diff --git a/docs/chinese-llm-setup.md b/docs/chinese-llm-setup.md
index 6638dbba1..ac03a7902 100644
--- a/docs/chinese-llm-setup.md
+++ b/docs/chinese-llm-setup.md
@@ -212,9 +212,9 @@ API Base URL: https://open.bigmodel.cn/api/paas/v4
 
 | 字段 | 值 | 说明 |
 |------|-----|------|
-| **Configuration Name** | `MiniMax M2.5` | 配置名称（自定义） |
+| **Configuration Name** | `MiniMax M3` | 配置名称（自定义） |
 | **Provider** | `MINIMAX` | 选择 MiniMax |
-| **Model Name** | `MiniMax-M2.5` | 推荐模型<br>其他选项: `MiniMax-M2.5-highspeed` |
+| **Model Name** | `MiniMax-M3` | 推荐模型<br>其他选项: `MiniMax-M2.7`、`MiniMax-M2.7-highspeed` |
 | **API Key** | `eyJ...` | 你的 MiniMax API Key |
 | **API Base URL** | `https://api.minimax.io/v1` | MiniMax API 地址 |
 | **Parameters** | `{"temperature": 1.0}` | 注意：temperature 必须在 (0.0, 1.0] 范围内，不能为 0 |
@@ -222,22 +222,23 @@ API Base URL: https://open.bigmodel.cn/api/paas/v4
 ### 示例配置
 
 ```
-Configuration Name: MiniMax M2.5
+Configuration Name: MiniMax M3
 Provider: MINIMAX
-Model Name: MiniMax-M2.5
+Model Name: MiniMax-M3
 API Key: eyJxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
 API Base URL: https://api.minimax.io/v1
 ```
 
 ### 可用模型
 
-- **MiniMax-M2.5**: 高性能通用模型，204K 上下文窗口（推荐）
-- **MiniMax-M2.5-highspeed**: 高速推理版本，204K 上下文窗口
+- **MiniMax-M3**: 旗舰模型，512K 上下文窗口（推荐）
+- **MiniMax-M2.7**: 上一代通用模型，204K 上下文窗口
+- **MiniMax-M2.7-highspeed**: 上一代高速推理版本，204K 上下文窗口
 
 ### 注意事项
 
 - **temperature 参数**: MiniMax 要求 temperature 必须在 (0.0, 1.0] 范围内，不能设置为 0。建议使用 1.0。
-- 两个模型都支持 204K 超长上下文窗口，适合处理长文本任务。
+- M3 支持 512K 超长上下文，M2.7 系列保留 204K，适合按需求选择。
 
 ### 定价
 - 请访问 [MiniMax 定价页面](https://platform.minimaxi.com/document/Price) 查看最新价格
@@ -315,8 +316,8 @@ docker compose logs backend | grep -i "error"
 |---------|---------|------|
 | **文档摘要** | Qwen-Plus, GLM-4 | 平衡性能和成本 |
 | **代码分析** | DeepSeek-Coder | 代码专用 |
-| **长文本处理** | Kimi 128K, MiniMax-M2.5 (204K) | 超长上下文 |
-| **快速响应** | Qwen-Turbo, GLM-4-Flash, MiniMax-M2.5-highspeed | 速度优先 |
+| **长文本处理** | Kimi 128K, MiniMax-M3 (512K) | 超长上下文 |
+| **快速响应** | Qwen-Turbo, GLM-4-Flash, MiniMax-M2.7-highspeed | 速度优先 |
 
 ### 2. 成本优化
 
diff --git a/surfsense_backend/alembic/versions/144_add_automation_tables.py b/surfsense_backend/alembic/versions/144_add_automation_tables.py
index 39f927417..296c33585 100644
--- a/surfsense_backend/alembic/versions/144_add_automation_tables.py
+++ b/surfsense_backend/alembic/versions/144_add_automation_tables.py
@@ -25,34 +25,60 @@ depends_on: str | Sequence[str] | None = None
 
 
 def upgrade() -> None:
-    # ENUM types (PostgreSQL requires types created before tables that use them)
+    # Guard every object so the migration is safe to re-run after a partial
+    # apply (the types/tables outlive a failed run that never advanced
+    # alembic_version). Types must precede the tables that reference them.
     op.execute(
         """
-        CREATE TYPE automation_status AS ENUM (
-            'active', 'paused', 'archived'
-        );
+        DO $$
+        BEGIN
+            IF NOT EXISTS (
+                SELECT 1 FROM pg_type WHERE typname = 'automation_status'
+            ) THEN
+                CREATE TYPE automation_status AS ENUM (
+                    'active', 'paused', 'archived'
+                );
+            END IF;
+        END
+        $$;
         """
     )
     op.execute(
         """
-        CREATE TYPE automation_trigger_type AS ENUM (
-            'schedule', 'manual'
-        );
+        DO $$
+        BEGIN
+            IF NOT EXISTS (
+                SELECT 1 FROM pg_type WHERE typname = 'automation_trigger_type'
+            ) THEN
+                CREATE TYPE automation_trigger_type AS ENUM (
+                    'schedule', 'manual'
+                );
+            END IF;
+        END
+        $$;
         """
     )
     op.execute(
         """
-        CREATE TYPE automation_run_status AS ENUM (
-            'pending', 'running', 'succeeded', 'failed',
-            'cancelled', 'timed_out'
-        );
+        DO $$
+        BEGIN
+            IF NOT EXISTS (
+                SELECT 1 FROM pg_type WHERE typname = 'automation_run_status'
+            ) THEN
+                CREATE TYPE automation_run_status AS ENUM (
+                    'pending', 'running', 'succeeded', 'failed',
+                    'cancelled', 'timed_out'
+                );
+            END IF;
+        END
+        $$;
         """
     )
 
     # automations — the editable, versioned automation definition
     op.execute(
         """
-        CREATE TABLE automations (
+        CREATE TABLE IF NOT EXISTS automations (
             id SERIAL PRIMARY KEY,
             search_space_id INTEGER NOT NULL
                 REFERENCES searchspaces(id) ON DELETE CASCADE,
@@ -69,19 +95,25 @@ def upgrade() -> None:
         """
     )
     op.execute(
-        "CREATE INDEX ix_automations_search_space_id ON automations(search_space_id);"
+        "CREATE INDEX IF NOT EXISTS ix_automations_search_space_id ON automations(search_space_id);"
     )
     op.execute(
-        "CREATE INDEX ix_automations_created_by_user_id ON automations(created_by_user_id);"
+        "CREATE INDEX IF NOT EXISTS ix_automations_created_by_user_id ON automations(created_by_user_id);"
+    )
+    op.execute(
+        "CREATE INDEX IF NOT EXISTS ix_automations_status ON automations(status);"
+    )
+    op.execute(
+        "CREATE INDEX IF NOT EXISTS ix_automations_created_at ON automations(created_at);"
+    )
+    op.execute(
+        "CREATE INDEX IF NOT EXISTS ix_automations_updated_at ON automations(updated_at);"
     )
-    op.execute("CREATE INDEX ix_automations_status ON automations(status);")
-    op.execute("CREATE INDEX ix_automations_created_at ON automations(created_at);")
-    op.execute("CREATE INDEX ix_automations_updated_at ON automations(updated_at);")
 
     # automation_triggers — one row per (automation, trigger-instance) pair
     op.execute(
         """
-        CREATE TABLE automation_triggers (
+        CREATE TABLE IF NOT EXISTS automation_triggers (
             id SERIAL PRIMARY KEY,
             automation_id INTEGER NOT NULL
                 REFERENCES automations(id) ON DELETE CASCADE,
@@ -96,20 +128,22 @@ def upgrade() -> None:
         """
     )
     op.execute(
-        "CREATE INDEX ix_automation_triggers_automation_id ON automation_triggers(automation_id);"
-    )
-    op.execute("CREATE INDEX ix_automation_triggers_type ON automation_triggers(type);")
-    op.execute(
-        "CREATE INDEX ix_automation_triggers_enabled ON automation_triggers(enabled);"
+        "CREATE INDEX IF NOT EXISTS ix_automation_triggers_automation_id ON automation_triggers(automation_id);"
     )
     op.execute(
-        "CREATE INDEX ix_automation_triggers_created_at ON automation_triggers(created_at);"
+        "CREATE INDEX IF NOT EXISTS ix_automation_triggers_type ON automation_triggers(type);"
+    )
+    op.execute(
+        "CREATE INDEX IF NOT EXISTS ix_automation_triggers_enabled ON automation_triggers(enabled);"
+    )
+    op.execute(
+        "CREATE INDEX IF NOT EXISTS ix_automation_triggers_created_at ON automation_triggers(created_at);"
     )
     # Partial index for the schedule tick: only enabled schedule triggers
     # with a scheduled next fire are ever scanned for due rows.
     op.execute(
         """
-        CREATE INDEX ix_automation_triggers_due
+        CREATE INDEX IF NOT EXISTS ix_automation_triggers_due
             ON automation_triggers (next_fire_at)
             WHERE enabled = true
               AND type = 'schedule'
@@ -120,7 +154,7 @@ def upgrade() -> None:
     # automation_runs — the immutable per-fire execution record
     op.execute(
         """
-        CREATE TABLE automation_runs (
+        CREATE TABLE IF NOT EXISTS automation_runs (
             id SERIAL PRIMARY KEY,
             automation_id INTEGER NOT NULL
                 REFERENCES automations(id) ON DELETE CASCADE,
@@ -140,14 +174,16 @@ def upgrade() -> None:
         """
     )
     op.execute(
-        "CREATE INDEX ix_automation_runs_automation_id ON automation_runs(automation_id);"
+        "CREATE INDEX IF NOT EXISTS ix_automation_runs_automation_id ON automation_runs(automation_id);"
     )
     op.execute(
-        "CREATE INDEX ix_automation_runs_trigger_id ON automation_runs(trigger_id);"
+        "CREATE INDEX IF NOT EXISTS ix_automation_runs_trigger_id ON automation_runs(trigger_id);"
     )
-    op.execute("CREATE INDEX ix_automation_runs_status ON automation_runs(status);")
     op.execute(
-        "CREATE INDEX ix_automation_runs_created_at ON automation_runs(created_at);"
+        "CREATE INDEX IF NOT EXISTS ix_automation_runs_status ON automation_runs(status);"
+    )
+    op.execute(
+        "CREATE INDEX IF NOT EXISTS ix_automation_runs_created_at ON automation_runs(created_at);"
     )
 
 
diff --git a/surfsense_backend/alembic/versions/148_add_automation_runs_to_zero_publication.py b/surfsense_backend/alembic/versions/148_add_automation_runs_to_zero_publication.py
new file mode 100644
index 000000000..1b25be753
--- /dev/null
+++ b/surfsense_backend/alembic/versions/148_add_automation_runs_to_zero_publication.py
@@ -0,0 +1,177 @@
+"""add automation_runs to zero_publication with thin column list
+
+Publishes ``automation_runs`` so the dashboard can replace polling with a
+live run status + per-step ticker. Only the columns the list and ticker
+read are exposed (``id, automation_id, trigger_id, status, step_results,
+started_at, finished_at, created_at``); heavy JSONB
+(``definition_snapshot``, ``inputs``, ``output``, ``artifacts``, ``error``)
+stays on REST and is fetched lazily on detail expand.
+
+Uses the canonical ``ALTER PUBLICATION ... SET TABLE`` + ``COMMENT``
+bookend pattern (see migration 143) -- the shape Zero ``>=1.0`` requires
+to fire its schema-change hook. Existing tables are re-emitted unchanged.
+
+Revision ID: 148
+Revises: 147
+"""
+
+from collections.abc import Sequence
+
+import sqlalchemy as sa
+
+from alembic import op
+
+revision: str = "148"
+down_revision: str | None = "147"
+branch_labels: str | Sequence[str] | None = None
+depends_on: str | Sequence[str] | None = None
+
+PUBLICATION_NAME = "zero_publication"
+
+# Mirrors migration 143. Kept in sync explicitly: any change to these lists
+# must be re-emitted in a new resync migration with COMMENT bookends.
+DOCUMENT_COLS = [
+    "id",
+    "title",
+    "document_type",
+    "search_space_id",
+    "folder_id",
+    "created_by_id",
+    "status",
+    "created_at",
+    "updated_at",
+]
+
+USER_COLS = [
+    "id",
+    "pages_limit",
+    "pages_used",
+    "premium_credit_micros_limit",
+    "premium_credit_micros_used",
+]
+
+# Thin set: status + lightweight progress only. Heavy JSONB stays on REST.
+AUTOMATION_RUN_COLS = [
+    "id",
+    "automation_id",
+    "trigger_id",
+    "status",
+    "step_results",
+    "started_at",
+    "finished_at",
+    "created_at",
+]
+
+
+def _has_zero_version(conn, table: str) -> bool:
+    return (
+        conn.execute(
+            sa.text(
+                "SELECT 1 FROM information_schema.columns "
+                "WHERE table_name = :tbl AND column_name = '_0_version'"
+            ),
+            {"tbl": table},
+        ).fetchone()
+        is not None
+    )
+
+
+def _build_set_table_ddl(
+    *, documents_has_zero_ver: bool, user_has_zero_ver: bool
+) -> str:
+    doc_cols = DOCUMENT_COLS + (['"_0_version"'] if documents_has_zero_ver else [])
+    user_cols = USER_COLS + (['"_0_version"'] if user_has_zero_ver else [])
+    doc_col_list = ", ".join(doc_cols)
+    user_col_list = ", ".join(user_cols)
+    run_col_list = ", ".join(AUTOMATION_RUN_COLS)
+    return (
+        f"ALTER PUBLICATION {PUBLICATION_NAME} SET TABLE "
+        f"notifications, "
+        f"documents ({doc_col_list}), "
+        f"folders, "
+        f"search_source_connectors, "
+        f"new_chat_messages, "
+        f"chat_comments, "
+        f"chat_session_state, "
+        f'"user" ({user_col_list}), '
+        f"automation_runs ({run_col_list})"
+    )
+
+
+def upgrade() -> None:
+    conn = op.get_bind()
+
+    exists = conn.execute(
+        sa.text("SELECT 1 FROM pg_publication WHERE pubname = :name"),
+        {"name": PUBLICATION_NAME},
+    ).fetchone()
+    if not exists:
+        return
+
+    documents_has_zero_ver = _has_zero_version(conn, "documents")
+    user_has_zero_ver = _has_zero_version(conn, "user")
+
+    # COMMENT-ALTER-COMMENT trio must be one transaction so Zero observes
+    # them as one schema-change event. Matches the SAVEPOINT pattern used
+    # in migrations 117 / 139 / 140 / 143.
+    tx = conn.begin_nested() if conn.in_transaction() else conn.begin()
+    with tx:
+        conn.execute(
+            sa.text(f"COMMENT ON PUBLICATION {PUBLICATION_NAME} IS 'pre-148-resync'")
+        )
+        conn.execute(
+            sa.text(
+                _build_set_table_ddl(
+                    documents_has_zero_ver=documents_has_zero_ver,
+                    user_has_zero_ver=user_has_zero_ver,
+                )
+            )
+        )
+        conn.execute(
+            sa.text(f"COMMENT ON PUBLICATION {PUBLICATION_NAME} IS 'post-148-resync'")
+        )
+
+
+def downgrade() -> None:
+    """Re-emit migration 143's shape (no automation_runs)."""
+    conn = op.get_bind()
+
+    exists = conn.execute(
+        sa.text("SELECT 1 FROM pg_publication WHERE pubname = :name"),
+        {"name": PUBLICATION_NAME},
+    ).fetchone()
+    if not exists:
+        return
+
+    documents_has_zero_ver = _has_zero_version(conn, "documents")
+    user_has_zero_ver = _has_zero_version(conn, "user")
+
+    doc_cols = DOCUMENT_COLS + (['"_0_version"'] if documents_has_zero_ver else [])
+    user_cols = USER_COLS + (['"_0_version"'] if user_has_zero_ver else [])
+    doc_col_list = ", ".join(doc_cols)
+    user_col_list = ", ".join(user_cols)
+    ddl = (
+        f"ALTER PUBLICATION {PUBLICATION_NAME} SET TABLE "
+        f"notifications, "
+        f"documents ({doc_col_list}), "
+        f"folders, "
+        f"search_source_connectors, "
+        f"new_chat_messages, "
+        f"chat_comments, "
+        f"chat_session_state, "
+        f'"user" ({user_col_list})'
+    )
+
+    tx = conn.begin_nested() if conn.in_transaction() else conn.begin()
+    with tx:
+        conn.execute(
+            sa.text(
+                f"COMMENT ON PUBLICATION {PUBLICATION_NAME} IS 'pre-148-downgrade'"
+            )
+        )
+        conn.execute(sa.text(ddl))
+        conn.execute(
+            sa.text(
+                f"COMMENT ON PUBLICATION {PUBLICATION_NAME} IS 'post-148-downgrade'"
+            )
+        )
diff --git a/surfsense_backend/app/automations/actions/builtin/agent_task/dependencies.py b/surfsense_backend/app/automations/actions/builtin/agent_task/dependencies.py
index e3736cc95..4ef8c52bf 100644
--- a/surfsense_backend/app/automations/actions/builtin/agent_task/dependencies.py
+++ b/surfsense_backend/app/automations/actions/builtin/agent_task/dependencies.py
@@ -85,23 +85,16 @@ async def build_dependencies(
     connector_service, firecrawl_api_key = await setup_connector_and_firecrawl(
         session, search_space_id=search_space_id
     )
-    # Quick fix: use an in-memory checkpointer for automation runs.
+    # Per-task InMemorySaver: the shared Postgres checkpointer's connection
+    # pool binds connections to the loop that opened them, but Celery uses a
+    # fresh loop per task, so the next task hangs 30s on a dead-loop connection
+    # (`PoolTimeout`). InMemorySaver has no pool and dies with the task — fine
+    # while runs are one-shot (the checkpoint only spans one graph execution).
     #
-    # The shared Postgres checkpointer caches DB connections in a
-    # module-level pool. Each cached connection is bound to the asyncio
-    # loop that opened it. Celery throws away the loop after every task,
-    # so the pool ends up full of connections pointing to a dead loop,
-    # and the next Celery task (running on a fresh loop) can't use any
-    # of them — it hangs 30s and fails with
-    # `PoolTimeout: couldn't get a connection after 30.00 sec`.
-    #
-    # InMemorySaver has no cached connections, no loop binding — each
-    # Celery task creates one and drops it on exit.
-    #
-    # TODO(checkpointer): proper fix is to dispose the checkpointer
-    # pool around each Celery task in `run_async_celery_task`, the same
-    # way `_dispose_shared_db_engine` already does for the SQLAlchemy
-    # pool. Then this site can switch back to the shared checkpointer.
+    # TODO(checkpointer): when runs need durability (crash-resume or HITL
+    # interrupt/resume across tasks), dispose the checkpointer pool around each
+    # Celery task in `run_async_celery_task` — as `_dispose_shared_db_engine`
+    # already does for the SQLAlchemy pool — then use the shared checkpointer.
     checkpointer = InMemorySaver()
     return AgentDependencies(
         llm=llm,
diff --git a/surfsense_backend/app/config/global_llm_config.example.yaml b/surfsense_backend/app/config/global_llm_config.example.yaml
index 83d556754..1c09a91ac 100644
--- a/surfsense_backend/app/config/global_llm_config.example.yaml
+++ b/surfsense_backend/app/config/global_llm_config.example.yaml
@@ -236,17 +236,17 @@ global_llm_configs:
     use_default_system_instructions: true
     citations_enabled: true
 
-  # Example: MiniMax M2.5 - High-performance with 204K context window
+  # Example: MiniMax M3 - High-performance with 512K context window
   - id: -8
-    name: "Global MiniMax M2.5"
-    description: "MiniMax M2.5 with 204K context window and competitive pricing"
+    name: "Global MiniMax M3"
+    description: "MiniMax M3 with 512K context window and competitive pricing"
     billing_tier: "free"
     anonymous_enabled: true
     seo_enabled: true
-    seo_slug: "minimax-m2.5"
+    seo_slug: "minimax-m3"
     quota_reserve_tokens: 4000
     provider: "MINIMAX"
-    model_name: "MiniMax-M2.5"
+    model_name: "MiniMax-M3"
     api_key: "your-minimax-api-key-here"
     api_base: "https://api.minimax.io/v1"
     rpm: 60
diff --git a/surfsense_web/app/dashboard/[search_space_id]/automations/[automation_id]/components/run-details-panel.tsx b/surfsense_web/app/dashboard/[search_space_id]/automations/[automation_id]/components/run-details-panel.tsx
index 164f156e5..1a54ac0e5 100644
--- a/surfsense_web/app/dashboard/[search_space_id]/automations/[automation_id]/components/run-details-panel.tsx
+++ b/surfsense_web/app/dashboard/[search_space_id]/automations/[automation_id]/components/run-details-panel.tsx
@@ -15,7 +15,7 @@ import { Collapsible, CollapsibleContent, CollapsibleTrigger } from "@/component
 import { ScrollArea } from "@/components/ui/scroll-area";
 import { Separator } from "@/components/ui/separator";
 import { Skeleton } from "@/components/ui/skeleton";
-import type { RunStepResult } from "@/contracts/types/automation.types";
+import type { RunStatus, RunStepResult } from "@/contracts/types/automation.types";
 import { useAutomationRun } from "@/hooks/use-automation-runs";
 import { cn } from "@/lib/utils";
 import { RunStepResultCard } from "./run-step-result-card";
@@ -23,44 +23,46 @@ import { RunStepResultCard } from "./run-step-result-card";
 interface RunDetailsPanelProps {
 	automationId: number;
 	runId: number;
+	/** Live step entries from Zero; rendered while the run is in-flight and
+	 * also kept as the authoritative source once it finishes. */
+	liveSteps: RunStepResult[];
+	/** Live run status from Zero. Used to hide diagnostic sections that
+	 * only make sense after the run reaches a terminal state. */
+	liveStatus: RunStatus;
 }
 
 /**
- * Expanded view of a single run. Fetches lazily — the parent only renders
- * this once the row is opened, so the list view stays cheap.
+ * Expanded view of a single run. Steps render immediately from the live
+ * Zero row so the panel updates as the run progresses; the heavy REST
+ * payload (output, artifacts, resolved inputs, run-level error) is
+ * fetched lazily and merged in when it arrives.
  *
- * We surface the run outcome readably: a run-level error first (when
- * present), then per-step cards that render the agent's markdown
- * ``final_message`` directly, and finally the structural artifacts/inputs.
- * The full ``definition_snapshot`` is omitted because it usually mirrors the
- * live definition — surfacing it would dominate the panel without informing
+ * Surfacing order is outcome-first: a run-level error (when present),
+ * then per-step cards that render the agent's markdown ``final_message``
+ * directly, and finally the structural artifacts/inputs. The full
+ * ``definition_snapshot`` is omitted because it usually mirrors the live
+ * definition — surfacing it would dominate the panel without informing
  * what the user is trying to learn ("did this work? what did it do?").
  */
-export function RunDetailsPanel({ automationId, runId }: RunDetailsPanelProps) {
-	const { data: run, isLoading, error } = useAutomationRun(automationId, runId);
+export function RunDetailsPanel({
+	automationId,
+	runId,
+	liveSteps,
+	liveStatus,
+}: RunDetailsPanelProps) {
+	const isTerminal = liveStatus !== "pending" && liveStatus !== "running";
+	// Defer the REST round-trip until the run can actually carry heavy
+	// fields — output/artifacts/error are only written at terminal mark.
+	const { data: run, isLoading, error } = useAutomationRun(automationId, runId, {
+		enabled: isTerminal,
+	});
 
-	if (isLoading) {
-		return (
-			<div className="flex flex-col gap-3 border-t border-border/60 bg-muted/20 p-4">
-				<Skeleton className="h-3 w-32" />
-				<Skeleton className="h-24 w-full" />
-			</div>
-		);
-	}
-
-	if (error || !run) {
-		return (
-			<div className="border-t border-border/60 bg-muted/20 p-4 text-xs text-muted-foreground">
-				Couldn't load run details{error?.message ? `: ${error.message}` : "."}
-			</div>
-		);
-	}
-
-	const runError = run.error && Object.keys(run.error).length > 0 ? run.error : null;
-	const hasOutput = run.output && Object.keys(run.output).length > 0;
-	const hasInputs = Object.keys(run.inputs ?? {}).length > 0;
-	const steps = run.step_results as RunStepResult[];
-	const hasDiagnostics = run.artifacts.length > 0 || hasInputs;
+	const runError = run?.error && Object.keys(run.error).length > 0 ? run.error : null;
+	const hasOutput = !!run?.output && Object.keys(run.output).length > 0;
+	const hasInputs = !!run && Object.keys(run.inputs ?? {}).length > 0;
+	const hasDiagnostics = !!run && (run.artifacts.length > 0 || hasInputs);
+	const heavyLoading = isTerminal && isLoading && !run;
+	const heavyError = isTerminal && !!error;
 
 	return (
 		<div className="flex flex-col gap-4 border-t border-border/60 bg-muted/20 p-4">
@@ -72,30 +74,40 @@ export function RunDetailsPanel({ automationId, runId }: RunDetailsPanelProps) {
 				</Section>
 			) : null}
 
-			<Section icon={GitCommitHorizontal} label={`Step results · ${steps.length}`}>
-				{steps.length === 0 ? (
-					<p className="text-xs text-muted-foreground">No steps recorded.</p>
+			<Section icon={GitCommitHorizontal} label={`Step results · ${liveSteps.length}`}>
+				{liveSteps.length === 0 ? (
+					<p className="text-xs text-muted-foreground">
+						{isTerminal ? "No steps recorded." : "Waiting for first step…"}
+					</p>
 				) : (
 					<div className="flex flex-col gap-2">
-						{steps.map((step, index) => (
+						{liveSteps.map((step, index) => (
 							<RunStepResultCard key={step.step_id ?? index} step={step} />
 						))}
 					</div>
 				)}
 			</Section>
 
-			{hasDiagnostics ? <Separator className="bg-border/60" /> : null}
-
-			{run.artifacts.length > 0 ? (
-				<Section icon={Package} label={`Artifacts · ${run.artifacts.length}`}>
-					<JsonBlock value={run.artifacts} />
-				</Section>
-			) : null}
-
-			{hasInputs ? (
-				<Section icon={Settings2} label="Resolved inputs">
-					<JsonBlock value={run.inputs} />
-				</Section>
+			{heavyLoading ? (
+				<Skeleton className="h-16 w-full" />
+			) : heavyError ? (
+				<p className="text-xs text-muted-foreground">
+					Couldn't load run details{error?.message ? `: ${error.message}` : "."}
+				</p>
+			) : hasDiagnostics ? (
+				<>
+					<Separator className="bg-border/60" />
+					{run && run.artifacts.length > 0 ? (
+						<Section icon={Package} label={`Artifacts · ${run.artifacts.length}`}>
+							<JsonBlock value={run.artifacts} />
+						</Section>
+					) : null}
+					{hasInputs ? (
+						<Section icon={Settings2} label="Resolved inputs">
+							<JsonBlock value={run?.inputs} />
+						</Section>
+					) : null}
+				</>
 			) : null}
 		</div>
 	);
diff --git a/surfsense_web/app/dashboard/[search_space_id]/automations/[automation_id]/components/run-row.tsx b/surfsense_web/app/dashboard/[search_space_id]/automations/[automation_id]/components/run-row.tsx
index 3f6a39c35..b48230e3f 100644
--- a/surfsense_web/app/dashboard/[search_space_id]/automations/[automation_id]/components/run-row.tsx
+++ b/surfsense_web/app/dashboard/[search_space_id]/automations/[automation_id]/components/run-row.tsx
@@ -1,21 +1,21 @@
 "use client";
 import { ChevronDown, ChevronRight, Hand } from "lucide-react";
 import { useState } from "react";
-import type { RunSummary } from "@/contracts/types/automation.types";
+import type { LiveRunSummary } from "@/hooks/use-automation-runs";
 import { formatDuration } from "@/lib/automations/run-duration";
 import { formatRelativeDate } from "@/lib/format-date";
 import { RunDetailsPanel } from "./run-details-panel";
 import { RunStatusBadge } from "./run-status-badge";
 
 interface RunRowProps {
-	run: RunSummary;
+	run: LiveRunSummary;
 	automationId: number;
 }
 
 /**
- * One run row. Click to expand → fetches the full run and shows the
- * details panel inline. State is local to each row so multiple panels
- * can be open at once (or none).
+ * One run row. Click to expand → renders the details panel inline.
+ * Status and step_results come live from the parent's Zero query; the
+ * panel itself only fetches the heavy REST fields on first expand.
  */
 export function RunRow({ run, automationId }: RunRowProps) {
 	const [open, setOpen] = useState(false);
@@ -47,7 +47,14 @@ export function RunRow({ run, automationId }: RunRowProps) {
 				</div>
 			</button>
 
-			{open && <RunDetailsPanel automationId={automationId} runId={run.id} />}
+			{open && (
+				<RunDetailsPanel
+					automationId={automationId}
+					runId={run.id}
+					liveSteps={run.step_results}
+					liveStatus={run.status}
+				/>
+			)}
 		</div>
 	);
 }
diff --git a/surfsense_web/content/docs/how-to/zero-sync.mdx b/surfsense_web/content/docs/how-to/zero-sync.mdx
index 1a7762f23..7007e6637 100644
--- a/surfsense_web/content/docs/how-to/zero-sync.mdx
+++ b/surfsense_web/content/docs/how-to/zero-sync.mdx
@@ -104,6 +104,7 @@ Zero syncs the following tables for real-time features:
 | `new_chat_messages` | Live chat message sync for shared chats |
 | `chat_comments` | Real-time comment threads on AI responses |
 | `chat_session_state` | Collaboration indicators (who is typing) |
+| `automation_runs` | Live run status and per-step progress (thin column set; heavy fields stay on REST) |
 
 ## Troubleshooting
 
diff --git a/surfsense_web/contracts/enums/llm-models.ts b/surfsense_web/contracts/enums/llm-models.ts
index 74cc056f3..9647c9d31 100644
--- a/surfsense_web/contracts/enums/llm-models.ts
+++ b/surfsense_web/contracts/enums/llm-models.ts
@@ -1528,14 +1528,20 @@ export const LLM_MODELS: LLMModel[] = [
 
 	// MiniMax
 	{
-		value: "MiniMax-M2.5",
-		label: "MiniMax M2.5",
+		value: "MiniMax-M3",
+		label: "MiniMax M3",
+		provider: "MINIMAX",
+		contextWindow: "512K",
+	},
+	{
+		value: "MiniMax-M2.7",
+		label: "MiniMax M2.7",
 		provider: "MINIMAX",
 		contextWindow: "204K",
 	},
 	{
-		value: "MiniMax-M2.5-highspeed",
-		label: "MiniMax M2.5 Highspeed",
+		value: "MiniMax-M2.7-highspeed",
+		label: "MiniMax M2.7 Highspeed",
 		provider: "MINIMAX",
 		contextWindow: "204K",
 	},
diff --git a/surfsense_web/contracts/enums/llm-providers.ts b/surfsense_web/contracts/enums/llm-providers.ts
index ce2b6afe9..c04a44923 100644
--- a/surfsense_web/contracts/enums/llm-providers.ts
+++ b/surfsense_web/contracts/enums/llm-providers.ts
@@ -184,8 +184,8 @@ export const LLM_PROVIDERS: LLMProvider[] = [
 	{
 		value: "MINIMAX",
 		label: "MiniMax",
-		example: "MiniMax-M2.5, MiniMax-M2.5-highspeed",
-		description: "High-performance models with 204K context",
+		example: "MiniMax-M3, MiniMax-M2.7",
+		description: "High-performance models with up to 512K context",
 		apiBase: "https://api.minimax.io/v1",
 	},
 	{
diff --git a/surfsense_web/hooks/use-automation-runs.ts b/surfsense_web/hooks/use-automation-runs.ts
index c91c7bd6e..69e51ddc6 100644
--- a/surfsense_web/hooks/use-automation-runs.ts
+++ b/surfsense_web/hooks/use-automation-runs.ts
@@ -1,42 +1,109 @@
 "use client";
-import { useQuery } from "@tanstack/react-query";
-import type { Run, RunListResponse } from "@/contracts/types/automation.types";
+import { useQuery as useZeroQuery } from "@rocicorp/zero/react";
+import { useQuery as useReactQuery } from "@tanstack/react-query";
+import { useMemo } from "react";
+import type { Run, RunStepResult, RunSummary } from "@/contracts/types/automation.types";
 import { automationsApiService } from "@/lib/apis/automations-api.service";
 import { cacheKeys } from "@/lib/query-client/cache-keys";
+import { queries } from "@/zero/queries";
 
 const DEFAULT_LIMIT = 50;
-const DEFAULT_OFFSET = 0;
+
+/**
+ * Thin live row sourced from Zero. Strict superset of {@link RunSummary} —
+ * existing consumers that only look at the summary fields keep working,
+ * while the run detail panel can read ``step_results`` directly for the
+ * live step ticker without a second REST round-trip.
+ */
+export interface LiveRunSummary extends RunSummary {
+	step_results: RunStepResult[];
+}
 
 export interface UseAutomationRunsOptions {
 	limit?: number;
-	offset?: number;
-	enabled?: boolean;
 }
 
-/** Paginated run history for one automation. Newest-first per backend. */
+interface UseAutomationRunsResult {
+	data: { items: LiveRunSummary[]; total: number } | undefined;
+	isLoading: boolean;
+	error: Error | null;
+}
+
+/**
+ * Live run history for one automation, newest-first. Sourced from Zero's
+ * thin ``automation_runs`` publication so status and per-step progress
+ * tick in real time without polling. Heavy fields (output, artifacts,
+ * inputs, error, definition_snapshot) are still fetched lazily via
+ * {@link useAutomationRun}.
+ */
 export function useAutomationRuns(
 	automationId: number | undefined,
-	{ limit = DEFAULT_LIMIT, offset = DEFAULT_OFFSET, enabled = true }: UseAutomationRunsOptions = {}
-) {
-	return useQuery<RunListResponse, Error>({
-		queryKey: cacheKeys.automations.runs(automationId ?? 0, limit, offset),
-		queryFn: () => automationsApiService.listRuns(automationId as number, { limit, offset }),
-		enabled: enabled && !!automationId,
-		staleTime: 30_000,
-	});
+	{ limit = DEFAULT_LIMIT }: UseAutomationRunsOptions = {}
+): UseAutomationRunsResult {
+	const [rows, result] = useZeroQuery(
+		queries.automationRuns.byAutomation({ automationId: automationId ?? -1 })
+	);
+
+	const items = useMemo<LiveRunSummary[]>(() => {
+		if (!automationId) return [];
+		return rows.slice(0, limit).map(toLiveRunSummary);
+	}, [automationId, rows, limit]);
+
+	const total = automationId ? rows.length : 0;
+
+	// Pre-hydration window: nothing visible AND Zero hasn't confirmed
+	// completeness yet. After the first sync (even an empty set) we stop
+	// showing the skeleton so the empty-state copy can take over.
+	const isLoading = !!automationId && result.type !== "complete" && rows.length === 0;
+
+	return {
+		data: automationId ? { items, total } : undefined,
+		isLoading,
+		error: null,
+	};
 }
 
-/** Single run with the full snapshot, step results, output and artifacts. */
+/**
+ * Full run record (definition snapshot, inputs, output, artifacts, error).
+ * Stays on REST: these fields are large and largely static after the run
+ * finishes, so they're not worth replicating to every connected client.
+ */
 export function useAutomationRun(
 	automationId: number | undefined,
 	runId: number | undefined,
 	options: { enabled?: boolean } = {}
 ) {
 	const { enabled = true } = options;
-	return useQuery<Run, Error>({
+	return useReactQuery<Run, Error>({
 		queryKey: cacheKeys.automations.run(automationId ?? 0, runId ?? 0),
 		queryFn: () => automationsApiService.getRun(automationId as number, runId as number),
 		enabled: enabled && !!automationId && !!runId,
 		staleTime: 30_000,
 	});
 }
+
+interface ZeroAutomationRunRow {
+	id: number;
+	automationId: number;
+	triggerId?: number | null;
+	status: string;
+	stepResults: unknown;
+	startedAt?: number | null;
+	finishedAt?: number | null;
+	createdAt: number;
+}
+
+/** Adapt a Zero camelCase row (epoch ms timestamps) to the snake_case
+ * ISO-string ``RunSummary`` shape the existing UI already consumes. */
+function toLiveRunSummary(row: ZeroAutomationRunRow): LiveRunSummary {
+	return {
+		id: row.id,
+		automation_id: row.automationId,
+		trigger_id: row.triggerId ?? null,
+		status: row.status as RunSummary["status"],
+		started_at: row.startedAt ? new Date(row.startedAt).toISOString() : null,
+		finished_at: row.finishedAt ? new Date(row.finishedAt).toISOString() : null,
+		created_at: new Date(row.createdAt).toISOString(),
+		step_results: Array.isArray(row.stepResults) ? (row.stepResults as RunStepResult[]) : [],
+	};
+}
diff --git a/surfsense_web/zero/queries/automations.ts b/surfsense_web/zero/queries/automations.ts
new file mode 100644
index 000000000..27b274acc
--- /dev/null
+++ b/surfsense_web/zero/queries/automations.ts
@@ -0,0 +1,14 @@
+import { defineQuery } from "@rocicorp/zero";
+import { z } from "zod";
+import { zql } from "../schema/index";
+
+// Mirrors chat byThread: client passes the parent id, the REST route still
+// authorizes via `automation_id -> search_space`. No search_space_id on the
+// table by design.
+export const automationRunQueries = {
+	byAutomation: defineQuery(
+		z.object({ automationId: z.number() }),
+		({ args: { automationId } }) =>
+			zql.automation_runs.where("automationId", automationId).orderBy("createdAt", "desc")
+	),
+};
diff --git a/surfsense_web/zero/queries/index.ts b/surfsense_web/zero/queries/index.ts
index fbf1bd76e..fe711f5d3 100644
--- a/surfsense_web/zero/queries/index.ts
+++ b/surfsense_web/zero/queries/index.ts
@@ -1,4 +1,5 @@
 import { defineQueries } from "@rocicorp/zero";
+import { automationRunQueries } from "./automations";
 import { chatSessionQueries, commentQueries, messageQueries } from "./chat";
 import { connectorQueries, documentQueries } from "./documents";
 import { folderQueries } from "./folders";
@@ -14,4 +15,5 @@ export const queries = defineQueries({
 	comments: commentQueries,
 	chatSession: chatSessionQueries,
 	user: userQueries,
+	automationRuns: automationRunQueries,
 });
diff --git a/surfsense_web/zero/schema/automations.ts b/surfsense_web/zero/schema/automations.ts
new file mode 100644
index 000000000..4d6ebfac7
--- /dev/null
+++ b/surfsense_web/zero/schema/automations.ts
@@ -0,0 +1,18 @@
+import { json, number, string, table } from "@rocicorp/zero";
+
+// Thin live row: status + per-step progress only. Heavy fields
+// (definition_snapshot, inputs, output, artifacts, error) stay on REST
+// (`GET /automations/{id}/runs/{run_id}`) and load on detail expand.
+// Mirrors the publication shape in migration 148.
+export const automationRunTable = table("automation_runs")
+	.columns({
+		id: number(),
+		automationId: number().from("automation_id"),
+		triggerId: number().optional().from("trigger_id"),
+		status: string(),
+		stepResults: json().from("step_results"),
+		startedAt: number().optional().from("started_at"),
+		finishedAt: number().optional().from("finished_at"),
+		createdAt: number().from("created_at"),
+	})
+	.primaryKey("id");
diff --git a/surfsense_web/zero/schema/index.ts b/surfsense_web/zero/schema/index.ts
index 3cca0f24a..d6731e371 100644
--- a/surfsense_web/zero/schema/index.ts
+++ b/surfsense_web/zero/schema/index.ts
@@ -1,4 +1,5 @@
 import { createBuilder, createSchema, relationships } from "@rocicorp/zero";
+import { automationRunTable } from "./automations";
 import { chatCommentTable, chatSessionStateTable, newChatMessageTable } from "./chat";
 import { documentTable, searchSourceConnectorTable } from "./documents";
 import { folderTable } from "./folders";
@@ -36,6 +37,7 @@ export const schema = createSchema({
 		chatCommentTable,
 		chatSessionStateTable,
 		userTable,
+		automationRunTable,
 	],
 	relationships: [chatCommentRelationships, newChatMessageRelationships],
 });