add internet search: ?query escape + web_search tool (SearXNG)

README.md

@@ -1,18 +1,19 @@
 # METATRON — lightweight Ollama terminal harness
 
 A minimal terminal harness for talking to Ollama (hosted on nightmare `10.30.20.29:11434`)
-with full system access. Type `metatron` to launch.
+with full system access and internet search. Type `metatron` to launch.
 
 ## Flow
 1. **Model selection** — fetches the live model list from nightmare, numbered menu.
 2. **Chat** — streaming REPL against the selected model.
-3. **System access** — two ways:
-   - `!command` — run a shell command directly (e.g. `!dir`, `!whoami`).
-   - The model calls `run_command` natively (function calling) to execute commands itself.
+3. **System access** — `!command` runs a shell command directly, and the model can also
+   call `run_command` natively (function calling).
+4. **Search** — `?query` searches the internet (SearXNG), and the model can call
+   `web_search` natively.
 
 ## Requirements
 - Python 3.11 (stdlib only — `urllib`, `subprocess`, `json`). No pip deps.
-- Network reach to `10.30.20.29:11434`.
+- Network reach to `10.30.20.29:11434` and `10.30.20.35:6969`.
 
 ## Install (Commando VM)
 ```
@@ -22,13 +23,20 @@ setx PATH "%PATH%;C:\Tools\metatron"
 ```
 Then `metatron` from any shell.
 
-## Hardcoded
-- Ollama URL: `http://10.30.20.29:11434`
-- System prompt + `run_command` tool definition
-- Only the model list is dynamic (fetched from `/api/tags`).
-
 ## Controls
 | Input | Action |
 |-------|--------|
 | `exit` / `quit` | leave |
 | `!<command>` | run a shell command directly |
+| `?<query>` | search the internet (SearXNG) directly |
+
+## Hardcoded
+- Ollama URL: `http://10.30.20.29:11434`
+- SearXNG URL: `http://10.30.20.35:6969` (self-hosted meta-search, CT 516)
+- System prompt + `run_command` / `web_search` tool definitions
+- Only the model list is dynamic (fetched from `/api/tags`).
+
+## Search notes
+- `?<query>` is the guaranteed path (no LLM needed).
+- Native `web_search` tool-calling can degenerate on nightmare `.29`'s qwen3.8fast —
+  the `?` escape always works.

metatron.py

@@ -13,14 +13,17 @@ Only the model list is dynamic.
 import json
 import subprocess
 import sys
+import urllib.parse
 import urllib.request
 
 OLLAMA = "http://10.30.20.29:11434"   # nightmare
+SEARXNG = "http://10.30.20.35:6969"   # self-hosted meta-search (CT 516)
 
 SYSTEM_PROMPT = (
     "You are METATRON, a system-access AI harness running on a Windows Commando "
     "VM. You have full system access through the run_command tool — execute shell "
-    "commands when asked and report their output. Be concise and direct."
+    "commands when asked and report their output. Use web_search for any up-to-date "
+    "or factual internet information. Be concise and direct."
 )
 
 TOOLS = [{
@@ -37,6 +40,20 @@ TOOLS = [{
             "required": ["command"],
         },
     },
+}, {
+    "type": "function",
+    "function": {
+        "name": "web_search",
+        "description": "Search the internet and return the top results (title, URL, snippet). "
+                       "Use for current events, facts, anything you don't already know.",
+        "parameters": {
+            "type": "object",
+            "properties": {
+                "query": {"type": "string", "description": "The search query"},
+            },
+            "required": ["query"],
+        },
+    },
 }]
 
 C = {"r": "\033[0m", "b": "\033[1m", "c": "\033[36m", "g": "\033[32m",
@@ -94,10 +111,26 @@ def run_command(cmd):
         return f"error: {e}"
 
 
+def web_search(query):
+    url = f"{SEARXNG}/search?q={urllib.parse.quote(query)}&format=json"
+    try:
+        d = json.load(urllib.request.urlopen(url, timeout=25))
+    except Exception as e:
+        return f"search error: {e}"
+    results = d.get("results", [])
+    if not results:
+        return "no results"
+    lines = []
+    for r in results[:8]:
+        lines.append(f"- {r.get('title', '')}\n  {r.get('url', '')}\n"
+                     f"  {(r.get('content') or '')[:220]}")
+    return "\n".join(lines)
+
+
 def chat(model):
     messages = [{"role": "system", "content": SYSTEM_PROMPT}]
     print(f"\n{C['b']}  METATRON {C['c']}:: {model}{C['r']}  "
-          f"{C['y']}(!cmd = run a command, exit = quit){C['r']}\n")
+          f"{C['y']}(!cmd = run, ?query = search, exit = quit){C['r']}\n")
     while True:
         try:
             user = input(f"{C['g']}you>{C['r']} ").strip()
@@ -111,6 +144,11 @@ def chat(model):
             print(f"{C['y']}  $ {user[1:]}{C['r']}")
             print(f"  {run_command(user[1:])}\n")
             continue
+        if user.startswith("?"):
+            q = user[1:].strip()
+            print(f"{C['c']}  [search] {q}{C['r']}")
+            print(f"  {web_search(q)}\n")
+            continue
 
         messages.append({"role": "user", "content": user})
         # agentic loop: let the model call run_command until it's satisfied
@@ -138,16 +176,23 @@ def chat(model):
                                  "tool_calls": tool_calls})
                 for tc in tool_calls:
                     fn = tc.get("function", {})
-                    if fn.get("name") == "run_command":
+                    name = fn.get("name")
                     try:
                         args = json.loads(fn.get("arguments", "{}"))
                     except Exception:
                         args = {}
+                    if name == "run_command":
                         cmd = args.get("command", "")
                         print(f"\n{C['y']}  $ {cmd}{C['r']}")
                         out = run_command(cmd)
                         print(f"  {out}")
                         messages.append({"role": "tool", "content": out})
+                    elif name == "web_search":
+                        q = args.get("query", "")
+                        print(f"\n{C['c']}  [search] {q}{C['r']}")
+                        out = web_search(q)
+                        print(f"  {out[:600]}")
+                        messages.append({"role": "tool", "content": out})
                 continue  # re-send to model with tool results
             else:
                 messages.append({"role": "assistant", "content": buf})