 @@ aside{background:var(--side);border-right:1px solid var(--line);display:flex;fle
 .local-card{display:flex;align-items:center;gap:10px;padding:10px 12px;border-radius:12px;background:var(--panel);border:1px solid var(--line);font-size:.8rem;color:var(--muted)}
 .local-card b{color:var(--ok);font-weight:600}
 .me{display:flex;align-items:center;gap:12px;padding:4px 4px}
-.avatar{width:42px;height:42px;border-radius:10px;display:grid;place-items:center;font-weight:700;color:#fff;background:linear-gradient(135deg,#5AA9F5,#8B6CFF);flex:none}
+.avatar{width:42px;height:42px;border-radius:12px;display:grid;place-items:center;font-weight:700;color:#fff;background:linear-gradient(135deg,#5AA9F5,#8B6CFF);flex:none;overflow:hidden}
+.avatar img{width:100%;height:100%;object-fit:cover}
+.avatar.lg{width:72px;height:72px;border-radius:18px;font-size:1.6rem}
+#menu{display:none}
+#app.collapsed #menu{display:grid}
 .me .n{flex:1;min-width:0;font-weight:600;white-space:nowrap;overflow:hidden;text-overflow:ellipsis}
 .me small{display:block;font-weight:400;color:var(--muted);font-size:.78rem}
 
@@ -241,6 +244,25 @@ input[type=range]{width:100%;accent-color:var(--accent)}
 .codeline code{flex:1;min-width:0;overflow-x:auto;white-space:nowrap}
 .toast{position:fixed;left:50%;bottom:24px;transform:translateX(-50%);z-index:50;background:var(--text);color:var(--bg);padding:8px 14px;border-radius:10px;font-size:.86rem;box-shadow:0 10px 30px -10px rgba(0,0,0,.5)}
 
+/* search dialog (ChatGPT style) */
+.search-sheet{width:min(680px,100%);max-height:72vh;background:var(--panel);border:1px solid var(--line2);border-radius:20px;display:flex;flex-direction:column;overflow:hidden;box-shadow:0 30px 80px -20px rgba(0,0,0,.6)}
+.search-top{display:flex;align-items:center;gap:10px;padding:16px 16px 16px 22px;border-bottom:1px solid var(--line)}
+.search-top input{flex:1;border:0;background:none;outline:none;font-size:1.05rem;color:var(--text)}
+.search-top input::placeholder{color:var(--faint)}
+.search-list{overflow-y:auto;padding:8px 10px 14px}
+.search-list h5{margin:14px 12px 6px;font-size:.8rem;font-weight:500;color:var(--faint)}
+.sr{display:flex;align-items:flex-start;gap:14px;width:100%;border:0;background:none;text-align:left;padding:10px 12px;border-radius:12px;color:var(--text)}
+.sr:hover,.sr.on{background:var(--panel2)}
+.sr .ico{margin-top:2px;color:var(--muted)}
+.sr b{display:block;font-weight:500;white-space:nowrap;overflow:hidden;text-overflow:ellipsis}
+.sr small{display:block;color:var(--muted);font-size:.84rem;margin-top:2px;overflow:hidden;display:-webkit-box;-webkit-line-clamp:2;-webkit-box-orient:vertical}
+.sr mark{background:color-mix(in srgb,var(--accent) 30%,transparent);color:inherit;border-radius:3px;padding:0 1px}
+.sr span{min-width:0;flex:1}
+.search-empty{padding:28px 12px;color:var(--muted);text-align:center;font-size:.9rem}
+.av-row{display:flex;align-items:center;gap:14px}
+.av-row .btns{display:flex;gap:8px;flex-wrap:wrap}
+.auth-av{width:64px;height:64px;border-radius:16px;overflow:hidden;background:linear-gradient(135deg,#5AA9F5,#8B6CFF);display:grid;place-items:center;font-size:1.5rem;font-weight:700;color:#fff;margin-bottom:6px}
+.auth-av img{width:100%;height:100%;object-fit:cover}
 @media (max-width:900px){
   #auth{grid-template-columns:1fr}
   .auth-l{display:none}
@@ -248,6 +270,7 @@ input[type=range]{width:100%;accent-color:var(--accent)}
   aside{position:fixed;inset:0 auto 0 0;width:290px;z-index:20;transform:translateX(-100%);transition:transform .2s}
   #app.open aside{transform:none}
   #app.collapsed aside{visibility:visible}
+  #menu{display:grid}
   .mgrid{grid-template-columns:1fr}
   .chip.hide-sm{display:none}
 }
@@ -283,6 +306,7 @@ input[type=range]{width:100%;accent-color:var(--accent)}
   <symbol id="i-code" viewBox="0 0 24 24"><path d="m8 8-5 4 5 4M16 8l5 4-5 4"/></symbol>
   <symbol id="i-calc" viewBox="0 0 24 24"><rect x="5" y="3" width="14" height="18" rx="2"/><path d="M8 7h8M8 12h2M14 12h2M8 16h2M14 16h2"/></symbol>
   <symbol id="i-menu" viewBox="0 0 24 24"><path d="M4 7h16M4 12h16M4 17h16"/></symbol>
+  <symbol id="i-bubble" viewBox="0 0 24 24"><path d="M21 12a8 8 0 0 1-11.6 7.1L4 20.5l1.4-4.9A8 8 0 1 1 21 12Z"/></symbol>
 </svg>
 
 <!-- ============ sign-in ============ -->
@@ -291,7 +315,7 @@ input[type=range]{width:100%;accent-color:var(--accent)}
     <img src="/static/logo-light.png" alt="Ablate">
     <div>
       <h1>Unrestricted small models. On your machine.</h1>
-      <p>Open source by AI Anytime. No account, no cloud, no logs: your profile and chats are files on this computer.</p>
+      <p>Open source by AI Anytime. No account, no cloud, no logs: your profile and chats live in a database file on this computer.</p>
     </div>
   </div>
   <div class="auth-r">
@@ -311,12 +335,13 @@ input[type=range]{width:100%;accent-color:var(--accent)}
       <p class="auth-foot">Open source under Apache-2.0 · <a href="https://aianytime.github.io/ablate/" target="_blank" rel="noopener">About Ablate</a></p>
     </form>
     <form class="auth-card" id="auth-login" hidden>
+      <div class="auth-av" id="l-av" hidden></div>
       <h2>Welcome back.</h2>
       <p class="sub" id="l-sub">Unlock your local Ablate Chat.</p>
       <div class="field"><label for="l-pass">Passcode *</label><input id="l-pass" type="password" autocomplete="current-password" required></div>
       <div class="auth-err" id="l-err"></div>
       <button class="btn-light" type="submit">Continue</button>
-      <p class="auth-foot">Forgot it? Delete <code class="mono">~/.ablate/chat/profile.json</code> on this computer.</p>
+      <p class="auth-foot">Forgot it? Run <code class="mono">ablate chat --reset-passcode</code> on this computer.</p>
     </form>
   </div>
 </section>
@@ -339,11 +364,11 @@ input[type=range]{width:100%;accent-color:var(--accent)}
   <aside>
     <div class="side-top">
       <img class="logo-d" src="/static/logo-dark.png" alt="Ablate"><img class="logo-l" src="/static/logo-light.png" alt="Ablate">
-      <button class="iconbtn" id="collapse" title="Hide sidebar" aria-label="Hide sidebar"><svg class="ico"><use href="#i-panel"/></svg></button>
+      <div class="acts"><button class="iconbtn" id="open-search" title="Search chats (⌘K)" aria-label="Search chats"><svg class="ico"><use href="#i-search"/></svg></button>
+      <button class="iconbtn" id="collapse" title="Close sidebar" aria-label="Close sidebar"><svg class="ico"><use href="#i-panel"/></svg></button></div>
     </div>
     <button class="newchat" id="new"><svg class="ico"><use href="#i-plus"/></svg>New chat</button>
     <nav class="nav" aria-label="App">
-      <label><svg class="ico"><use href="#i-search"/></svg><input id="q" placeholder="Search" aria-label="Search chats"></label>
       <button id="open-models"><svg class="ico"><use href="#i-cube"/></svg>Models</button>
       <button id="open-memory"><svg class="ico"><use href="#i-brain"/></svg>Memory</button>
       <button id="open-settings"><svg class="ico"><use href="#i-gear"/></svg>Settings</button>
@@ -359,11 +384,10 @@ input[type=range]{width:100%;accent-color:var(--accent)}
 
   <main>
     <div class="topbar">
-      <button class="iconbtn" id="menu" title="Chats" aria-label="Show chats"><svg class="ico"><use href="#i-menu"/></svg></button>
+      <button class="iconbtn" id="menu" title="Open sidebar" aria-label="Open sidebar"><svg class="ico"><use href="#i-panel"/></svg></button>
       <button class="modelbtn" id="modelbtn"><span class="dot" id="mdot"></span><span id="mname">Choose a model</span><svg class="ico" style="width:14px;height:14px"><use href="#i-down"/></svg></button>
       <span class="sp"></span>
       <span class="chip hide-sm" id="ctx" title="Context window used by the last reply"><span id="ctx-t">0 / 4,096</span><span class="meter"><i id="ctx-bar"></i></span></span>
-      <span class="chip hide-sm"><svg class="ico" style="color:var(--ok)"><use href="#i-lock"/></svg>Private · on this device</span>
       <button class="iconbtn" id="export" title="Export chat as Markdown" aria-label="Export chat"><svg class="ico"><use href="#i-dl"/></svg></button>
       <button class="iconbtn" id="theme" title="Theme" aria-label="Switch theme"><svg class="ico"><use href="#i-moon"/></svg></button>
     </div>
@@ -409,17 +433,33 @@ input[type=range]{width:100%;accent-color:var(--accent)}
 <!-- memory -->
 <div class="modal" id="m-memory" hidden role="dialog" aria-modal="true" aria-labelledby="mem-t">
   <div class="sheet" style="width:min(620px,100%)">
-    <header><div><h2 id="mem-t">Memory</h2><p>Facts added to every chat so the model remembers them. Stored in <code class="mono">~/.ablate/chat/memory.json</code>.</p></div><span class="sp"></span><button class="iconbtn" data-close aria-label="Close"><svg class="ico"><use href="#i-x"/></svg></button></header>
+    <header><div><h2 id="mem-t">Memory</h2><p>Facts added to every chat so the model remembers them. Stored on this computer in <code class="mono">~/.ablate/chat</code>.</p></div><span class="sp"></span><button class="iconbtn" data-close aria-label="Close"><svg class="ico"><use href="#i-x"/></svg></button></header>
     <div class="toggle-row"><span>Use memory in chats</span><label class="switch"><input type="checkbox" id="use-mem"><i></i></label></div>
     <div class="memlist" id="memlist"></div>
     <div class="custom"><input id="mem-new" placeholder="e.g. I write Python and prefer short answers."><button class="btn primary" id="mem-add">Add</button></div>
   </div>
 </div>
 
+<!-- search -->
+<div class="modal" id="m-search" hidden role="dialog" aria-modal="true" aria-label="Search chats">
+  <div class="search-sheet">
+    <div class="search-top"><input id="sq" placeholder="Search chats…" autocomplete="off" aria-label="Search chats"><button class="iconbtn" data-close aria-label="Close"><svg class="ico"><use href="#i-x"/></svg></button></div>
+    <div class="search-list" id="slist"></div>
+  </div>
+</div>
+
 <!-- settings drawer -->
 <div class="drawer" id="settings" hidden role="dialog" aria-labelledby="set-t">
   <header><h2 id="set-t">Settings</h2><button class="iconbtn" data-close aria-label="Close"><svg class="ico"><use href="#i-x"/></svg></button></header>
   <div class="body">
+    <div class="group"><h4>Profile</h4>
+      <div class="av-row"><div class="avatar lg" id="p-av">A</div><div class="btns">
+        <button class="btn" type="button" id="p-av-up">Upload photo</button><button class="btn" type="button" id="p-av-rm">Remove</button>
+        <input type="file" id="p-av-file" accept="image/png,image/jpeg,image/webp" hidden></div></div>
+      <div class="set"><div class="lbl">Display name</div><input type="text" id="p-name" maxlength="40"></div>
+      <div class="set"><div class="lbl">New passcode</div><input type="password" id="p-pass" autocomplete="new-password" placeholder="Leave empty to keep it"><small>A passcode locks the app and is required with <code class="mono">--share</code>.</small></div>
+      <div><button class="btn" id="p-save">Save profile</button></div>
+    </div>
     <div class="group"><h4>Model behaviour</h4>
       <div class="set"><div class="lbl">System prompt</div><textarea id="s-system"></textarea></div>
       <div class="set"><div class="lbl">Context window <b id="v-ctx"></b></div><input type="range" id="s-ctx" min="512" max="32768" step="512"><small>How much of the conversation the model sees. Older messages are dropped first. Bigger uses more memory.</small></div>
@@ -432,11 +472,6 @@ input[type=range]{width:100%;accent-color:var(--accent)}
       <div class="toggle-row"><span>Local tools<small>Calculator, date and time, search your past chats.</small></span><label class="switch"><input type="checkbox" id="s-tools"><i></i></label></div>
       <div class="toggle-row"><span>Network tools<small>Lets the model fetch public web pages. Off keeps everything offline.</small></span><label class="switch"><input type="checkbox" id="s-net"><i></i></label></div>
     </div>
-    <div class="group"><h4>Profile</h4>
-      <div class="set"><div class="lbl">Display name</div><input type="text" id="p-name" maxlength="40"></div>
-      <div class="set"><div class="lbl">New passcode</div><input type="password" id="p-pass" autocomplete="new-password" placeholder="Leave empty to keep it"><small>A passcode locks the app and is required with <code class="mono">--share</code>.</small></div>
-      <div><button class="btn" id="p-save">Save profile</button></div>
-    </div>
     <div class="group"><h4>OpenAI-compatible API</h4>
       <small class="small">Point any OpenAI SDK, editor or agent at the loaded model.</small>
       <div class="codeline"><code id="api-base"></code><button class="iconbtn" data-copy="#api-base" aria-label="Copy base URL"><svg class="ico"><use href="#i-copy"/></svg></button></div>
@@ -514,7 +549,7 @@ async function boot() {
   const me = await api('/api/me');
   if (!me.profile) return showAuth('create');
   if (!me.authed && !me.profile.accepted_terms && !me.profile.has_passcode) return showGate();
-  if (!me.authed) return showAuth('login', me.profile.name);
+  if (!me.authed) { setAvatar($('#l-av'), me.profile); $('#l-av').hidden = false; return showAuth('login', me.profile.name); }
   start();
 }
 function showAuth(which, name) {
@@ -543,7 +578,8 @@ async function start() {
   $('#auth').hidden = true; $('#gate').hidden = true; $('#app').hidden = false;
   ST = await api('/api/state');
   toolsOn = !!ST.settings.tools;
-  $('#me-name').textContent = ST.profile.name; $('#avatar').textContent = (ST.profile.name || 'A').trim()[0].toUpperCase();
+  $('#me-name').textContent = ST.profile.name; setAvatar($('#avatar'), ST.profile); setAvatar($('#p-av'), ST.profile);
+  if (ST.settings.theme) applyTheme(ST.settings.theme);
   $('#me-dev').textContent = `100% local · ${({mps: 'Apple GPU', cuda: 'NVIDIA GPU', cpu: 'CPU'})[ST.device] || ST.device}`;
   $('#mm-dev').textContent = ({mps: 'Mac (Apple GPU)', cuda: 'NVIDIA GPU', cpu: 'CPU'})[ST.device] || 'computer';
   renderModelBtn(); renderChats(); renderTools(); fillSettings(); newChat();
@@ -557,15 +593,18 @@ function renderModelBtn(loading) {
   $('#hero-note').textContent = m ? `${m.name} · ${m.params} parameters · running on this ${ST.device === 'cpu' ? 'CPU' : 'GPU'}, offline.` : 'Pick a model to start. It downloads once, then runs offline.';
   $('#send').disabled = !m && !busy;
 }
+function setAvatar(el, p) {
+  el.innerHTML = '';
+  if (p && p.avatar) { const i = document.createElement('img'); i.src = p.avatar; i.alt = ''; el.appendChild(i); }
+  else el.textContent = ((p && p.name) || 'A').trim()[0].toUpperCase();
+}
 function renderChats() {
-  const q = $('#q').value.trim().toLowerCase();
-  const rows = ST.chats.filter(c => !q || c.title.toLowerCase().includes(q));
+  const rows = ST.chats;
   $('#chats').innerHTML = rows.length ? rows.map(c => `<div class="chat-row ${CHAT && CHAT.id === c.id ? 'on' : ''}" data-id="${esc(c.id)}">
       <button class="t" data-open title="${esc(c.title)}">${c.pinned ? '📌 ' : ''}${esc(c.title)}</button>
       <span class="acts"><button data-pin title="${c.pinned ? 'Unpin' : 'Pin'}">${ico('pin')}</button><button data-rename title="Rename">${ico('edit')}</button><button data-del title="Delete">${ico('trash')}</button></span></div>`).join('')
-    : `<div class="empty-chats">${q ? 'No matching chats.' : 'Your chats will show up here.'}</div>`;
+    : '<div class="empty-chats">Your chats will show up here.</div>';
 }
-$('#q').addEventListener('input', renderChats);
 $('#chats').addEventListener('click', async e => {
   const row = e.target.closest('.chat-row'); if (!row) return; const id = row.dataset.id; const c = ST.chats.find(x => x.id === id);
   if (e.target.closest('[data-open]')) return openChat(id);
@@ -641,7 +680,11 @@ $('#input').addEventListener('keydown', e => {
   if (e.key === 'Escape') { if (busy) stop(); else if (editAt !== null) { editAt = null; $('#hint').textContent = ''; $('#input').value = ''; autosize(); } }
 });
 $('#composer').addEventListener('submit', e => { e.preventDefault(); busy ? stop() : send(); });
-function renderTools() { $('#tools').classList.toggle('on', toolsOn); $('#tools').title = toolsOn ? 'Tools on: calculator, time, search chats' : 'Tools off'; }
+function renderTools() {
+  const ok = ST ? ST.tools_ok !== false : true;
+  $('#tools').classList.toggle('on', toolsOn && ok); $('#tools').disabled = !ok; $('#tools').style.opacity = ok ? '' : '.45';
+  $('#tools').title = !ok ? 'This model is too small for reliable tool use. Try Ablate Qwen 1.5B.' : toolsOn ? 'Tools on: used only when a message needs them (maths, date and time, past chats)' : 'Tools off';
+}
 $('#tools').addEventListener('click', async () => { toolsOn = !toolsOn; renderTools(); ST.settings = (await api('/api/settings', {tools: toolsOn})).settings; $('#s-tools').checked = toolsOn; });
 $('#attach').addEventListener('click', () => $('#file').click());
 $('#file').addEventListener('change', async e => {
@@ -730,7 +773,7 @@ $('#mgrid').addEventListener('click', async e => {
     if (act === 'load' || act === 'unload') {
       b.disabled = true; b.textContent = act === 'load' ? 'Loading…' : 'Unloading…'; renderModelBtn(act === 'load' ? ST.models.find(x => x.id === id).name : '');
       const j = await api('/api/models/load', {id: act === 'load' ? id : null}); ST.models = j.models; ST.loaded = j.loaded; ST.max_context = j.max_context;
-      if (j.device) ST.device = j.device; fillSettings(); if (act === 'load') { closeAll(); $('#input').focus(); }
+      if (j.device) ST.device = j.device; ST.tools_ok = j.tools_ok; renderTools(); fillSettings(); if (act === 'load') { closeAll(); $('#input').focus(); }
     }
   } catch (err) { toast(err.message); }
   renderModels(); renderModelBtn();
@@ -767,16 +810,59 @@ SL.forEach(([el, k, lab, f]) => $(el).addEventListener('input', e => { $(lab).te
 $('#s-system').addEventListener('input', e => saveSettings({system: e.target.value}));
 $('#s-tools').addEventListener('change', e => { toolsOn = e.target.checked; renderTools(); saveSettings({tools: toolsOn}); });
 $('#s-net').addEventListener('change', e => saveSettings({network_tools: e.target.checked}));
+$('#p-av-up').addEventListener('click', () => $('#p-av-file').click());
+$('#p-av-file').addEventListener('change', async e => {
+  const f = e.target.files[0]; e.target.value = ''; if (!f) return;
+  try {
+    const img = await createImageBitmap(f); const c = document.createElement('canvas'); c.width = c.height = 256;
+    const side = Math.min(img.width, img.height);
+    c.getContext('2d').drawImage(img, (img.width - side) / 2, (img.height - side) / 2, side, side, 0, 0, 256, 256);
+    ST.profile = (await api('/api/profile', {avatar: c.toDataURL('image/webp', 0.86)})).profile;
+    setAvatar($('#avatar'), ST.profile); setAvatar($('#p-av'), ST.profile); toast('Photo updated');
+  } catch (err) { toast(err.message || 'Could not read that image'); }
+});
+$('#p-av-rm').addEventListener('click', async () => { ST.profile = (await api('/api/profile', {avatar: ''})).profile; setAvatar($('#avatar'), ST.profile); setAvatar($('#p-av'), ST.profile); });
 $('#p-save').addEventListener('click', async () => { const body = {name: $('#p-name').value}; if ($('#p-pass').value) body.passcode = $('#p-pass').value; ST.profile = (await api('/api/profile', body)).profile; $('#p-pass').value = ''; $('#me-name').textContent = ST.profile.name; toast('Profile saved'); });
 $('#open-settings').addEventListener('click', () => { closeAll(); $('#settings').hidden = false; });
 
 /* ---------------- misc ---------------- */
-function closeAll() { ['#m-models', '#m-memory', '#settings'].forEach(s => $(s).hidden = true); }
+function closeAll() { ['#m-models', '#m-memory', '#settings', '#m-search'].forEach(s => $(s).hidden = true); }
+
+/* ---------------- search (ChatGPT style) ---------------- */
+let sTimer = 0, sItems = [], sOn = 0;
+function hl(text, q) { const t = esc(text); if (!q) return t; const i = t.toLowerCase().indexOf(esc(q).toLowerCase()); return i < 0 ? t : t.slice(0, i) + '<mark>' + t.slice(i, i + esc(q).length) + '</mark>' + t.slice(i + esc(q).length); }
+function srRow(r, q, k) { return `<button class="sr ${k === sOn ? 'on' : ''}" data-k="${k}">${ico(r.icon || 'bubble')}<span><b>${hl(r.title, q)}</b>${r.snippet ? `<small>${hl(r.snippet, q)}</small>` : ''}</span></button>`; }
+function renderSearch(q, results) {
+  sItems = []; let html = ''; sOn = 0;
+  if (!q) {
+    sItems.push({id: null, title: 'New chat', icon: 'edit'}); html += srRow(sItems[0], '', 0);
+    const opened = ST.chats.filter(c => c.opened).sort((a, b) => b.opened - a.opened)[0];
+    if (opened) { html += '<h5>Last opened</h5>'; sItems.push(opened); html += srRow(opened, '', sItems.length - 1); }
+    const recent = [...ST.chats].sort((a, b) => b.updated - a.updated).filter(c => !opened || c.id !== opened.id).slice(0, 8);
+    if (recent.length) { html += '<h5>Recent chats</h5>'; recent.forEach(c => { sItems.push(c); html += srRow(c, '', sItems.length - 1); }); }
+  } else if (!results.length) { html = `<div class="search-empty">No chats match “${esc(q)}”.</div>`; }
+  else { html += '<h5>Results</h5>'; results.forEach(r => { sItems.push(r); html += srRow(r, q, sItems.length - 1); }); }
+  $('#slist').innerHTML = html;
+}
+function openSearch() { closeAll(); $('#m-search').hidden = false; $('#sq').value = ''; renderSearch('', []); setTimeout(() => $('#sq').focus(), 20); closeSide(); }
+function pickSearch(k) { const it = sItems[k]; if (!it) return; closeAll(); if (it.id) openChat(it.id); else newChat(); }
+$('#open-search').addEventListener('click', openSearch);
+$('#sq').addEventListener('input', e => { const q = e.target.value.trim(); clearTimeout(sTimer); if (!q) return renderSearch('', []); sTimer = setTimeout(async () => { try { renderSearch(q, (await api('/api/search?q=' + encodeURIComponent(q))).results); } catch {} }, 160); });
+$('#sq').addEventListener('keydown', e => {
+  if (e.key === 'ArrowDown' || e.key === 'ArrowUp') { e.preventDefault(); sOn = Math.max(0, Math.min(sItems.length - 1, sOn + (e.key === 'ArrowDown' ? 1 : -1))); $$('#slist .sr').forEach(b => b.classList.toggle('on', +b.dataset.k === sOn)); $$('#slist .sr')[sOn]?.scrollIntoView({block: 'nearest'}); }
+  if (e.key === 'Enter') { e.preventDefault(); pickSearch(sOn); }
+});
+$('#slist').addEventListener('click', e => { const b = e.target.closest('.sr'); if (b) pickSearch(+b.dataset.k); });
+function applyTheme(t) { document.documentElement.setAttribute('data-theme', t); try { localStorage.setItem('ablate-chat-theme', t); } catch {} }
 function closeSide() { $('#app').classList.remove('open'); }
-document.addEventListener('keydown', e => { if (e.key === 'Escape') closeAll(); if ((e.metaKey || e.ctrlKey) && e.key.toLowerCase() === 'k') { e.preventDefault(); newChat(); } });
-$('#collapse').addEventListener('click', () => $('#app').classList.add('collapsed'));
-$('#menu').addEventListener('click', () => { const a = $('#app'); if (a.classList.contains('collapsed')) a.classList.remove('collapsed'); else a.classList.toggle('open'); });
-$('#theme').addEventListener('click', () => { const t = document.documentElement.getAttribute('data-theme') === 'light' ? 'dark' : 'light'; document.documentElement.setAttribute('data-theme', t); try { localStorage.setItem('ablate-chat-theme', t); } catch {} });
+document.addEventListener('keydown', e => {
+  if (e.key === 'Escape') closeAll();
+  if ((e.metaKey || e.ctrlKey) && e.key.toLowerCase() === 'k' && !$('#app').hidden) { e.preventDefault(); openSearch(); }
+  if ((e.metaKey || e.ctrlKey) && e.shiftKey && e.key.toLowerCase() === 'o') { e.preventDefault(); newChat(); }
+});
+$('#collapse').addEventListener('click', () => { const a = $('#app'); if (matchMedia('(max-width:900px)').matches) a.classList.remove('open'); else a.classList.add('collapsed'); });
+$('#menu').addEventListener('click', () => { const a = $('#app'); if (matchMedia('(max-width:900px)').matches) a.classList.toggle('open'); else a.classList.remove('collapsed'); });
+$('#theme').addEventListener('click', async () => { const t = document.documentElement.getAttribute('data-theme') === 'light' ? 'dark' : 'light'; applyTheme(t); try { ST.settings = (await api('/api/settings', {theme: t})).settings; } catch {} });
 $('#export').addEventListener('click', () => {
   if (!CHAT || !CHAT.messages.length) return toast('Nothing to export yet');
   const text = `# ${CHAT.title}\n\n` + CHAT.messages.map(m => `**${m.role === 'user' ? 'You' : 'Ablate'}:**\n\n${m.content}\n`).join('\n');

{ablate_llm-0.4.0 → ablate_llm-0.4.1}/src/ablate/chat/static/terms.html

@@ -46,7 +46,7 @@ nav{display:flex;gap:16px;flex-wrap:wrap;margin:18px 0 0}
   <h2 id="privacy">Privacy Policy</h2>
   <p><strong>We collect nothing.</strong> Ablate Chat has no accounts on our servers, no analytics and no telemetry.</p>
   <ul>
-    <li><strong>Your profile, chats, memories and settings</strong> are JSON files in <code>~/.ablate/chat</code> on your computer. Delete that folder to erase them.</li>
+    <li><strong>Your profile, chats, memories and settings</strong> are stored in a local SQLite database, <code>~/.ablate/chat/ablate.db</code>, on your computer. Delete that folder to erase them.</li>
     <li><strong>Prompts and answers</strong> are processed by the model on your own CPU or GPU. They are never sent to AI Anytime or anyone else.</li>
     <li><strong>Network use:</strong> the app goes online only to download a model from the HuggingFace Hub when you click Download, and to fetch a web page if you turn on network tools and the model calls them. HuggingFace's own privacy policy applies to downloads.</li>
     <li><strong>Sharing on your network:</strong> if you start the app with <code>--share</code>, other devices on your network can sign in with your passcode. Their chats are stored on your computer.</li>