feat, added ollama-pull

.env.example

@@ -30,7 +30,7 @@ OPENROUTER_BASE_URL=https://openrouter.ai/api/v1
 
 # Ollama (local; no API key required — run: ollama serve && ollama pull <model>)
 OLLAMA_BASE_URL=http://localhost:11434/v1
-OLLAMA_MODEL=llama3.2
+OLLAMA_MODEL=qwen3-coder-next:cloud
 
 # ------------------------------------------------------------------------------
 # Slack API Credentials  (https://api.slack.com/apps)

agent/config.py

@@ -84,7 +84,7 @@ class Config:
     OPENROUTER_MODEL: str = os.getenv("OPENROUTER_MODEL", "anthropic/claude-sonnet-4-5")
     OPENROUTER_BASE_URL: str = os.getenv("OPENROUTER_BASE_URL", "https://openrouter.ai/api/v1")
     OLLAMA_BASE_URL: str = os.getenv("OLLAMA_BASE_URL", "http://localhost:11434/v1")
-    OLLAMA_MODEL: str = os.getenv("OLLAMA_MODEL", "llama3.2")
+    OLLAMA_MODEL: str = os.getenv("OLLAMA_MODEL", "lfm2.5-thinking:latest")
 
     # Slack
     SLACK_BOT_TOKEN: str = os.getenv("SLACK_BOT_TOKEN", "")
@@ -123,7 +123,7 @@ class Config:
     TERMINAL_TOOL_CONFIG: Path = _CONFIG_DIR / "terminal_allowed_commands.yaml"
 
     # LLM request timeout
-    LLM_REQUEST_TIMEOUT_SECONDS: int = int(os.getenv("LLM_REQUEST_TIMEOUT_SECONDS", "90"))
+    LLM_REQUEST_TIMEOUT_SECONDS: int = int(os.getenv("LLM_REQUEST_TIMEOUT_SECONDS", "200"))
 
     # Deep Research Agent
     MAX_RESEARCH_ITERATIONS: int = int(os.getenv("MAX_RESEARCH_ITERATIONS", "20"))

agent/nodes/deep_researcher.py

@@ -41,7 +41,8 @@
 # Inject a SYSTEM NOTE into the next iteration when the same pattern is seen this many times.
 _SAME_PATTERN_WARN_THRESHOLD = 2
 # Force actions=[] (immediate hand-off) when the same pattern is seen this many times.
-_SAME_PATTERN_FORCE_PIVOT = 3
+# Lowered from 3 to 2 to catch loops earlier and avoid wasted iterations.
+_SAME_PATTERN_FORCE_PIVOT = 2
 # Inject a SYSTEM NOTE after this many consecutive iterations that added zero new files.
 _ZERO_NEW_FILES_INJECT_THRESHOLD = 3
 # Force exit (hand-off) after this many consecutive zero-gain iterations.
@@ -381,9 +382,16 @@ def _normalize_pattern(action: Dict[str, Any]) -> str:
     Produce a normalized fingerprint for a search action so we can detect
     when the planner is repeating the same query.  Lowercases and strips
     leading/trailing whitespace and quote characters.
+
+    Includes tool + path in the key to distinguish searches with same pattern
+    but different scopes (e.g., search_code vs find_definitions, or path="all" vs specific repo).
     """
-    raw = action.get("pattern") or action.get("symbol") or ""
-    return raw.lower().strip().strip("\"'")
+    tool = action.get("tool", "")
+    raw = action.get("pattern") or action.get("symbol", "")
+    path = action.get("path", "all")
+    # Normalize: tool|path|pattern for more accurate deduplication
+    normalized = f"{tool}|{path}|{raw}".lower().strip().strip("\"'")
+    return normalized
 
 
 # ---------------------------------------------------------------------------
@@ -538,6 +546,7 @@ def __call__(self, state: ConversationState) -> ConversationState:
                         if key:
                             pattern_seen_counts[key] = pattern_seen_counts.get(key, 0) + 1
                             count = pattern_seen_counts[key]
+                            print(f"[DeepResearcher]   🔍 Pattern tracker: {key[:70]!r} seen {count}x")
                             if count >= _SAME_PATTERN_FORCE_PIVOT:
                                 print(f"[DeepResearcher]   ⚠ Force-pivot: pattern seen {count}x — {key[:80]!r}")
                                 log.info("[DeepResearcher] force-pivot pattern=%r count=%d", key[:80], count)
@@ -746,7 +755,9 @@ def _execute_search_action(action: Dict[str, Any], _iter: int = iteration) -> tu
                 history_entry = (
                     f"Iteration {iteration}\n"
                     f"  Intent: {search_intent}\n"
-                    f"  Thinking: {(thinking or '').strip()[:220]}\n\n"
+                    f"  Thinking: {(thinking or '').strip()[:300]}\n"
+                    f"  Actions: {len(actions)} tool calls\n"
+                    f"  Results: {added_this_iter} new files found\n"
                     + "\n\n".join(action_blocks)
                 )
                 search_history_entries.append(history_entry)

docker-compose.yml

@@ -8,7 +8,7 @@
 #   docker compose -f docker-compose.yml -f docker-compose.build.yml up --build
 #
 # Ollama runs by default (service `ollama`). Set LLM_PROVIDER=ollama and pull a model, e.g.:
-#   docker compose exec ollama ollama pull llama3.2
+#   docker compose exec ollama ollama pull lfm2.5-thinking:latest
 #
 # Configuration lives in this file (no separate .env required).
 #
@@ -25,8 +25,8 @@ x-postgres-password: &postgres-password "${POSTGRES_PASSWORD:-slack_agent_dev}"
 # Shared application environment for the agent service.
 x-app-environment: &app-environment
   # --- Identity (optional; config/agent.yaml is the default source) ---
-  AGENT_NAME: ""
-  COMPANY_NAME: ""
+  AGENT_NAME: "Alex"
+  COMPANY_NAME: "OLake"
 
   # --- LLM Provider to use---
   LLM_PROVIDER: ollama
@@ -46,9 +46,9 @@ x-app-environment: &app-environment
   OPENROUTER_MODEL: anthropic/claude-sonnet-4-5
   OPENROUTER_BASE_URL: ${OPENROUTER_BASE_URL}
 
-  # --- Ollama ---
+  # --- Ollama (agent reaches the ollama service on the compose network) ---
   OLLAMA_BASE_URL: http://ollama:11434/v1
-  OLLAMA_MODEL: ${OLLAMA_MODEL}
+  OLLAMA_MODEL: ${OLLAMA_MODEL:-lfm2.5-thinking:latest}
 
   # --- Slack (required for the agent service) ---
   SLACK_BOT_TOKEN: ${SLACK_BOT_TOKEN}
@@ -71,7 +71,7 @@ x-app-environment: &app-environment
   HIGH_PRIORITY_CHANNELS: ""
 
   # --- Tuning ---
-  LLM_REQUEST_TIMEOUT_SECONDS: "90"
+  LLM_REQUEST_TIMEOUT_SECONDS: "300"
   MAX_RESEARCH_ITERATIONS: "20"
   MAX_CONTEXT_FILES: "15"
   RESEARCH_TIMEOUT_SECONDS: "120"
@@ -101,9 +101,9 @@ services:
       test: ["CMD-SHELL", "pg_isready -U slack_agent -d slack_agent"]
       interval: 5s
       timeout: 5s
-      retries: 10
+      retries: 3
 
-  # Local LLM — runs by default; use with LLM_PROVIDER=ollama (pull models into the volume first).
+  # Local LLM — image default CMD is `ollama serve` (listens on :11434). Data in ollama_data volume.
   ollama:
     image: ollama/ollama:latest
     restart: unless-stopped
@@ -115,9 +115,22 @@ services:
       test: ["CMD-SHELL", "ollama list >/dev/null 2>&1 || exit 1"]
       interval: 15s
       timeout: 5s
-      retries: 5
+      retries: 3
       start_period: 15s
 
+  # One-shot: pull OLLAMA_MODEL into the shared volume after the server is ready (first run may take a while).
+  ollama-pull:
+    image: ollama/ollama:latest
+    restart: "no"
+    volumes:
+      - ollama_data:/root/.ollama
+    environment:
+      OLLAMA_HOST: http://ollama:11434
+    entrypoint: ["ollama", "pull", "${OLLAMA_MODEL:-lfm2.5-thinking:latest}"]
+    depends_on:
+      ollama:
+        condition: service_healthy
+
   agent:
     image: siddharthchevella/slack-agent:latest
     restart: unless-stopped
@@ -133,11 +146,13 @@ services:
         condition: service_healthy
       ollama:
         condition: service_healthy
+      ollama-pull:
+        condition: service_completed_successfully
     healthcheck:
       test: ["CMD-SHELL", "python -c \"import urllib.request; urllib.request.urlopen('http://localhost:8001/health')\""]
       interval: 15s
       timeout: 5s
-      retries: 5
+      retries: 3
       start_period: 20s
 
 volumes: