AI: personal assistants without the dead API, and keep Ollama warm

Two things the field report asked for.

1) PERSONAL ASSISTANTS (replacing the sunset OpenAI Assistants API)

The old implementation gave three capabilities. Two are reimplemented here,
the third was confirmed unused and is deliberately not replaced:

 * per-user persona - it already lived in system_gpt_settings.json; it was
   only ever being shipped to OpenAI. It is now the system prompt.
 * per-user conversation thread - OpenAI held this server-side. It now lives
   in assistant_memory.json, keyed by discord user id, trimmed to the most
   recent turns (CONJURER_ASSISTANT_MEMORY_TURNS) and written atomically so a
   torn write cannot lose someone's history. Deliberately a plain trim, not
   the AI summarisation used for the bar's shared memory: these are private
   DMs and must not end up in a public "legend".
 * file_search - not replaced. Confirmed not in use.

The conversation goes through handle_response with request_type="NONE" and an
explicit message list, which keeps it out of the bar's shared memory. The big
win: create_chat_assistant hardcoded model="gpt-4o", so assistants were locked
to OpenAI. They now run on whatever $gadaj_teraz selects - Claude and Ollama
included.

create_chat_assistant / chat_with_assistant are gone, and with them the last
call to beta.threads in the startup path - so the cog cannot be killed by that
API again. (add_files_to_vector_store / delete_files_from_vector_store still
reference beta.assistants but are dead code - nothing calls them - so they
cannot crash anything; left alone rather than widening this change.)

2) KEEPING A SELF-HOSTED MODEL WARM

Loading is the slow part - the GPU is shared with other users - so we preload
via Ollama's documented mechanism: /api/generate with a model, a keep_alive
and NO prompt. It loads the model and generates nothing.

 * on switching to ollama, $gadaj_teraz fires a preload in the BACKGROUND
   (not awaited: loading can take minutes and the command must answer at
   once), so the wait lands on the operator rather than the first user;
 * a warm loop re-asserts keep_alive every CONJURER_OLLAMA_WARM_MINUTES.

Both are hard-guarded on the ACTIVE provider being ollama. Warming a metered
API would burn tokens and money for nothing, so that guard is pinned by a test
asserting the preload is never called for gpt/claude, and another asserting the
preload body carries no prompt (a prompt would make every warm-up generate).

Tests: 82 unit + 71 integration green.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

ai_commands.py

@@ -1,4 +1,5 @@
 # ai command cogs
+import asyncio
 import logging
 import re
 import sys
@@ -17,7 +18,7 @@ from communication_subroutine import AI_QUERY_Q
 
 import ai_functions
 from constants import (
-    ASSISTANTS,
+    OLLAMA_WARM_MINUTES,
     DATA,
     GRAPHICS_PATH,
     INITIAL_TIME_WAIT,
@@ -101,55 +102,45 @@ class Events(commands.Cog):
             text = text[1900:]
 
     async def cog_load(self):
-        # The AI query worker must run regardless of the OpenAI guard below - it
+        # The AI query worker answers via handle_response, so it works on every
-        # answers via handle_response, which works on Claude too. Start it first.
+        # backend. Start it first.
         if not self.ai_query_worker.is_running():
             self.ai_query_worker.start()
-        self.logger.info("Starting personal assistants")
+        # Keeps a self-hosted model resident; it no-ops on any other provider.
-        # Personal assistants use the OpenAI Assistants API (threads/runs), which
+        if not self.ollama_warm_loop.is_running():
-        # has no Anthropic equivalent - skip cleanly when OpenAI isn't wired up
+            self.ollama_warm_loop.start()
-        # (e.g. a Claude-only deployment) instead of crashing the cog load.
+        # NOTE: there is no OpenAI-Assistants bootstrap any more. It called a
-        if OPENAICLIENT is None:
+        # sunset API (beta threads), 404'd, and failed the WHOLE extension -
-            self.logger.warning(
+        # taking every AI command with it. Personal assistants now ride
-                "OPENAICLIENT niedostępny - osobiści asystenci (OpenAI Assistants API) wyłączeni"
+        # handle_response with per-user memory (ai_functions), so they work on
-            )
+        # Claude and Ollama too and nothing has to be created at startup.
-            return
+        self.logger.info("Osobiści asystenci: pamięć per-user, aktywny backend AI")
-        for superfryta_id, superfryta in SPECJALNE_ZIEMNIACZKI.items():
 
-            if superfryta[4] != "":
+    @tasks.loop(minutes=OLLAMA_WARM_MINUTES)
-                self.logger.info(
+    async def ollama_warm_loop(self):
-                    "Personal assistant for user: %s, exists id: %s,name: %s, owner: %s, special instructions: %s assistant id: %s ",
+        """Keep a self-hosted model resident so users don't pay the load wait.
-                    superfryta_id,
+
-                    superfryta[0],
+        Loading is the slow part on a GPU shared with other users, so we
-                    superfryta[1],
+        re-assert Ollama's keep_alive well inside its window. This preloads
-                    superfryta[2],
+        WITHOUT generating - no tokens, no cost.
-                    superfryta[3],
+
-                    superfryta[4],
+        Hard guard: it does nothing unless the ACTIVE backend is Ollama. Firing
-                )
+        warm-ups at a metered API would burn tokens and money for nothing.
-                thread = await OPENAICLIENT.beta.threads.create()
+        """
-                self.logger.info("Thread id: %s", thread.id)
+        try:
-                ASSISTANTS[superfryta[1]] = (
+            if ai_functions.active_provider() != "ollama":
-                    superfryta[2],
+                return
-                    superfryta[4],
+            await ai_functions.warm_active_model()
-                    superfryta[0],
+        except Exception as exc:  # pylint: disable=broad-exception-caught
-                    thread,
+            self.logger.info("Rozgrzewanie Ollamy nieudane (nieszkodliwe): %s", exc)
-                )
+
-            else:
+    @ollama_warm_loop.before_loop
-                self.logger.info(
+    async def before_ollama_warm_loop(self):
-                    "Creating personal assistant for user: %s, id: %s,name: %s, owner: %s, special instructions: %s",
+        await self.bot.wait_until_ready()
-                    superfryta_id,
-                    superfryta[0],
-                    superfryta[1],
-                    superfryta[2],
-                    superfryta[3],
-                )
-                await ai_functions.create_chat_assistant(
-                    superfryta_id, superfryta[0], superfryta[1], superfryta[2], superfryta[3]
-                )
-        self.logger.info("Started personal assistants")
 
     async def cog_unload(self):
         self.ai_query_worker.cancel()
+        self.ollama_warm_loop.cancel()
 
     @commands.hybrid_command(
         name="switch_dm_mode",
@@ -174,19 +165,57 @@ class Events(commands.Cog):
                 await ctx.reply("Nope. Nie wiesz jak użyć")
 
     @commands.hybrid_command(
-        name="gadaj_teraz",
+        name="modele_ai",
-        description="Pokaż/przełącz backend AI (bez argumentu = status). Przełączanie: Vykidailo.",
+        description="Pokaż modele dostępne dla danego backendu (Ollamę pyta na żywo).",
     )
-    async def gadaj_teraz(self, ctx, nazwa_konfigu: Optional[str] = None):
+    async def modele_ai(self, ctx, nazwa_konfigu: Optional[str] = None):
+        """Read-only model listing. For Ollama this queries the server, so it
+        shows exactly what is pulled on the box right now."""
+        async with ctx.channel.typing():
+            target = nazwa_konfigu or ai_functions.get_active_ai_config()
+            if target not in ai_functions.list_ai_configs():
+                await discord_friendly_reply(
+                    ctx,
+                    f"Nie znam configu '{target}'. Dostępne: "
+                    f"{', '.join(ai_functions.list_ai_configs())}",
+                )
+                return
+            try:
+                models = await ai_functions.list_provider_models(target)
+            except ai_functions.AIError as exc:
+                await discord_friendly_reply(
+                    ctx, f"Nie mogę pobrać modeli dla '{target}': {exc}"
+                )
+                return
+            if not models:
+                await discord_friendly_reply(ctx, f"Brak modeli dla '{target}'.")
+                return
+            await discord_friendly_reply(
+                ctx,
+                f"Modele dla **{target}**: {', '.join(models)}\n"
+                f"Wepniesz przez `$gadaj_teraz {target} <model>` (tylko Vykidailo).",
+            )
+
+    @commands.hybrid_command(
+        name="gadaj_teraz",
+        description="Pokaż/przełącz backend AI i model (bez argumentu = status). Przełączanie: Vykidailo.",
+    )
+    async def gadaj_teraz(
+        self, ctx, nazwa_konfigu: Optional[str] = None, model: Optional[str] = None
+    ):
         async with ctx.channel.typing():
             available = ai_functions.list_ai_configs()
             # No argument -> report the active backend (read-only, open to all).
             if not nazwa_konfigu:
                 active = ai_functions.get_active_ai_config()
+                active_cfg = ai_functions.AI_CONFIGS.get(active, {})
                 await discord_friendly_reply(
                     ctx,
-                    f"Teraz gadam przez **{active}**. Dostępne: {', '.join(available)}. "
+                    f"Teraz gadam przez **{active}** "
-                    "Przełączysz przez `$gadaj_teraz <config>` (tylko Vykidailo).",
+                    f"({active_cfg.get('provider')} / {active_cfg.get('latest_model')}). "
+                    f"Dostępne: {', '.join(available)}. "
+                    "Przełączysz przez `$gadaj_teraz <config> [model]` (tylko Vykidailo), "
+                    "modele zobaczysz przez `$modele_ai`.",
                 )
                 return
             is_admin = isinstance(ctx.author, discord.Member) and any(
@@ -208,13 +237,69 @@ class Events(commands.Cog):
                     ctx, f"Nie mogę przełączyć na '{nazwa_konfigu}': {exc}"
                 )
                 return
-            self.logger.info(
+
-                "Przełączono AI na config %s (%s)", nazwa_konfigu, cfg.get("provider")
+            # Optional second argument pins the model. For a backend we can
-            )
+            # enumerate (Ollama), reject an unknown id up front with the list -
+            # otherwise the typo only surfaces later as a failed reply.
+            if model:
+                try:
+                    known = await ai_functions.list_provider_models(nazwa_konfigu)
+                except ai_functions.AIError:
+                    known = []  # cannot enumerate -> accept verbatim
+                if known and cfg.get("provider") == "ollama" and model not in known:
                     await discord_friendly_reply(
                         ctx,
-                f"Teraz gadam przez **{nazwa_konfigu}** — {cfg.get('provider')} / {cfg.get('latest_model')}.",
+                        f"Model '{model}' nie jest wgrany na Ollamie. "
+                        f"Dostępne: {', '.join(known)}",
                     )
+                    return
+                try:
+                    cfg = ai_functions.set_active_model(model, nazwa_konfigu)
+                except (KeyError, ValueError) as exc:
+                    await discord_friendly_reply(
+                        ctx, f"Nie mogę wpiąć modelu '{model}': {exc}"
+                    )
+                    return
+
+            self.logger.info(
+                "Przełączono AI na config %s (%s / %s)",
+                nazwa_konfigu, cfg.get("provider"), cfg.get("latest_model"),
+            )
+            message = (
+                f"Teraz gadam przez **{nazwa_konfigu}** — "
+                f"{cfg.get('provider')} / {cfg.get('latest_model')}."
+            )
+            if cfg.get("provider") == "ollama":
+                # Pay the (slow, shared-GPU) load cost NOW, in the background,
+                # so it lands on the operator switching backends rather than on
+                # whoever sends the first message. Not awaited: loading can take
+                # minutes and the command must answer immediately.
+                asyncio.create_task(ai_functions.warm_active_model())
+                message += (
+                    "\nRozgrzewam model w tle — pierwsza odpowiedź może chwilę potrwać."
+                )
+            # Switched without pinning a model: show what else is on offer.
+            if not model:
+                try:
+                    others = await ai_functions.list_provider_models(nazwa_konfigu)
+                except ai_functions.AIError:
+                    others = []
+                current = cfg.get("latest_model")
+                if others and current not in others:
+                    # The configured/pinned model is not on the server: every
+                    # reply would fail with "model not found" and nothing would
+                    # say why. Flag it here, where the list is already in hand.
+                    message += (
+                        f"\n⚠ Uwaga: '{current}' nie jest wgrany na serwerze. "
+                        f"Dostępne: {', '.join(others)} "
+                        f"(`$gadaj_teraz {nazwa_konfigu} <model>`)."
+                    )
+                elif len(others) > 1:
+                    message += (
+                        f"\nDostępne modele: {', '.join(others)} "
+                        f"(`$gadaj_teraz {nazwa_konfigu} <model>`)."
+                    )
+            await discord_friendly_reply(ctx, message)
 
     @commands.hybrid_command(
         name="armia_hammera",
@@ -326,8 +411,10 @@ class Events(commands.Cog):
                 if message.author.id == superfryta[0]:
                     self.logger.info("Specjalny ziemniak")
                     if self.armia[message.author.id] == Dm_Mode.SPECJALNY_ZIEMNIACZEK:
-                        #await self.bot.process_commands(message)
+                        # superfryta = [discord_id, assistant_name, owner, instructions, legacy_assistant_id]
-                        await ai_functions.chat_with_assistant(message, superfryta[1])
+                        await ai_functions.chat_with_personal_assistant(
+                            message, superfryta[2], superfryta[3]
+                        )
                         return
                     elif self.armia[message.author.id] == Dm_Mode.ECHO_ECHO:
                         await ai_functions.echo(message)

ai_functions.py

@@ -1,15 +1,21 @@
 import asyncio
 import json
 import logging
+import os
 import random
+import tempfile
 
 import openai
 import tiktoken
 import time
 from other_functions import discord_friendly_send
+import requests
+
 from constants import (
     AI_CONFIGS,
-    ASSISTANTS,
+    AI_TIMEOUT_SECONDS,
+    ASSISTANT_MEMORY_FILE,
+    ASSISTANT_MEMORY_TURNS,
     CLAUDECLIENT,
     CYCLIC_WORDS,
     DEFAULT_AI_CONFIG,
@@ -19,6 +25,10 @@ from constants import (
     MEMORY_FIVE_SIARA,
     MESSAGE_TABLE,
     MESSAGE_TABLE_MUZYKA,
+    OLLAMACLIENT,
+    OLLAMA_KEEP_ALIVE,
+    OLLAMA_PRELOAD_TIMEOUT,
+    OLLAMA_URL,
     OPENAICLIENT,
     SYSTEM_GPT_SETTINGS,
     WORD_REACTIONS,
@@ -92,12 +102,54 @@ def set_active_ai_config(name: str) -> dict:
         raise RuntimeError("klient Anthropic nie jest skonfigurowany (brak ANTHROPIC_API_KEY)")
     if provider == "openai" and OPENAICLIENT is None:
         raise RuntimeError("klient OpenAI nie jest skonfigurowany (brak OPENAI_API_KEY)")
+    if provider == "ollama" and OLLAMACLIENT is None:
+        raise RuntimeError("Ollama nie jest skonfigurowana (ustaw CONJURER_OLLAMA_URL)")
     _ACTIVE_CONFIG_NAME = name
     _persist_active_ai_config(name)
     return cfg
 
 
-def _persist_active_ai_config(name: str) -> None:
+async def list_provider_models(name: str = None):
+    """Model ids selectable for a config.
+
+    For Ollama this ASKS THE SERVER (its OpenAI-compatible /v1/models), so the
+    picker always reflects what is actually pulled on the box rather than a
+    hardcoded list. Hosted providers are not enumerated - we only report what
+    the config is wired to.
+    """
+    cfg = AI_CONFIGS.get(name or _ACTIVE_CONFIG_NAME) or _active_config()
+    if cfg.get("provider") == "ollama":
+        if OLLAMACLIENT is None:
+            raise AIError(
+                "auth",
+                RuntimeError("Ollama nie jest skonfigurowana (ustaw CONJURER_OLLAMA_URL)"),
+            )
+        try:
+            resp = await OLLAMACLIENT.models.list()
+        except Exception as exc:  # pylint: disable=broad-except
+            raise _map_openai_error(exc)
+        return sorted({item.id for item in resp.data})
+    return [m for m in (cfg.get("latest_model"), cfg.get("cheap_model")) if m]
+
+
+def set_active_model(model: str, name: str = None) -> dict:
+    """Pin the model a config uses for normal replies, and persist it.
+
+    Only ``latest_model`` is changed; ``cheap_model`` stays as configured so the
+    MUSIC path keeps its cheaper backend.
+    """
+    cfg_name = name or _ACTIVE_CONFIG_NAME
+    if cfg_name not in AI_CONFIGS:
+        raise KeyError(cfg_name)
+    if not model or not model.strip():
+        raise ValueError("pusta nazwa modelu")
+    cfg = AI_CONFIGS[cfg_name]
+    cfg["latest_model"] = model.strip()
+    _persist_active_ai_config(_ACTIVE_CONFIG_NAME, model_for=cfg_name)
+    return cfg
+
+
+def _persist_active_ai_config(name: str, model_for: str = None) -> None:
     """Best-effort write of the active-config choice into system_gpt_settings.json.
 
     Keeps the historical two-element structure intact: updates index 2 if it
@@ -117,7 +169,16 @@ def _persist_active_ai_config(name: str) -> None:
         return
     if len(data) > 2 and isinstance(data[2], dict):
         data[2]["active"] = name
-        data[2].setdefault("configs", AI_CONFIGS)
+        # Write back ONLY what this process actually changed. Assigning the whole
+        # in-memory AI_CONFIGS here would clobber operator hand-edits (the only
+        # way to change cheap_model/temperature/max_tokens) and re-seed configs
+        # deliberately deleted from the file, because AI_CONFIGS is the built-in
+        # defaults merged under the file. setdefault alone is not enough either:
+        # it would drop a model pinned via set_active_model, hence model_for.
+        block = data[2].setdefault("configs", AI_CONFIGS)
+        if model_for and model_for in AI_CONFIGS:
+            entry = block.setdefault(model_for, dict(AI_CONFIGS[model_for]))
+            entry["latest_model"] = AI_CONFIGS[model_for]["latest_model"]
     else:
         data = data[:2] + [{"active": name, "configs": AI_CONFIGS}]
     try:
@@ -214,12 +275,83 @@ async def _anthropic_call(messages, model, cfg):
     return text.strip()
 
 
+async def _ollama_call(messages, model, cfg):
+    """Self-hosted counterpart of openai_call. Returns a plain string.
+
+    Ollama exposes an OpenAI-compatible /v1 surface, so the same message format
+    and the same error mapping apply - only the base_url and the model ids
+    differ. Chat Completions (not the Responses API) is what Ollama implements.
+    """
+    if OLLAMACLIENT is None:
+        raise AIError(
+            "auth",
+            RuntimeError("Ollama nie jest skonfigurowana (ustaw CONJURER_OLLAMA_URL)"),
+        )
+    try:
+        resp = await OLLAMACLIENT.chat.completions.create(
+            model=model,
+            messages=messages,
+            temperature=float(cfg.get("temperature", 0.2)),
+        )
+    except Exception as exc:  # pylint: disable=broad-except
+        raise _map_openai_error(exc)
+    return (resp.choices[0].message.content or "").strip()
+
+
+def _ollama_preload(model, keep_alive=None) -> bool:
+    """Load ``model`` into Ollama and keep it resident, generating NOTHING.
+
+    Ollama's /api/generate with a model and no prompt is the documented preload:
+    it pays the (slow, GPU-shared) load cost once and returns, producing no
+    tokens. Used to warm up on switch and to re-assert keep_alive periodically.
+
+    Blocking on purpose - callers wrap it in asyncio.to_thread.
+    """
+    if not OLLAMA_URL:
+        return False
+    logger = logging.getLogger("discord")
+    try:
+        resp = requests.post(
+            f"{OLLAMA_URL}/api/generate",
+            json={"model": model, "keep_alive": keep_alive or OLLAMA_KEEP_ALIVE},
+            timeout=OLLAMA_PRELOAD_TIMEOUT,
+        )
+        ok = resp.status_code == 200
+        logger.info("Ollama preload %s -> HTTP %s", model, resp.status_code)
+        return ok
+    except requests.exceptions.RequestException as exc:
+        logger.info("Ollama preload %s failed: %s", model, exc)
+        return False
+
+
+def active_provider() -> str:
+    """Provider of the active config - the guard every warm-up must check.
+
+    Preloading only makes sense for a self-hosted model; firing it at a metered
+    API would burn tokens (and money) for nothing.
+    """
+    return (_active_config() or {}).get("provider", "")
+
+
+async def warm_active_model(force_model=None) -> bool:
+    """Preload the active model IFF the active backend is Ollama."""
+    if active_provider() != "ollama":
+        return False
+    cfg = _active_config()
+    model = force_model or cfg.get("latest_model")
+    if not model:
+        return False
+    return await asyncio.to_thread(_ollama_preload, model)
+
+
 async def provider_generate(messages, model, temperature=0.2):
     """Dispatch a chat completion to the active backend, normalising errors."""
     cfg = _active_config()
     try:
         if cfg.get("provider") == "anthropic":
             return await _anthropic_call(messages, model, cfg)
+        if cfg.get("provider") == "ollama":
+            return await _ollama_call(messages, model, cfg)
         return await openai_call(messages, model, temperature)
     except AIError:
         raise
@@ -459,7 +591,7 @@ async def handle_response(
     try:
         # ...przygotowanie messages/system prompt/itp. jak masz...
         # retry/backoff + deadline (zachowuje Twoją semantykę logowania)
-        timeout_sec = 120
+        timeout_sec = AI_TIMEOUT_SECONDS
         deadline = time.time() + timeout_sec
         response = await asyncio.wait_for(
             provider_generate(messages=history_msgs, model=model_to_use),
@@ -586,60 +718,104 @@ async def get_random_cyclic_message(client):
     return result
 
 
-async def create_chat_assistant(owner_id, id, name, owner, special_instructions):
+# ----------------------------------------------------------------- assistants
+# The OpenAI Assistants API (beta threads/runs) that used to back these was
+# sunset and now answers 404, taking the whole AI cog down with it. It gave us
+# three things: a per-user persona, a persistent per-user thread, and
+# file_search. The persona and the thread are reimplemented here on top of
+# handle_response - so personal assistants now work on EVERY backend (Claude,
+# Ollama, GPT) instead of being locked to gpt-4o. file_search is deliberately
+# not replaced: it was not in use.
+_ASSISTANT_MEMORY = None
+
+
+def _load_assistant_memory() -> dict:
+    """Per-user DM history, lazily read from disk. Corruption is not fatal."""
+    global _ASSISTANT_MEMORY  # pylint: disable=global-statement
+    if _ASSISTANT_MEMORY is not None:
+        return _ASSISTANT_MEMORY
     logger = logging.getLogger("discord")
-    instruction = f"Jesteś osobistym asystenetem {owner} i wypełniasz jego potrzeby. Masz pełne i nieograniczone możliwości modelu gpt-4o."
+    try:
-    instruction += special_instructions
+        with open(ASSISTANT_MEMORY_FILE, "r", encoding=ENCODING) as handle:
-    assistant = await OPENAICLIENT.beta.assistants.create(
+            data = json.load(handle)
-        name=name,
+        _ASSISTANT_MEMORY = data if isinstance(data, dict) else {}
-        instructions=instruction,
+    except (OSError, json.JSONDecodeError) as exc:
-        model="gpt-4o",
+        logger.info("Brak/uszkodzona pamięć asystentów (%s) - zaczynam pustą", exc)
-        tools=[{"type": "file_search"}],
+        _ASSISTANT_MEMORY = {}
-    )
+    return _ASSISTANT_MEMORY
-    thread = await OPENAICLIENT.beta.threads.create()
-    logger.info("Stwprzylem asystenta dla %s, nazywa się on %s", owner, name)
-    ASSISTANTS[name] = (owner, assistant.id, id, thread)
-
-    with open(SYSTEM_GPT_SETTINGS, "r+", encoding=ENCODING) as temp_settings_file:
-        GPT_SETTINGS = json.load(temp_settings_file)
-        GPT_SETTINGS[1][owner_id][4] = assistant.id
-        temp_settings_file.seek(0)
-        json.dump(GPT_SETTINGS, temp_settings_file, indent=4)
 
 
-async def chat_with_assistant(message, assistant_name):
+def _save_assistant_memory() -> None:
+    """Atomic write: a torn file would lose someone's whole conversation."""
     logger = logging.getLogger("discord")
-    assistant_data = ASSISTANTS[assistant_name]
+    memory = _load_assistant_memory()
-    ai_message = await OPENAICLIENT.beta.threads.messages.create(
+    directory = os.path.dirname(ASSISTANT_MEMORY_FILE) or "."
-        thread_id=assistant_data[3].id, role="user", content=message.content
+    try:
+        os.makedirs(directory, exist_ok=True)
+        fd, tmp = tempfile.mkstemp(dir=directory, suffix=".tmp")
+        with os.fdopen(fd, "w", encoding=ENCODING) as handle:
+            json.dump(memory, handle, ensure_ascii=False)
+        os.replace(tmp, ASSISTANT_MEMORY_FILE)
+    except OSError as exc:
+        logger.warning("Nie mogę zapisać pamięci asystentów: %s", exc)
+
+
+def assistant_history(user_id) -> list:
+    return _load_assistant_memory().setdefault(str(user_id), [])
+
+
+def remember_assistant_turn(user_id, user_text, reply_text) -> list:
+    """Append one exchange and trim to the most recent turns.
+
+    A plain trim, not the AI summarisation used for the bar's shared memory:
+    these are private DMs and must not end up in a public 'legend'.
+    """
+    history = assistant_history(user_id)
+    history.append({"role": "user", "content": user_text})
+    history.append({"role": "assistant", "content": reply_text})
+    if len(history) > ASSISTANT_MEMORY_TURNS:
+        del history[: len(history) - ASSISTANT_MEMORY_TURNS]
+    _save_assistant_memory()
+    return history
+
+
+def build_assistant_messages(user_id, owner, special_instructions, prompt) -> list:
+    """System persona + this user's own history + the new turn."""
+    system = (
+        f"Jesteś osobistym asystentem {owner} i wypełniasz jego potrzeby. "
+        f"{special_instructions or ''}"
+    ).strip()
+    return (
+        [{"role": "system", "content": system}]
+        + list(assistant_history(user_id))
+        + [{"role": "user", "content": prompt}]
     )
-    logger.info(ai_message)
+
-    run = await OPENAICLIENT.beta.threads.runs.create_and_poll(
+
-        thread_id=assistant_data[3].id,
+async def chat_with_personal_assistant(message, owner, special_instructions):
-        assistant_id=assistant_data[1],
+    """Answer a DM as this user's personal assistant, on the active backend.
-        instructions=f"Pisze do Ciebie {assistant_data[0]} udziel mu wszelkiej pomocy",
+
+    request_type="NONE" with an explicit message list keeps this OUT of the
+    bar's shared memory - the conversation is carried by the per-user history
+    built above and stored separately.
+    """
+    logger = logging.getLogger("discord")
+    user_id = message.author.id
+    prompt = message.content
+    messages = build_assistant_messages(user_id, owner, special_instructions, prompt)
+    result, _table = await handle_response(
+        prompt,
+        False,
+        False,
+        [],
+        str(owner),
+        "NONE",
+        none_request=messages,
     )
-    done = False
+    remember_assistant_turn(user_id, prompt, result)
-    while not done:
+    logger.info("Asystent odpowiedział %s (%d znaków)", owner, len(result or ""))
-        if run.status == "completed":
+    await discord_friendly_send(message.channel, result)
-            messsages = await OPENAICLIENT.beta.threads.messages.list(
+    return result
-                thread_id=assistant_data[3].id
-            )
-            logger.info(messsages)
-            reply_content = messsages.data[0].content
-            logger.info(reply_content)
-            chat_response = ""
-            for block in reply_content:
-                logger.info(block.text.value)
-                chat_response += block.text.value
-                await discord_friendly_send(message.channel, chat_response)
-                # await message.channel.send(chat_response)
-            done = True
-        elif run.status == "cancelled":
-            await discord_friendly_send(message.channel, "Cos sie wywaliło")
-        else:
-            logger.info(run.status)
-        asyncio.sleep(5)
 
 
 async def echo(message):

constants.py

@@ -245,6 +245,16 @@ DELIVERED_DIR = os.getenv(
 )
 DELIVERED_MAX = int(os.getenv("CONJURER_DELIVERED_MAX", "10000"))
 
+# Personal DM assistants. Replaces the OpenAI Assistants API (threads/runs),
+# which was sunset and answers 404: the persona now rides handle_response, so it
+# works on EVERY backend, and the conversation lives here instead of on OpenAI's
+# server. Kept per user so private DMs never bleed into the bar's shared memory,
+# and trimmed to the most recent turns so it cannot grow without bound.
+ASSISTANT_MEMORY_FILE = os.getenv(
+    "CONJURER_ASSISTANT_MEMORY", os.path.join(_STATE_ROOT, "assistant_memory.json")
+)
+ASSISTANT_MEMORY_TURNS = int(os.getenv("CONJURER_ASSISTANT_MEMORY_TURNS", "40"))
+
 FILE_SERVICE_ADDRESS = os.getenv("CONJURER_FILE_SERVICE", "http://192.168.1.15:5000")
 RADIO_HARBOR_ADDRESS = os.getenv("CONJURER_RADIO_HARBOR", "http://192.168.1.15:54321")
 # Betoniarka (radio-operator service colocated with Liquidsoap). Falls back to
@@ -400,6 +410,38 @@ if anthropic and ANTHROPIC_API_KEY:
 else:
     CLAUDECLIENT = None
 
+# Ollama (self-hosted models). There is no API key - the endpoint IS the whole
+# configuration, so the feature stays dormant until CONJURER_OLLAMA_URL is set
+# (same pattern as the Conan bridge). We talk to Ollama's OpenAI-COMPATIBLE
+# surface (/v1) with the openai SDK we already depend on, which means the
+# existing message format and _map_openai_error handling work unchanged.
+# How long handle_response waits for ANY backend before giving up. 120s was
+# hardcoded and is fine for hosted APIs, but a self-hosted model on a modest GPU
+# can legitimately take longer, so it is now tunable.
+AI_TIMEOUT_SECONDS = int(os.getenv("CONJURER_AI_TIMEOUT_SECONDS", "120"))
+
+OLLAMA_URL = os.getenv("CONJURER_OLLAMA_URL", "").rstrip("/")
+# Keeping a self-hosted model resident. Loading it is the slow part (it is
+# offloaded to a GPU shared with other users), so we preload it - Ollama's
+# /api/generate with a model and NO prompt loads it and generates nothing, which
+# costs no tokens and no money. KEEP_ALIVE is how long Ollama should then hold
+# it; the warm loop re-asserts that well inside the window.
+# STRICTLY Ollama-only: doing this against a paid API would burn tokens for
+# nothing, so every caller checks the active provider first.
+OLLAMA_KEEP_ALIVE = os.getenv("CONJURER_OLLAMA_KEEP_ALIVE", "30m")
+OLLAMA_WARM_MINUTES = float(os.getenv("CONJURER_OLLAMA_WARM_MINUTES", "10"))
+# A preload waits for the model to finish loading, which on a shared GPU is the
+# slow path we are trying to move off the user's first message.
+OLLAMA_PRELOAD_TIMEOUT = int(os.getenv("CONJURER_OLLAMA_PRELOAD_TIMEOUT", "600"))
+
+OLLAMA_LATEST_MODEL = os.getenv("CONJURER_OLLAMA_MODEL", "llama3.1:8b")
+OLLAMA_CHEAP_MODEL = os.getenv("CONJURER_OLLAMA_CHEAP_MODEL", OLLAMA_LATEST_MODEL)
+if openai and OLLAMA_URL:
+    # api_key is required by the SDK but ignored by Ollama.
+    OLLAMACLIENT = openai.AsyncOpenAI(base_url=f"{OLLAMA_URL}/v1", api_key="ollama")
+else:
+    OLLAMACLIENT = None
+
 TOKEN = _resolve_token("discord", "DISCORD_TOKEN")
 
 # Voice recognition (AssemblyAI). None = the voice cog reports and disables.
@@ -489,6 +531,12 @@ def _default_ai_configs():
             # sent for Claude (Opus 4.8 / Sonnet 5 reject sampling params).
             "max_tokens": 2048,
         },
+        "ollama": {
+            "provider": "ollama",
+            "latest_model": OLLAMA_LATEST_MODEL,
+            "cheap_model": OLLAMA_CHEAP_MODEL,
+            "temperature": 0.2,
+        },
         # Template for wiring further providers. Copy it, rename the key, point
         # "provider" at a backend ai_functions.provider_generate implements, and
         # fill in the model ids. Keys starting with "_" are treated as inert
@@ -508,7 +556,12 @@ _ai_block = (
     if isinstance(GPT_SETTINGS, list) and len(GPT_SETTINGS) > 2 and isinstance(GPT_SETTINGS[2], dict)
     else {}
 )
-AI_CONFIGS = _ai_block.get("configs") or _default_ai_configs()
+# Built-in defaults FIRST, then whatever the settings file defines on top. The
+# file cannot simply win outright: every provider switch persists a "configs"
+# block, so a file written by an older build would permanently hide providers
+# added later (ollama) from the picker.
+AI_CONFIGS = _default_ai_configs()
+AI_CONFIGS.update(_ai_block.get("configs") or {})
 # Single switch: env var wins, then the settings-file "active" key, then "gpt".
 DEFAULT_AI_CONFIG = (
     os.getenv("CONJURER_AI_CONFIG")

docker/env/bot.env.example

@@ -13,10 +13,29 @@ CONJURER_NETRC_FILE=/secrets/.netrc
 
 # --- AI backend switch --------------------------------------------------
 # Which AI config from system_gpt_settings.json is active at startup
-# (e.g. "gpt" or "claude"). Runtime switch: $gadaj_teraz <config>. Unset =
+# (e.g. "gpt", "claude" or "ollama"). Runtime switch:
-# whatever the settings file's "active" key says, falling back to "gpt".
+# $gadaj_teraz <config> [model]. Unset = whatever the settings file's "active"
+# key says, falling back to "gpt".
 # CONJURER_AI_CONFIG=gpt
 
+# --- Ollama (self-hosted models) ----------------------------------------
+# The endpoint IS the whole configuration - no API key. Leave unset and the
+# "ollama" backend simply refuses to be selected. In-cluster, use the Service
+# DNS name; from outside, host:port. Port 11434 is Ollama's default.
+#   CONJURER_OLLAMA_URL=http://ollama.ollama.svc.cluster.local:11434
+# CONJURER_OLLAMA_URL=
+# DEFAULT model for normal replies. $modele_ai lists what the server actually
+# has pulled. Pinning one at runtime with `$gadaj_teraz ollama <model>` is
+# persisted into system_gpt_settings.json and from then on WINS over this
+# variable - the pin is the more recent, more explicit choice. Clear the
+# "latest_model" of the ollama entry in that file to fall back to this default.
+# CONJURER_OLLAMA_MODEL=llama3.1:8b
+# Model used for the cheaper MUSIC path; defaults to CONJURER_OLLAMA_MODEL.
+# CONJURER_OLLAMA_CHEAP_MODEL=
+# How long to wait for ANY backend to answer. 120s suits hosted APIs; a
+# self-hosted model on a modest GPU may need more.
+# CONJURER_AI_TIMEOUT_SECONDS=120
+
 # --- Data ---------------------------------------------------------------
 # Single mounted volume; all writable state is rooted here.
 CONJURER_DATA_DIR=/data

tests/integration/test_librarian_heartbeat.py

@@ -60,25 +60,49 @@ def test_current_search_registration_round_trip():
         assert not lib._current_search
 
 
-def test_search_fills_progress_with_live_positions_and_total(tmp_path, monkeypatch):
+def _write_two_chunks(tmp_path):
-    # End to end against the real scan: total_bytes matches the chunk files on
-    # disk, and once finished the recorded offsets cover them.
-    monkeypatch.setattr(search_bot, "DATABASE_PATH", str(tmp_path) + "/")
     (tmp_path / "0_chunk.txt").write_text("10.1/a\n10.1/b\n", encoding="utf-8")
     (tmp_path / "1_chunk.txt").write_text("10.1/c\n", encoding="utf-8")
-    expected_total = sum(
+    return sum(
         (tmp_path / name).stat().st_size for name in ("0_chunk.txt", "1_chunk.txt")
     )
 
+
+def test_search_fills_progress_and_reaches_full_coverage(tmp_path, monkeypatch):
+    # Coverage must be measured on a search that CANNOT stop early. Once every
+    # queried DOI is found the consumer signals TERM and the producers stop
+    # mid-file, so a search for a DOI that exists reaches an arbitrary offset -
+    # asserting 100% there is a race (it failed roughly one run in two).
+    # An absent DOI forces the whole database to be read.
+    monkeypatch.setattr(search_bot, "DATABASE_PATH", str(tmp_path) + "/")
+    expected_total = _write_two_chunks(tmp_path)
+
+    progress = {}
+    search_bot.search_for_doi([("10.9/absent", "DATA")], [], _LOG, progress=progress)
+
+    assert progress["total_bytes"] == expected_total
+    assert progress["chunk_files"] == 2
+    done, total, percent = lib._progress_summary(progress)
+    assert total == expected_total
+    assert done == expected_total          # nothing stopped it: whole DB scanned
+    assert percent == pytest.approx(100.0)
+
+
+def test_progress_is_populated_for_a_search_that_finds_its_target(tmp_path, monkeypatch):
+    # The early-termination case: the target is found, so coverage is whatever
+    # the producers reached. Assert what IS deterministic - the total is known,
+    # progress is bounded and sane, and the hit is reported.
+    monkeypatch.setattr(search_bot, "DATABASE_PATH", str(tmp_path) + "/")
+    expected_total = _write_two_chunks(tmp_path)
+
     progress = {}
     result, _positions, _interrupted = search_bot.search_for_doi(
         [("10.1/c", "DATA")], [], _LOG, progress=progress
     )
 
     assert progress["total_bytes"] == expected_total
-    assert progress["chunk_files"] == 2
     done, total, percent = lib._progress_summary(progress)
     assert total == expected_total
-    assert done == expected_total          # whole DB scanned
+    assert 0 <= done <= total              # bounded, never nonsense
-    assert percent == pytest.approx(100.0)
+    assert 0.0 <= percent <= 100.0
     assert [r for r in result if r["DOI"] == "10.1/c" and r["exists"]]

tests/unit/test_ai_provider_switch.py

@@ -136,3 +136,401 @@ def test_map_openai_error_categories():
     assert ai_functions._map_openai_error(_bare(openai.RateLimitError)).category == "rate_limit"
     assert ai_functions._map_openai_error(_bare(openai.APITimeoutError)).category == "timeout"
     assert ai_functions._map_openai_error(ValueError("x")).category == "api"
+
+
+# ---------------------------------------------------------------- Ollama
+# The self-hosted backend is wired through Ollama's OpenAI-compatible surface,
+# so it reuses the message format and the error mapping above. What is new and
+# worth pinning: it must appear in the picker even on an upgraded settings file,
+# models come from the SERVER, and pinning one must stick.
+import asyncio  # noqa: E402
+
+
+class _FakeModel:
+    def __init__(self, ident):
+        self.id = ident
+
+
+class _FakeModels:
+    def __init__(self, ids, raises=None):
+        self._ids = ids
+        self._raises = raises
+
+    async def list(self):
+        if self._raises:
+            raise self._raises
+        return types.SimpleNamespace(data=[_FakeModel(i) for i in self._ids])
+
+
+class _FakeOllamaClient:
+    def __init__(self, ids=(), raises=None):
+        self.models = _FakeModels(list(ids), raises)
+
+
+def test_ollama_config_is_offered_in_the_picker():
+    assert "ollama" in ai_functions.list_ai_configs()
+    cfg = ai_functions.AI_CONFIGS["ollama"]
+    assert cfg["provider"] == "ollama"
+
+
+def test_select_model_uses_ollama_models_when_active(monkeypatch):
+    monkeypatch.setitem(
+        ai_functions.AI_CONFIGS,
+        "ollama",
+        {"provider": "ollama", "latest_model": "llama3.1:8b", "cheap_model": "qwen2.5:3b"},
+    )
+    _reset_active("ollama")
+    try:
+        # the legacy gpt-4o default must auto-map, not leak to Ollama
+        assert ai_functions.select_model("GENERAL", "gpt-4o") == "llama3.1:8b"
+        assert ai_functions.select_model("MUSIC", "gpt-4o") == "qwen2.5:3b"
+        # an explicit id is still honoured verbatim
+        assert ai_functions.select_model("GENERAL", "mistral:7b") == "mistral:7b"
+    finally:
+        _reset_active("gpt")
+
+
+def test_list_provider_models_queries_the_ollama_server(monkeypatch):
+    monkeypatch.setattr(
+        ai_functions, "OLLAMACLIENT", _FakeOllamaClient(["b:2", "a:1", "a:1"])
+    )
+    models = asyncio.run(ai_functions.list_provider_models("ollama"))
+    assert models == ["a:1", "b:2"]  # sorted + de-duplicated
+
+
+def test_list_provider_models_for_hosted_provider_reports_configured_ids():
+    models = asyncio.run(ai_functions.list_provider_models("gpt"))
+    assert models == [
+        ai_functions.AI_CONFIGS["gpt"]["latest_model"],
+        ai_functions.AI_CONFIGS["gpt"]["cheap_model"],
+    ]
+
+
+def test_list_provider_models_wraps_server_failure(monkeypatch):
+    monkeypatch.setattr(
+        ai_functions, "OLLAMACLIENT", _FakeOllamaClient(raises=ValueError("boom"))
+    )
+    try:
+        asyncio.run(ai_functions.list_provider_models("ollama"))
+    except ai_functions.AIError as exc:
+        assert exc.category == "api"
+    else:
+        raise AssertionError("a server failure must surface as AIError")
+
+
+def test_list_provider_models_without_endpoint_is_an_auth_error(monkeypatch):
+    # Contrast, so the assertion cannot pass vacuously: with a client present the
+    # call succeeds, and ONLY setting it to None turns it into an auth error.
+    monkeypatch.setattr(ai_functions, "OLLAMACLIENT", _FakeOllamaClient(["a:1"]))
+    assert asyncio.run(ai_functions.list_provider_models("ollama")) == ["a:1"]
+
+    monkeypatch.setattr(ai_functions, "OLLAMACLIENT", None)
+    try:
+        asyncio.run(ai_functions.list_provider_models("ollama"))
+    except ai_functions.AIError as exc:
+        assert exc.category == "auth"
+    else:
+        raise AssertionError("an unconfigured Ollama must surface as AIError")
+
+
+def test_set_active_model_pins_latest_and_keeps_cheap(monkeypatch):
+    monkeypatch.setitem(
+        ai_functions.AI_CONFIGS,
+        "ollama",
+        {"provider": "ollama", "latest_model": "old:1", "cheap_model": "cheap:1"},
+    )
+    written = {}
+    monkeypatch.setattr(
+        ai_functions,
+        "_persist_active_ai_config",
+        lambda name, model_for=None: written.update(name=name, model_for=model_for),
+    
+    )
+    ai_functions.set_active_model("mistral:7b", "ollama")
+    # Assert on the shared registry, not on the returned object - that object IS
+    # the mutated dict, so asserting on it would pass even if nothing was stored.
+    stored = ai_functions.AI_CONFIGS["ollama"]
+    assert stored["latest_model"] == "mistral:7b"
+    assert stored["cheap_model"] == "cheap:1"   # MUSIC path untouched
+    assert written["name"]                      # the choice was persisted...
+    assert written["model_for"] == "ollama"     # ...scoped to the config we changed
+
+
+def test_set_active_model_rejects_blank_and_unknown_config(monkeypatch):
+    monkeypatch.setattr(
+        ai_functions, "_persist_active_ai_config", lambda _n, model_for=None: None
+    )
+    for bad in ("", "   "):
+        try:
+            ai_functions.set_active_model(bad, "gpt")
+        except ValueError:
+            pass
+        else:
+            raise AssertionError("a blank model id must be rejected")
+    try:
+        ai_functions.set_active_model("x", "nie-ma-takiego")
+    except KeyError:
+        pass
+    else:
+        raise AssertionError("an unknown config must be rejected")
+
+
+def test_switching_to_ollama_without_endpoint_explains_itself(monkeypatch):
+    monkeypatch.setattr(ai_functions, "OLLAMACLIENT", None)
+    try:
+        ai_functions.set_active_ai_config("ollama")
+    except RuntimeError as exc:
+        assert "CONJURER_OLLAMA_URL" in str(exc)
+    else:
+        raise AssertionError("switching to an unconfigured Ollama must raise")
+    finally:
+        _reset_active("gpt")
+
+
+def test_provider_generate_routes_to_ollama(monkeypatch):
+    monkeypatch.setitem(
+        ai_functions.AI_CONFIGS,
+        "ollama",
+        {"provider": "ollama", "latest_model": "m:1", "cheap_model": "m:1"},
+    )
+    _reset_active("ollama")
+    seen = {}
+
+    async def _fake_ollama_call(messages, model, cfg):
+        seen.update(model=model, messages=messages)
+        return "odpowiedź z domu"
+
+    monkeypatch.setattr(ai_functions, "_ollama_call", _fake_ollama_call)
+    try:
+        out = asyncio.run(
+            ai_functions.provider_generate([{"role": "user", "content": "hej"}], "m:1")
+        )
+    finally:
+        _reset_active("gpt")
+    assert out == "odpowiedź z domu"
+    assert seen["model"] == "m:1"
+
+
+# ------------------------------------------------- persistence (real disk path)
+# This path had NO coverage, which is exactly how a config-clobbering regression
+# got in: persisting the whole in-memory AI_CONFIGS (built-in defaults merged
+# under the file) overwrote operator hand-edits and resurrected deleted configs.
+import json  # noqa: E402
+
+
+def _settings_file(tmp_path, configs, active="gpt"):
+    path = tmp_path / "system_gpt_settings.json"
+    path.write_text(
+        json.dumps(
+            [
+                {"role": "system", "content": "sys"},
+                {"someuser": [1, "a", "b", "c", "asst_x"]},
+                {"active": active, "configs": configs},
+            ]
+        ),
+        encoding="utf-8",
+    )
+    return path
+
+
+def test_persist_writes_the_pin_without_clobbering_operator_edits(tmp_path, monkeypatch):
+    # The file is authoritative for everything the bot does not itself change:
+    # a hand-tuned cheap_model, and a config deliberately deleted from it.
+    settings = _settings_file(
+        tmp_path,
+        {"gpt": {"provider": "openai", "latest_model": "gpt-4.1", "cheap_model": "hand-tuned"}},
+    )
+    monkeypatch.setattr(ai_functions, "SYSTEM_GPT_SETTINGS", str(settings))
+    monkeypatch.setitem(
+        ai_functions.AI_CONFIGS,
+        "ollama",
+        {"provider": "ollama", "latest_model": "mistral:7b", "cheap_model": "c:1"},
+    )
+
+    ai_functions._persist_active_ai_config("ollama", model_for="ollama")
+
+    data = json.loads(settings.read_text(encoding="utf-8"))
+    configs = data[2]["configs"]
+    assert data[2]["active"] == "ollama"
+    assert configs["ollama"]["latest_model"] == "mistral:7b"   # the pin landed
+    assert configs["gpt"]["latest_model"] == "gpt-4.1"         # edit survived
+    assert configs["gpt"]["cheap_model"] == "hand-tuned"       # edit survived
+    assert "claude" not in configs   # a deleted config is NOT resurrected
+    assert data[0]["content"] == "sys" and "someuser" in data[1]  # rest intact
+
+
+def test_plain_switch_leaves_the_configs_block_untouched(tmp_path, monkeypatch):
+    original = {"gpt": {"provider": "openai", "latest_model": "gpt-4.1", "cheap_model": "hand-tuned"}}
+    settings = _settings_file(tmp_path, original, active="claude")
+    monkeypatch.setattr(ai_functions, "SYSTEM_GPT_SETTINGS", str(settings))
+
+    # Switching backend without pinning a model must only move "active".
+    ai_functions._persist_active_ai_config("gpt")
+
+    data = json.loads(settings.read_text(encoding="utf-8"))
+    assert data[2]["active"] == "gpt"
+    assert data[2]["configs"] == original
+
+
+def test_pinned_model_survives_a_restart(tmp_path, monkeypatch):
+    # The whole point of persisting: re-reading the file must yield the pin.
+    settings = _settings_file(tmp_path, {"ollama": {"provider": "ollama", "latest_model": "old:1"}})
+    monkeypatch.setattr(ai_functions, "SYSTEM_GPT_SETTINGS", str(settings))
+    monkeypatch.setitem(
+        ai_functions.AI_CONFIGS,
+        "ollama",
+        {"provider": "ollama", "latest_model": "new:2", "cheap_model": "c:1"},
+    )
+
+    ai_functions._persist_active_ai_config("ollama", model_for="ollama")
+
+    reread = json.loads(settings.read_text(encoding="utf-8"))[2]
+    assert reread["configs"]["ollama"]["latest_model"] == "new:2"
+
+
+def test_persist_survives_an_unreadable_settings_file(tmp_path, monkeypatch):
+    # Best-effort by contract: a broken file must not raise into the command.
+    broken = tmp_path / "broken.json"
+    broken.write_text("{ not json", encoding="utf-8")
+    monkeypatch.setattr(ai_functions, "SYSTEM_GPT_SETTINGS", str(broken))
+    ai_functions._persist_active_ai_config("gpt", model_for="gpt")  # must not raise
+
+
+# ----------------------------------------------- keep-warm (Ollama ONLY) ----
+# The money guard: preloading a self-hosted model is free, but firing the same
+# thing at a metered API would burn tokens for nothing. These pin that it can
+# only ever happen for Ollama.
+
+
+def test_warm_active_model_is_a_noop_for_paid_providers(monkeypatch):
+    called = []
+    monkeypatch.setattr(
+        ai_functions, "_ollama_preload", lambda *a, **k: called.append(a) or True
+    )
+    for paid in ("gpt", "claude"):
+        _reset_active(paid)
+        try:
+            assert asyncio.run(ai_functions.warm_active_model()) is False
+        finally:
+            _reset_active("gpt")
+    assert called == [], "a paid backend must never be preloaded"
+
+
+def test_warm_active_model_preloads_when_ollama_is_active(monkeypatch):
+    monkeypatch.setitem(
+        ai_functions.AI_CONFIGS,
+        "ollama",
+        {"provider": "ollama", "latest_model": "qwen2.5:7b", "cheap_model": "c"},
+    )
+    seen = {}
+    monkeypatch.setattr(
+        ai_functions, "_ollama_preload", lambda model, *a, **k: seen.update(model=model) or True
+    )
+    _reset_active("ollama")
+    try:
+        assert asyncio.run(ai_functions.warm_active_model()) is True
+    finally:
+        _reset_active("gpt")
+    assert seen["model"] == "qwen2.5:7b"
+
+
+def test_active_provider_reports_the_switch():
+    _reset_active("gpt")
+    assert ai_functions.active_provider() == "openai"
+    _reset_active("claude")
+    try:
+        assert ai_functions.active_provider() == "anthropic"
+    finally:
+        _reset_active("gpt")
+
+
+def test_preload_sends_no_prompt_so_it_generates_nothing(monkeypatch):
+    # Ollama's documented preload: a model and keep_alive, and NO prompt. If a
+    # prompt ever crept in, every warm-up would silently generate tokens.
+    sent = {}
+
+    class _Resp:
+        status_code = 200
+
+    monkeypatch.setattr(ai_functions, "OLLAMA_URL", "http://ollama:11434")
+    monkeypatch.setattr(
+        ai_functions.requests, "post",
+        lambda url, json=None, timeout=None: sent.update(url=url, body=json) or _Resp(),
+    )
+    assert ai_functions._ollama_preload("qwen2.5:7b") is True
+    assert sent["url"].endswith("/api/generate")
+    assert sent["body"]["model"] == "qwen2.5:7b"
+    assert "keep_alive" in sent["body"]
+    assert "prompt" not in sent["body"], "a preload must not generate"
+
+
+def test_preload_without_endpoint_is_a_noop(monkeypatch):
+    monkeypatch.setattr(ai_functions, "OLLAMA_URL", "")
+    assert ai_functions._ollama_preload("x") is False
+
+
+# ------------------------------------------- personal assistants (per user) --
+# Replaces the sunset OpenAI Assistants API. The two properties that matter:
+# each user's DM history is ISOLATED (private DMs must not leak into another
+# user's context or the bar's shared memory), and it stays BOUNDED.
+
+
+def _fresh_assistant_memory(tmp_path, monkeypatch, turns=40):
+    monkeypatch.setattr(
+        ai_functions, "ASSISTANT_MEMORY_FILE", str(tmp_path / "assistant_memory.json")
+    )
+    monkeypatch.setattr(ai_functions, "ASSISTANT_MEMORY_TURNS", turns)
+    monkeypatch.setattr(ai_functions, "_ASSISTANT_MEMORY", None)
+
+
+def test_assistant_history_is_isolated_per_user(tmp_path, monkeypatch):
+    _fresh_assistant_memory(tmp_path, monkeypatch)
+    ai_functions.remember_assistant_turn(111, "sekret Anny", "ok Anna")
+    ai_functions.remember_assistant_turn(222, "sekret Bartka", "ok Bartek")
+
+    anna = ai_functions.assistant_history(111)
+    bartek = ai_functions.assistant_history(222)
+    assert [m["content"] for m in anna] == ["sekret Anny", "ok Anna"]
+    assert [m["content"] for m in bartek] == ["sekret Bartka", "ok Bartek"]
+    assert "sekret Anny" not in str(bartek)   # no cross-user bleed
+
+
+def test_assistant_history_is_trimmed_to_the_bound(tmp_path, monkeypatch):
+    _fresh_assistant_memory(tmp_path, monkeypatch, turns=4)
+    for i in range(10):
+        ai_functions.remember_assistant_turn(1, f"u{i}", f"a{i}")
+    history = ai_functions.assistant_history(1)
+    assert len(history) == 4                          # bounded
+    assert history[-1]["content"] == "a9"             # newest kept
+    assert all("u0" != m["content"] for m in history)  # oldest dropped
+
+
+def test_assistant_history_survives_a_restart(tmp_path, monkeypatch):
+    _fresh_assistant_memory(tmp_path, monkeypatch)
+    ai_functions.remember_assistant_turn(7, "pamietaj", "pamietam")
+    # Simulate a restart: drop the in-memory cache, re-read from disk.
+    monkeypatch.setattr(ai_functions, "_ASSISTANT_MEMORY", None)
+    assert [m["content"] for m in ai_functions.assistant_history(7)] == [
+        "pamietaj",
+        "pamietam",
+    ]
+
+
+def test_assistant_messages_carry_persona_history_and_new_turn(tmp_path, monkeypatch):
+    _fresh_assistant_memory(tmp_path, monkeypatch)
+    ai_functions.remember_assistant_turn(5, "wczoraj", "odpowiedz")
+    msgs = ai_functions.build_assistant_messages(
+        5, "Towarzysz Młotek", "Mówisz po polsku.", "dzisiaj"
+    )
+    assert msgs[0]["role"] == "system"
+    assert "Towarzysz Młotek" in msgs[0]["content"]
+    assert "Mówisz po polsku." in msgs[0]["content"]
+    assert [m["content"] for m in msgs[1:]] == ["wczoraj", "odpowiedz", "dzisiaj"]
+
+
+def test_corrupt_assistant_memory_starts_empty_instead_of_crashing(tmp_path, monkeypatch):
+    path = tmp_path / "assistant_memory.json"
+    path.write_text("{ not json", encoding="utf-8")
+    monkeypatch.setattr(ai_functions, "ASSISTANT_MEMORY_FILE", str(path))
+    monkeypatch.setattr(ai_functions, "_ASSISTANT_MEMORY", None)
+    assert ai_functions.assistant_history(1) == []