diff --git a/docker-compose.yml b/docker-compose.yml
new file mode 100644
index 000000000..9b0f7be31
--- /dev/null
+++ b/docker-compose.yml
@@ -0,0 +1,30 @@
+# docker-compose.yml — Local PostgreSQL for CleverAgents server-mode development
+#
+# Usage:
+#   docker compose up -d          # start PostgreSQL
+#   docker compose down -v        # tear down and remove data volume
+#
+# Connect from CleverAgents:
+#   export CLEVERAGENTS_SERVER_MODE=true
+#   export CLEVERAGENTS_DATABASE_URL=postgresql://cleveragents:cleveragents@localhost:5432/cleveragents
+
+services:
+  postgres:
+    image: postgres:16-alpine
+    environment:
+      POSTGRES_USER: cleveragents
+      # WARNING: Development credentials only — do not use in production
+      POSTGRES_PASSWORD: cleveragents
+      POSTGRES_DB: cleveragents
+    ports:
+      - "5432:5432"
+    volumes:
+      - pgdata:/var/lib/postgresql/data
+    healthcheck:
+      test: ["CMD-SHELL", "pg_isready -U cleveragents"]
+      interval: 5s
+      timeout: 5s
+      retries: 5
+
+volumes:
+  pgdata:
diff --git a/features/postgresql_backend.feature b/features/postgresql_backend.feature
new file mode 100644
index 000000000..361e06262
--- /dev/null
+++ b/features/postgresql_backend.feature
@@ -0,0 +1,111 @@
+@phase1 @database @postgresql @server_mode
+Feature: PostgreSQL Storage Backend for Server Mode
+  As a platform operator
+  I want CleverAgents to support PostgreSQL as its storage backend
+  So that server mode can handle multi-user collaborative access
+
+  Background:
+    Given I have imported the database infrastructure modules
+
+  # ── Settings & URL Selection ──────────────────────────────────────
+
+  @settings @database_url
+  Scenario: Default database URL is SQLite in local mode
+    Given I create settings with server_mode disabled
+    Then the resolved database URL should start with "sqlite"
+
+  @settings @database_url
+  Scenario: Server mode with default URL raises ValueError
+    Given I create settings with server_mode enabled
+    And the database_url is the default SQLite value
+    Then resolving the database URL should raise a ValueError
+
+  @settings @database_url
+  Scenario: Server mode with explicit PostgreSQL URL uses that URL
+    Given I create settings with server_mode enabled
+    And the database_url is "postgresql://myhost/mydb"
+    Then the resolved database URL should be "postgresql://myhost/mydb"
+
+  @settings @database_url
+  Scenario: is_postgresql returns True for PostgreSQL URLs
+    Given I create settings with server_mode enabled
+    And the database_url is "postgresql://localhost/cleveragents"
+    Then is_postgresql should return True
+
+  @settings @database_url
+  Scenario: is_postgresql returns False for SQLite URLs
+    Given I create settings with server_mode disabled
+    Then is_postgresql should return False
+
+  @settings @database_url
+  Scenario: is_postgresql returns False when server mode raises ValueError for SQLite URL
+    Given I create settings with server_mode enabled
+    And the database_url is the default SQLite value
+    Then is_postgresql should return False
+
+  # ── Pool Configuration ────────────────────────────────────────────
+
+  @settings @pool
+  Scenario: Default pool settings are appropriate for multi-user
+    Given I create settings with server_mode enabled
+    Then the db_pool_size should be 5
+    And the db_max_overflow should be 10
+    And the db_pool_recycle should be 1800
+
+  @settings @pool
+  Scenario: Pool settings can be overridden via environment
+    Given I create settings with custom pool configuration
+      | setting         | value |
+      | db_pool_size    | 20    |
+      | db_max_overflow | 30    |
+      | db_pool_recycle | 900   |
+    Then the db_pool_size should be 20
+    And the db_max_overflow should be 30
+    And the db_pool_recycle should be 900
+
+  # ── UnitOfWork Engine Creation ────────────────────────────────────
+
+  @engine @sqlite
+  Scenario: UnitOfWork creates SQLite engine for local mode
+    Given I create a UnitOfWork with URL "sqlite:///:memory:"
+    When I access the engine
+    Then the engine dialect should be "sqlite"
+
+  @engine @pool
+  Scenario: UnitOfWork passes pool parameters to non-SQLite engines
+    Given I create a UnitOfWork with URL "sqlite:///:memory:" and pool_size 8
+    Then the UnitOfWork pool_size should be 8
+    And the UnitOfWork max_overflow should be 10
+    And the UnitOfWork pool_recycle should be 1800
+
+  # ── ORM Model Dialect Compatibility ───────────────────────────────
+
+  @models @dialect
+  Scenario: ORM models use only dialect-agnostic column types
+    Given I inspect all ORM models from the Base metadata
+    Then no column should use a SQLite-specific type
+    And all columns should use standard SQLAlchemy types
+
+  @models @dialect
+  Scenario: ORM models define no raw SQL or sqlite3 imports
+    Given I read the models module source code
+    Then the source should not contain "sqlite3"
+    And the source should not contain "PRAGMA"
+    And the source should not contain "raw_sql"
+
+  @models @dialect
+  Scenario: All table names are valid PostgreSQL identifiers
+    Given I inspect all ORM models from the Base metadata
+    Then all table names should be valid SQL identifiers
+
+  # ── Migration Runner ──────────────────────────────────────────────
+
+  @migration
+  Scenario: Migration runner handles SQLite URLs
+    Given I create a MigrationRunner with URL "sqlite:///:memory:"
+    Then the migration runner database_url should start with "sqlite"
+
+  @migration
+  Scenario: Migration runner handles PostgreSQL URLs
+    Given I create a MigrationRunner with URL "postgresql://localhost/test"
+    Then the migration runner database_url should start with "postgresql"
diff --git a/features/steps/postgresql_backend_steps.py b/features/steps/postgresql_backend_steps.py
new file mode 100644
index 000000000..22a0154cc
--- /dev/null
+++ b/features/steps/postgresql_backend_steps.py
@@ -0,0 +1,369 @@
+"""Step definitions for PostgreSQL storage backend BDD scenarios.
+
+Steps specific to the ``postgresql_backend.feature`` file.  The shared
+``I have imported the database infrastructure modules`` step lives in
+``database_infrastructure_steps.py`` and is reused here via Behave's
+step-sharing mechanism.
+"""
+
+from __future__ import annotations
+
+import inspect
+import re
+
+from behave import given, then, when
+from behave.runner import Context
+
+# ---------------------------------------------------------------------------
+# Settings helpers
+# ---------------------------------------------------------------------------
+
+
+@given("I create settings with server_mode disabled")
+def step_settings_local_mode(context: Context) -> None:
+    """Create settings with local (SQLite) mode.
+
+    Clears ``CLEVERAGENTS_SERVER_MODE`` and restores the default
+    ``database_url`` so Settings resolves to SQLite.
+    """
+    import os
+
+    from cleveragents.config.settings import Settings
+
+    saved_db_url = os.environ.pop("CLEVERAGENTS_DATABASE_URL", None)
+    os.environ.pop("CLEVERAGENTS_SERVER_MODE", None)
+    try:
+        context.settings = Settings()
+    finally:
+        if saved_db_url is not None:
+            os.environ["CLEVERAGENTS_DATABASE_URL"] = saved_db_url
+
+
+@given("I create settings with server_mode enabled")
+def step_settings_server_mode(context: Context) -> None:
+    """Create settings with server mode enabled.
+
+    The ``before_scenario`` hook injects a per-scenario SQLite URL into
+    ``CLEVERAGENTS_DATABASE_URL``.  We must clear it so that Settings
+    falls back to its declared default (``sqlite:///cleveragents.db``),
+    allowing ``resolve_database_url`` to detect the default and switch to
+    PostgreSQL.
+    """
+    import os
+
+    from cleveragents.config.settings import Settings
+
+    saved_db_url = os.environ.pop("CLEVERAGENTS_DATABASE_URL", None)
+    os.environ["CLEVERAGENTS_SERVER_MODE"] = "true"
+    try:
+        context.settings = Settings()
+    finally:
+        os.environ.pop("CLEVERAGENTS_SERVER_MODE", None)
+        if saved_db_url is not None:
+            os.environ["CLEVERAGENTS_DATABASE_URL"] = saved_db_url
+
+
+@given("the database_url is the default SQLite value")
+def step_database_url_is_default(context: Context) -> None:
+    """Ensure database_url is the SQLite default.
+
+    The default is already ``sqlite:///cleveragents.db`` so this is a
+    no-op guard ensuring the previous step didn't override it.
+    """
+    assert context.settings.database_url.startswith("sqlite")
+
+
+@given('the database_url is "{url}"')
+def step_database_url_is(context: Context, url: str) -> None:
+    """Recreate settings with an explicit database_url."""
+    import os
+
+    from cleveragents.config.settings import Settings
+
+    os.environ["CLEVERAGENTS_DATABASE_URL"] = url
+    server = "true" if context.settings.server_mode else "false"
+    os.environ["CLEVERAGENTS_SERVER_MODE"] = server
+    try:
+        context.settings = Settings()
+    finally:
+        os.environ.pop("CLEVERAGENTS_DATABASE_URL", None)
+        os.environ.pop("CLEVERAGENTS_SERVER_MODE", None)
+
+
+@given("I create settings with custom pool configuration")
+def step_settings_custom_pool(context: Context) -> None:
+    """Create settings with custom pool values from the step table.
+
+    Pydantic-settings reads from environment variables, so we set the
+    ``CLEVERAGENTS_*`` env vars before constructing the Settings instance
+    and clean them up afterward.  The ``before_scenario`` per-scenario
+    database URL is temporarily cleared so the default applies.
+    """
+    import os
+
+    from cleveragents.config.settings import Settings
+
+    env_map: dict[str, str] = {
+        "db_pool_size": "CLEVERAGENTS_DB_POOL_SIZE",
+        "db_max_overflow": "CLEVERAGENTS_DB_MAX_OVERFLOW",
+        "db_pool_recycle": "CLEVERAGENTS_DB_POOL_RECYCLE",
+    }
+    set_vars: list[str] = []
+
+    saved_db_url = os.environ.pop("CLEVERAGENTS_DATABASE_URL", None)
+    os.environ["CLEVERAGENTS_SERVER_MODE"] = "true"
+    set_vars.append("CLEVERAGENTS_SERVER_MODE")
+
+    if context.table is not None:
+        for row in context.table:
+            key = row["setting"]
+            value = row["value"]
+            env_name = env_map.get(key, f"CLEVERAGENTS_{key.upper()}")
+            os.environ[env_name] = str(value)
+            set_vars.append(env_name)
+
+    try:
+        context.settings = Settings()
+    finally:
+        for var in set_vars:
+            os.environ.pop(var, None)
+        if saved_db_url is not None:
+            os.environ["CLEVERAGENTS_DATABASE_URL"] = saved_db_url
+
+
+# ---------------------------------------------------------------------------
+# Settings assertions
+# ---------------------------------------------------------------------------
+
+
+@then("resolving the database URL should raise a ValueError")
+def step_resolve_url_raises_value_error(context: Context) -> None:
+    """Assert that resolve_database_url raises ValueError in server mode."""
+    try:
+        context.settings.resolve_database_url()
+    except ValueError:
+        pass  # expected
+    else:
+        raise AssertionError("Expected ValueError from resolve_database_url()")
+
+
+@then('the resolved database URL should start with "{prefix}"')
+def step_resolved_url_starts_with(context: Context, prefix: str) -> None:
+    """Assert the resolved database URL prefix."""
+    url = context.settings.resolve_database_url()
+    assert url.startswith(prefix), f"Expected URL to start with '{prefix}', got '{url}'"
+
+
+@then('the resolved database URL should be "{expected}"')
+def step_resolved_url_equals(context: Context, expected: str) -> None:
+    """Assert the resolved database URL matches exactly."""
+    url = context.settings.resolve_database_url()
+    assert url == expected, f"Expected '{expected}', got '{url}'"
+
+
+@then("is_postgresql should return True")
+def step_is_postgresql_true(context: Context) -> None:
+    """Assert is_postgresql returns True."""
+    assert context.settings.is_postgresql(), "Expected is_postgresql() to be True"
+
+
+@then("is_postgresql should return False")
+def step_is_postgresql_false(context: Context) -> None:
+    """Assert is_postgresql returns False."""
+    assert not context.settings.is_postgresql(), "Expected is_postgresql() to be False"
+
+
+@then("the db_pool_size should be {value:d}")
+def step_pool_size_equals(context: Context, value: int) -> None:
+    """Assert db_pool_size setting."""
+    assert context.settings.db_pool_size == value, (
+        f"Expected {value}, got {context.settings.db_pool_size}"
+    )
+
+
+@then("the db_max_overflow should be {value:d}")
+def step_max_overflow_equals(context: Context, value: int) -> None:
+    """Assert db_max_overflow setting."""
+    assert context.settings.db_max_overflow == value, (
+        f"Expected {value}, got {context.settings.db_max_overflow}"
+    )
+
+
+@then("the db_pool_recycle should be {value:d}")
+def step_pool_recycle_equals(context: Context, value: int) -> None:
+    """Assert db_pool_recycle setting."""
+    assert context.settings.db_pool_recycle == value, (
+        f"Expected {value}, got {context.settings.db_pool_recycle}"
+    )
+
+
+# ---------------------------------------------------------------------------
+# UnitOfWork engine
+# ---------------------------------------------------------------------------
+
+_VALID_SQL_IDENTIFIER = re.compile(r"^[a-z_][a-z0-9_]*$")
+
+
+@given('I create a UnitOfWork with URL "{url}"')
+def step_create_uow(context: Context, url: str) -> None:
+    """Create a UnitOfWork instance with the given URL."""
+    from cleveragents.infrastructure.database.unit_of_work import UnitOfWork
+
+    context.uow = UnitOfWork(
+        url,
+        prompt_for_migration=lambda _msg: True,
+    )
+
+
+@given('I create a UnitOfWork with URL "{url}" and pool_size {size:d}')
+def step_create_uow_with_pool(context: Context, url: str, size: int) -> None:
+    """Create a UnitOfWork with explicit pool_size."""
+    from cleveragents.infrastructure.database.unit_of_work import UnitOfWork
+
+    context.uow = UnitOfWork(
+        url,
+        prompt_for_migration=lambda _msg: True,
+        pool_size=size,
+    )
+
+
+@when("I access the engine")
+def step_access_engine(context: Context) -> None:
+    """Trigger engine creation."""
+    context.engine = context.uow.engine
+
+
+@then('the engine dialect should be "{dialect}"')
+def step_engine_dialect(context: Context, dialect: str) -> None:
+    """Assert the engine dialect name."""
+    actual = context.engine.dialect.name
+    assert actual == dialect, f"Expected dialect '{dialect}', got '{actual}'"
+
+
+@then("the UnitOfWork pool_size should be {value:d}")
+def step_uow_pool_size(context: Context, value: int) -> None:
+    """Assert UnitOfWork.pool_size via read-only property."""
+    assert context.uow.pool_size == value
+
+
+@then("the UnitOfWork max_overflow should be {value:d}")
+def step_uow_max_overflow(context: Context, value: int) -> None:
+    """Assert UnitOfWork.max_overflow via read-only property."""
+    assert context.uow.max_overflow == value
+
+
+@then("the UnitOfWork pool_recycle should be {value:d}")
+def step_uow_pool_recycle(context: Context, value: int) -> None:
+    """Assert UnitOfWork.pool_recycle via read-only property."""
+    assert context.uow.pool_recycle == value
+
+
+# ---------------------------------------------------------------------------
+# ORM Model dialect-agnostic checks
+# ---------------------------------------------------------------------------
+
+# Standard SQLAlchemy types that are portable across all dialects.
+_PORTABLE_TYPE_NAMES = frozenset(
+    {
+        "VARCHAR",
+        "STRING",
+        "TEXT",
+        "INTEGER",
+        "BIGINTEGER",
+        "SMALLINTEGER",
+        "FLOAT",
+        "NUMERIC",
+        "BOOLEAN",
+        "DATE",
+        "DATETIME",
+        "TIME",
+        "TIMESTAMP",
+        "JSON",
+        "JSONB",
+        "BLOB",
+        "LARGEBINARY",
+        "ENUM",
+        "UUID",
+        "INTERVAL",
+        "ARRAY",
+    }
+)
+
+
+@given("I inspect all ORM models from the Base metadata")
+def step_inspect_orm_models(context: Context) -> None:
+    """Collect all tables from the ORM metadata."""
+    from cleveragents.infrastructure.database.models import Base
+
+    context.metadata = Base.metadata
+    context.tables = list(Base.metadata.tables.values())
+
+
+@then("no column should use a SQLite-specific type")
+def step_no_sqlite_specific_types(context: Context) -> None:
+    """Assert no column uses a SQLite-only type."""
+    for table in context.tables:
+        for col in table.columns:
+            type_name = type(col.type).__name__.upper()
+            # Known SQLite-only types that are NOT portable
+            sqlite_only = {"SQLITEINTEGER", "SQLITETEXT", "SQLITEBLOB"}
+            assert type_name not in sqlite_only, (
+                f"Column {table.name}.{col.name} uses SQLite-specific type {type_name}"
+            )
+
+
+@then("all columns should use standard SQLAlchemy types")
+def step_all_standard_types(context: Context) -> None:
+    """Assert all columns use portable SQLAlchemy types."""
+    for table in context.tables:
+        for col in table.columns:
+            type_name = type(col.type).__name__.upper()
+            # Allow common base types and their variants
+            assert type_name in _PORTABLE_TYPE_NAMES or any(
+                base in type_name for base in _PORTABLE_TYPE_NAMES
+            ), f"Column {table.name}.{col.name} uses non-standard type {type_name}"
+
+
+@then("all table names should be valid SQL identifiers")
+def step_valid_table_names(context: Context) -> None:
+    """Assert all table names are valid (lowercase) SQL identifiers."""
+    for table in context.tables:
+        assert _VALID_SQL_IDENTIFIER.match(table.name), (
+            f"Table name '{table.name}' is not a valid SQL identifier"
+        )
+
+
+@given("I read the models module source code")
+def step_read_models_source(context: Context) -> None:
+    """Read the models.py source for textual analysis."""
+    from cleveragents.infrastructure.database import models
+
+    context.models_source = inspect.getsource(models)
+
+
+@then('the source should not contain "{forbidden}"')
+def step_source_not_contain(context: Context, forbidden: str) -> None:
+    """Assert the source code does not contain a forbidden string."""
+    assert forbidden not in context.models_source, (
+        f"models.py should not contain '{forbidden}'"
+    )
+
+
+# ---------------------------------------------------------------------------
+# Migration runner
+# ---------------------------------------------------------------------------
+
+
+@given('I create a MigrationRunner with URL "{url}"')
+def step_create_migration_runner(context: Context, url: str) -> None:
+    """Create a MigrationRunner instance."""
+    from cleveragents.infrastructure.database.migration_runner import MigrationRunner
+
+    context.migration_runner = MigrationRunner(url)
+
+
+@then('the migration runner database_url should start with "{prefix}"')
+def step_runner_url_starts_with(context: Context, prefix: str) -> None:
+    """Assert migration runner database_url prefix."""
+    url = context.migration_runner.database_url
+    assert url.startswith(prefix), f"Expected '{prefix}', got '{url}'"
diff --git a/features/steps/uow_coverage_boost_steps.py b/features/steps/uow_coverage_boost_steps.py
index 9a5bf3698..74cf1db00 100644
--- a/features/steps/uow_coverage_boost_steps.py
+++ b/features/steps/uow_coverage_boost_steps.py
@@ -53,20 +53,31 @@ def step_uow_non_sqlite_url(context: Context) -> None:
 @when("I access the engine property with mocked create_engine")
 def step_access_engine_property_mocked(context: Context) -> None:
     """Access the engine property, mocking _ensure_database_initialized and create_engine."""
+    import sys as _sys
+
     from sqlalchemy.engine import Engine
 
     mock_engine = MagicMock(spec=Engine)
 
-    with (
-        patch.object(context.uow, "_ensure_database_initialized"),
-        patch(
-            "cleveragents.infrastructure.database.unit_of_work.create_engine",
-            return_value=mock_engine,
-        ) as mock_create,
-    ):
-        engine = context.uow.engine
-        context.result_engine = engine
-        context.mock_create_engine = mock_create
+    # Ensure psycopg2 can be imported (it may not be installed in test env)
+    _needs_psycopg2_stub = "psycopg2" not in _sys.modules
+    if _needs_psycopg2_stub:
+        _sys.modules["psycopg2"] = MagicMock()
+
+    try:
+        with (
+            patch.object(context.uow, "_ensure_database_initialized"),
+            patch(
+                "cleveragents.infrastructure.database.unit_of_work.create_engine",
+                return_value=mock_engine,
+            ) as mock_create,
+        ):
+            engine = context.uow.engine
+            context.result_engine = engine
+            context.mock_create_engine = mock_create
+    finally:
+        if _needs_psycopg2_stub:
+            _sys.modules.pop("psycopg2", None)
 
 
 @then("the engine should be created via the non-SQLite code path")
@@ -232,20 +243,31 @@ def step_uow_postgresql_url(context: Context) -> None:
 @when("I trigger engine creation with mocked dependencies")
 def step_trigger_engine_creation_mocked(context: Context) -> None:
     """Access engine property with mocked create_engine and migration runner."""
+    import sys as _sys
+
     from sqlalchemy.engine import Engine
 
     mock_engine = MagicMock(spec=Engine)
 
-    with (
-        patch.object(context.uow, "_ensure_database_initialized"),
-        patch(
-            "cleveragents.infrastructure.database.unit_of_work.create_engine",
-            return_value=mock_engine,
-        ) as mock_create,
-    ):
-        engine = context.uow.engine
-        context.result_engine = engine
-        context.mock_create_engine = mock_create
+    # Ensure psycopg2 can be imported (it may not be installed in test env)
+    _needs_psycopg2_stub = "psycopg2" not in _sys.modules
+    if _needs_psycopg2_stub:
+        _sys.modules["psycopg2"] = MagicMock()
+
+    try:
+        with (
+            patch.object(context.uow, "_ensure_database_initialized"),
+            patch(
+                "cleveragents.infrastructure.database.unit_of_work.create_engine",
+                return_value=mock_engine,
+            ) as mock_create,
+        ):
+            engine = context.uow.engine
+            context.result_engine = engine
+            context.mock_create_engine = mock_create
+    finally:
+        if _needs_psycopg2_stub:
+            _sys.modules.pop("psycopg2", None)
 
 
 @then("create_engine should have been called without isolation_level")
@@ -299,3 +321,145 @@ def step_second_engine_same_as_first(context: Context) -> None:
     if hasattr(context, "saved_memory_engines"):
         MEMORY_ENGINES.clear()
         MEMORY_ENGINES.update(context.saved_memory_engines)
+
+
+# =========================================================================
+# Argument validation in __init__ (lines 68-75)
+# =========================================================================
+
+
+@when("I try to create a UnitOfWork with an empty database_url")
+def step_uow_empty_database_url(context: Context) -> None:
+    """Try to create UnitOfWork with empty database_url; capture ValueError."""
+    try:
+        UnitOfWork(database_url="")
+        context.last_error = None
+    except ValueError as exc:
+        context.last_error = exc
+
+
+@when("I try to create a UnitOfWork with pool_size set to {value:d}")
+def step_uow_invalid_pool_size(context: Context, value: int) -> None:
+    """Try to create UnitOfWork with invalid pool_size; capture ValueError."""
+    try:
+        UnitOfWork(database_url="sqlite:///test.db", pool_size=value)
+        context.last_error = None
+    except ValueError as exc:
+        context.last_error = exc
+
+
+@when("I try to create a UnitOfWork with max_overflow set to {value:d}")
+def step_uow_invalid_max_overflow(context: Context, value: int) -> None:
+    """Try to create UnitOfWork with invalid max_overflow; capture ValueError."""
+    try:
+        UnitOfWork(database_url="sqlite:///test.db", max_overflow=value)
+        context.last_error = None
+    except ValueError as exc:
+        context.last_error = exc
+
+
+@when("I try to create a UnitOfWork with pool_recycle set to {value:d}")
+def step_uow_invalid_pool_recycle(context: Context, value: int) -> None:
+    """Try to create UnitOfWork with invalid pool_recycle; capture ValueError."""
+    try:
+        UnitOfWork(database_url="sqlite:///test.db", pool_recycle=value)
+        context.last_error = None
+    except ValueError as exc:
+        context.last_error = exc
+
+
+# =========================================================================
+# dispose() method (lines 196-205)
+# =========================================================================
+
+
+@given("a UnitOfWork with a mocked non-memory engine")
+def step_uow_mocked_non_memory_engine(context: Context) -> None:
+    """Create a UnitOfWork with a pre-set mocked engine to test dispose()."""
+    from sqlalchemy.engine import Engine
+
+    context.uow = UnitOfWork(
+        database_url="postgresql://user:pass@localhost:5432/testdb"
+    )
+    context.mock_engine = MagicMock(spec=Engine)
+    context.uow._engine = context.mock_engine
+
+
+@when("I call dispose on the UnitOfWork")
+def step_call_dispose_on_uow(context: Context) -> None:
+    """Call dispose() on the UnitOfWork stored in context."""
+    context.uow.dispose()
+
+
+@then("the internal engine should have been disposed and cleared")
+def step_engine_disposed_and_cleared(context: Context) -> None:
+    """Verify dispose() called engine.dispose() and cleared internal references."""
+    context.mock_engine.dispose.assert_called_once()
+    assert context.uow._engine is None, (
+        f"Expected _engine to be None after dispose(), got {context.uow._engine}"
+    )
+    assert context.uow._session_factory is None, (
+        "Expected _session_factory to be None after dispose()"
+    )
+
+
+@then("the dispose call should complete without error")
+def step_dispose_no_error(context: Context) -> None:
+    """Verify dispose() completes without raising (False branch — engine is None)."""
+    assert context.uow._engine is None, (
+        "Expected _engine to remain None when dispose() is a no-op"
+    )
+
+
+# =========================================================================
+# Context manager protocol (__enter__ / __exit__) (lines 207-213)
+# =========================================================================
+
+
+@when("I use the UnitOfWork as a context manager")
+def step_use_uow_as_context_manager(context: Context) -> None:
+    """Use the UnitOfWork with a with-statement to exercise __enter__/__exit__."""
+    with context.uow as entered_uow:
+        context.entered_uow = entered_uow
+
+
+@then("the engine should be disposed on context manager exit")
+def step_engine_disposed_on_context_exit(context: Context) -> None:
+    """Verify __enter__ returned self and __exit__ triggered dispose()."""
+    assert context.entered_uow is context.uow, "__enter__ should return self"
+    context.mock_engine.dispose.assert_called_once()
+
+
+# =========================================================================
+# psycopg2 ImportError hint (lines 142-148)
+# =========================================================================
+
+
+@when("I access the engine with psycopg2 forcibly unavailable")
+def step_access_engine_no_psycopg2(context: Context) -> None:
+    """Access engine property while psycopg2 is blocked via sys.modules."""
+    import sys as _sys
+
+    try:
+        with (
+            patch.dict(_sys.modules, {"psycopg2": None}),
+            patch.object(context.uow, "_ensure_database_initialized"),
+        ):
+            _ = context.uow.engine
+        context.raised_exception = None
+    except ImportError as exc:
+        context.raised_exception = exc
+
+
+@then("an ImportError should be raised mentioning the server extra")
+def step_import_error_server_extra(context: Context) -> None:
+    """Verify ImportError was raised containing the install-hint message."""
+    assert context.raised_exception is not None, (
+        "Expected an ImportError to be raised but no exception was captured"
+    )
+    assert isinstance(context.raised_exception, ImportError), (
+        f"Expected ImportError, got {type(context.raised_exception)}"
+    )
+    assert "server" in str(context.raised_exception).lower(), (
+        f"Expected 'server' in exception message: '{context.raised_exception}'"
+    )
diff --git a/features/uow_coverage_boost.feature b/features/uow_coverage_boost.feature
index 19c6a1bc0..1a832d2bd 100644
--- a/features/uow_coverage_boost.feature
+++ b/features/uow_coverage_boost.feature
@@ -74,3 +74,63 @@ Feature: UnitOfWork coverage boost for uncovered code paths
     And a first UnitOfWork has already created an in-memory engine
     When a second UnitOfWork accesses its engine property
     Then the second engine should be the same object as the first
+
+  # ---------------------------------------------------------------------------
+  # Argument validation in __init__ (ValueError raises)
+  # ---------------------------------------------------------------------------
+
+  @uow_validate_empty_url
+  Scenario: UnitOfWork rejects an empty database_url
+    When I try to create a UnitOfWork with an empty database_url
+    Then a ValueError should be raised containing "database_url must be a non-empty string"
+
+  @uow_validate_pool_size
+  Scenario: UnitOfWork rejects pool_size less than 1
+    When I try to create a UnitOfWork with pool_size set to 0
+    Then a ValueError should be raised containing "pool_size must be >= 1"
+
+  @uow_validate_max_overflow
+  Scenario: UnitOfWork rejects negative max_overflow
+    When I try to create a UnitOfWork with max_overflow set to -1
+    Then a ValueError should be raised containing "max_overflow must be >= 0"
+
+  @uow_validate_pool_recycle
+  Scenario: UnitOfWork rejects pool_recycle below -1
+    When I try to create a UnitOfWork with pool_recycle set to -2
+    Then a ValueError should be raised containing "pool_recycle must be >= -1"
+
+  # ---------------------------------------------------------------------------
+  # dispose() method branches (lines 196-205)
+  # ---------------------------------------------------------------------------
+
+  @uow_dispose_active_engine
+  Scenario: UnitOfWork dispose clears a non-memory engine
+    Given a UnitOfWork with a mocked non-memory engine
+    When I call dispose on the UnitOfWork
+    Then the internal engine should have been disposed and cleared
+
+  @uow_dispose_no_engine
+  Scenario: UnitOfWork dispose is a no-op when engine has not been created
+    Given a UnitOfWork configured with a non-SQLite database URL
+    When I call dispose on the UnitOfWork
+    Then the dispose call should complete without error
+
+  # ---------------------------------------------------------------------------
+  # Context manager protocol (__enter__ / __exit__) (lines 207-213)
+  # ---------------------------------------------------------------------------
+
+  @uow_context_manager_protocol
+  Scenario: UnitOfWork context manager disposes engine on exit
+    Given a UnitOfWork with a mocked non-memory engine
+    When I use the UnitOfWork as a context manager
+    Then the engine should be disposed on context manager exit
+
+  # ---------------------------------------------------------------------------
+  # psycopg2 ImportError hint (lines 142-148)
+  # ---------------------------------------------------------------------------
+
+  @uow_psycopg2_missing
+  Scenario: UnitOfWork raises ImportError with install hint when psycopg2 is unavailable
+    Given a UnitOfWork configured with a non-SQLite database URL
+    When I access the engine with psycopg2 forcibly unavailable
+    Then an ImportError should be raised mentioning the server extra
diff --git a/pyproject.toml b/pyproject.toml
index 05e62c699..bb3240fad 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -52,6 +52,9 @@ dependencies = [
 ]
 
 [project.optional-dependencies]
+server = [
+    "psycopg2-binary>=2.9.9",  # PostgreSQL adapter for server-mode storage
+]
 tui = [
     "textual>=1.0.0,<2.0.0",
 ]
diff --git a/robot/postgresql_backend.robot b/robot/postgresql_backend.robot
new file mode 100644
index 000000000..c16548e86
--- /dev/null
+++ b/robot/postgresql_backend.robot
@@ -0,0 +1,166 @@
+*** Settings ***
+Documentation     Integration tests for PostgreSQL storage backend abstraction layer.
+...               Tests the dialect-agnostic ORM models, connection configuration,
+...               and database URL selection logic using SQLite as the test backend.
+...               Tests tagged @requires_postgresql are excluded from default CI runs.
+Library           OperatingSystem
+Library           String
+Library           Collections
+Library           Process
+Library           indentation_library.py
+Resource          ${CURDIR}/common.resource
+
+*** Variables ***
+${PYTHON}         python
+
+*** Test Cases ***
+
+Settings Resolve Database URL Local Mode
+    [Documentation]    Verify resolve_database_url returns SQLite in local mode
+    ${script}=    Catenate    SEPARATOR=\n
+    ...    import os
+    ...    os.environ.pop('CLEVERAGENTS_SERVER_MODE', None)
+    ...    os.environ.pop('CLEVERAGENTS_DATABASE_URL', None)
+    ...    from cleveragents.config.settings import Settings
+    ...    s = Settings()
+    ...    print(s.resolve_database_url())
+    ${result}=    Run Process    ${PYTHON}    -c    ${script}
+    ...    cwd=${CURDIR}/..
+    Should Be Equal As Integers    ${result.rc}    0
+    Should Start With    ${result.stdout.strip()}    sqlite
+
+Settings Resolve Database URL Server Mode
+    [Documentation]    Verify resolve_database_url returns PostgreSQL when server_mode is on
+    ...               and CLEVERAGENTS_DATABASE_URL points to a PostgreSQL instance.
+    ${script}=    Catenate    SEPARATOR=\n
+    ...    import os
+    ...    os.environ['CLEVERAGENTS_SERVER_MODE'] = 'true'
+    ...    os.environ['CLEVERAGENTS_DATABASE_URL'] = 'postgresql://localhost/cleveragents'
+    ...    from cleveragents.config.settings import Settings
+    ...    s = Settings()
+    ...    print(s.resolve_database_url())
+    ${result}=    Run Process    ${PYTHON}    -c    ${script}
+    ...    cwd=${CURDIR}/..
+    Should Be Equal As Integers    ${result.rc}    0
+    Should Start With    ${result.stdout.strip()}    postgresql
+
+Settings Is PostgreSQL Returns True For PG URL
+    [Documentation]    Verify is_postgresql detects PostgreSQL URLs
+    ${script}=    Catenate    SEPARATOR=\n
+    ...    import os
+    ...    os.environ['CLEVERAGENTS_SERVER_MODE'] = 'true'
+    ...    os.environ['CLEVERAGENTS_DATABASE_URL'] = 'postgresql://localhost/db'
+    ...    from cleveragents.config.settings import Settings
+    ...    s = Settings()
+    ...    print(s.is_postgresql())
+    ${result}=    Run Process    ${PYTHON}    -c    ${script}
+    ...    cwd=${CURDIR}/..
+    Should Be Equal As Integers    ${result.rc}    0
+    Should Be Equal As Strings    ${result.stdout.strip()}    True
+
+Settings Is PostgreSQL Returns False For SQLite URL
+    [Documentation]    Verify is_postgresql returns False for SQLite URLs
+    ${script}=    Catenate    SEPARATOR=\n
+    ...    import os
+    ...    os.environ.pop('CLEVERAGENTS_SERVER_MODE', None)
+    ...    os.environ.pop('CLEVERAGENTS_DATABASE_URL', None)
+    ...    from cleveragents.config.settings import Settings
+    ...    s = Settings()
+    ...    print(s.is_postgresql())
+    ${result}=    Run Process    ${PYTHON}    -c    ${script}
+    ...    cwd=${CURDIR}/..
+    Should Be Equal As Integers    ${result.rc}    0
+    Should Be Equal As Strings    ${result.stdout.strip()}    False
+
+UnitOfWork Creates SQLite Engine With Pool Parameters
+    [Documentation]    UnitOfWork stores pool parameters and creates an SQLite engine
+    ${result}=    Run Process    ${PYTHON}    -c
+    ...    from cleveragents.infrastructure.database.unit_of_work import UnitOfWork; u \= UnitOfWork('sqlite:///:memory:', prompt_for_migration\=lambda m: True, pool_size\=8, max_overflow\=15, pool_recycle\=900); print(u._pool_size, u._max_overflow, u._pool_recycle, u.engine.dialect.name)
+    ...    cwd=${CURDIR}/..
+    Should Be Equal As Integers    ${result.rc}    0
+    Should Contain    ${result.stdout.strip()}    8 15 900 sqlite
+
+ORM Models Use Dialect-Agnostic Types
+    [Documentation]    All ORM column types are portable across SQLite and PostgreSQL
+    ${script}=    Catenate    SEPARATOR=\n
+    ...    from cleveragents.infrastructure.database.models import Base
+    ...    PORTABLE = {'VARCHAR','STRING','TEXT','INTEGER','BIGINTEGER','SMALLINTEGER',
+    ...    'FLOAT','NUMERIC','BOOLEAN','DATE','DATETIME','TIME','TIMESTAMP','JSON',
+    ...    'BLOB','LARGEBINARY','ENUM'}
+    ...    for t in Base.metadata.tables.values():
+    ...    ${SPACE}${SPACE}${SPACE}${SPACE}for c in t.columns:
+    ...    ${SPACE}${SPACE}${SPACE}${SPACE}${SPACE}${SPACE}${SPACE}${SPACE}tn = type(c.type).__name__.upper()
+    ...    ${SPACE}${SPACE}${SPACE}${SPACE}${SPACE}${SPACE}${SPACE}${SPACE}if tn not in PORTABLE and not any(b in tn for b in PORTABLE):
+    ...    ${SPACE}${SPACE}${SPACE}${SPACE}${SPACE}${SPACE}${SPACE}${SPACE}${SPACE}${SPACE}${SPACE}${SPACE}raise AssertionError(f'{t.name}.{c.name} has non-portable type {tn}')
+    ...    print('ALL_PORTABLE')
+    ${result}=    Run Process    ${PYTHON}    -c    ${script}
+    ...    cwd=${CURDIR}/..
+    Should Be Equal As Integers    ${result.rc}    0
+    Should Contain    ${result.stdout.strip()}    ALL_PORTABLE
+
+ORM Models Source Contains No SQLite Specific Code
+    [Documentation]    models.py should not import sqlite3 or use PRAGMAs
+    ${script}=    Catenate    SEPARATOR=\n
+    ...    import inspect
+    ...    from cleveragents.infrastructure.database import models
+    ...    src = inspect.getsource(models)
+    ...    for forbidden in ['sqlite3', 'PRAGMA']:
+    ...    ${SPACE}${SPACE}${SPACE}${SPACE}assert forbidden not in src, f'Found {forbidden}'
+    ...    print('CLEAN')
+    ${result}=    Run Process    ${PYTHON}    -c    ${script}
+    ...    cwd=${CURDIR}/..
+    Should Be Equal As Integers    ${result.rc}    0
+    Should Contain    ${result.stdout.strip()}    CLEAN
+
+Table Names Are Valid PostgreSQL Identifiers
+    [Documentation]    All table names are lowercase SQL identifiers
+    ${script}=    Catenate    SEPARATOR=\n
+    ...    import re
+    ...    from cleveragents.infrastructure.database.models import Base
+    ...    pat = re.compile(r'^[a-z_][a-z0-9_]*$')
+    ...    for t in Base.metadata.tables.values():
+    ...    ${SPACE}${SPACE}${SPACE}${SPACE}assert pat.match(t.name), f'Invalid table: {t.name}'
+    ...    print('VALID')
+    ${result}=    Run Process    ${PYTHON}    -c    ${script}
+    ...    cwd=${CURDIR}/..
+    Should Be Equal As Integers    ${result.rc}    0
+    Should Contain    ${result.stdout.strip()}    VALID
+
+Docker Compose File Exists For PostgreSQL Development
+    [Documentation]    docker-compose.yml exists at project root
+    File Should Exist    ${CURDIR}/../docker-compose.yml
+
+Pool Defaults Are Multi-User Appropriate
+    [Documentation]    Default pool settings: pool_size=5, max_overflow=10, pool_recycle=1800
+    ${script}=    Catenate    SEPARATOR=\n
+    ...    import os
+    ...    os.environ['CLEVERAGENTS_SERVER_MODE'] = 'true'
+    ...    os.environ.pop('CLEVERAGENTS_DATABASE_URL', None)
+    ...    from cleveragents.config.settings import Settings
+    ...    s = Settings()
+    ...    print(s.db_pool_size, s.db_max_overflow, s.db_pool_recycle)
+    ${result}=    Run Process    ${PYTHON}    -c    ${script}
+    ...    cwd=${CURDIR}/..
+    Should Be Equal As Integers    ${result.rc}    0
+    Should Be Equal As Strings    ${result.stdout.strip()}    5 10 1800
+
+Migration Runner Accepts PostgreSQL URL
+    [Documentation]    MigrationRunner can be constructed with a PostgreSQL URL
+    ${result}=    Run Process    ${PYTHON}    -c
+    ...    from cleveragents.infrastructure.database.migration_runner import MigrationRunner; r \= MigrationRunner('postgresql://localhost/test'); print(r.database_url)
+    ...    cwd=${CURDIR}/..
+    Should Be Equal As Integers    ${result.rc}    0
+    Should Start With    ${result.stdout.strip()}    postgresql
+
+PostgreSQL Live Connection Test
+    [Documentation]    Connect to a real PostgreSQL instance and run migrations.
+    ...               Requires a running PostgreSQL server and CLEVERAGENTS_DATABASE_URL
+    ...               pointing to it. Excluded from default CI runs.
+    [Tags]    requires_postgresql
+    ${pg_url}=    Get Environment Variable    CLEVERAGENTS_DATABASE_URL    default=SKIP
+    Skip If    '${pg_url}' == 'SKIP'    PostgreSQL not available
+    ${result}=    Run Process    ${PYTHON}    -c
+    ...    from cleveragents.infrastructure.database.migration_runner import MigrationRunner; r \= MigrationRunner('${pg_url}'); r.init_or_upgrade(); print('OK')
+    ...    cwd=${CURDIR}/..
+    Should Be Equal As Integers    ${result.rc}    0
+    Should Contain    ${result.stdout.strip()}    OK
diff --git a/src/cleveragents/application/container.py b/src/cleveragents/application/container.py
index ab1b47517..03a79d5f4 100644
--- a/src/cleveragents/application/container.py
+++ b/src/cleveragents/application/container.py
@@ -251,6 +251,9 @@ def get_database_url() -> str:
     relative to ``CLEVERAGENTS_HOME`` (if set) rather than CWD, ensuring
     test isolation and correct data placement.
 
+    .. todo:: (#878) Use ``Settings.resolve_database_url()`` once
+       PostgreSQL wiring is complete so that *server_mode* is honoured.
+
     Returns:
         SQLAlchemy database URL with absolute path for SQLite
     """
diff --git a/src/cleveragents/cli/commands/db.py b/src/cleveragents/cli/commands/db.py
index 0d641d23c..bdee2b7b5 100644
--- a/src/cleveragents/cli/commands/db.py
+++ b/src/cleveragents/cli/commands/db.py
@@ -37,6 +37,7 @@ def _get_runner() -> MigrationRunner:
     from cleveragents.application.container import get_database_url
     from cleveragents.infrastructure.database.migration_runner import MigrationRunner
 
+    # TODO(#878): Use Settings.resolve_database_url() once PostgreSQL wiring is complete
     return MigrationRunner(get_database_url())
 
 
diff --git a/src/cleveragents/cli/commands/project.py b/src/cleveragents/cli/commands/project.py
index 7b95d6942..9fb55de18 100644
--- a/src/cleveragents/cli/commands/project.py
+++ b/src/cleveragents/cli/commands/project.py
@@ -126,6 +126,7 @@ def _store_project_extras(
 
     from cleveragents.application.container import get_database_url
 
+    # TODO(#878): Use Settings.resolve_database_url() once PostgreSQL wiring is complete
     db_url = get_database_url()
     engine = create_engine(db_url, echo=False)
     session = sessionmaker(bind=engine, expire_on_commit=False)()
diff --git a/src/cleveragents/cli/commands/system.py b/src/cleveragents/cli/commands/system.py
index 31a3dccc8..63282ab9a 100644
--- a/src/cleveragents/cli/commands/system.py
+++ b/src/cleveragents/cli/commands/system.py
@@ -413,6 +413,8 @@ def _check_stale_locks() -> dict[str, Any]:
     try:
         from sqlalchemy.orm import Session, sessionmaker
 
+        # TODO(#878): Use Settings.resolve_database_url()
+        # once PostgreSQL wiring is complete.
         db_url = get_database_url()
         engine = create_engine(db_url, echo=False)
 
diff --git a/src/cleveragents/config/settings.py b/src/cleveragents/config/settings.py
index 36ee04d48..87e52b802 100644
--- a/src/cleveragents/config/settings.py
+++ b/src/cleveragents/config/settings.py
@@ -386,6 +386,38 @@ class Settings(BaseSettings):
         validation_alias=AliasChoices("CLEVERAGENTS_TEST_DATABASE_URL"),
     )
 
+    # Server mode toggle — when True, the application operates as a
+    # collaborative multi-user server, which selects a PostgreSQL-compatible
+    # database URL by default instead of SQLite.
+    server_mode: bool = Field(
+        default=False,
+        validation_alias=AliasChoices("CLEVERAGENTS_SERVER_MODE"),
+        description="Enable server mode for multi-user collaborative access.",
+    )
+
+    # PostgreSQL connection-pool tuning (server mode)
+    db_pool_size: int = Field(
+        default=5,
+        ge=1,
+        validation_alias=AliasChoices("CLEVERAGENTS_DB_POOL_SIZE"),
+        description="SQLAlchemy connection-pool size for PostgreSQL.",
+    )
+    db_max_overflow: int = Field(
+        default=10,
+        ge=0,
+        validation_alias=AliasChoices("CLEVERAGENTS_DB_MAX_OVERFLOW"),
+        description="Extra connections above pool_size allowed under load.",
+    )
+    db_pool_recycle: int = Field(
+        default=1800,
+        ge=-1,
+        validation_alias=AliasChoices("CLEVERAGENTS_DB_POOL_RECYCLE"),
+        description=(
+            "Seconds after which a pooled connection is recycled. "
+            "-1 disables recycling."
+        ),
+    )
+
     # Context tier budgets (ACMS #208)
     context_max_tokens_hot: int = Field(
         default=32000,
@@ -893,6 +925,42 @@ class Settings(BaseSettings):
             return self._derive_test_database_url(self.database_url)
         return self.database_url
 
+    def resolve_database_url(self, *, test: bool = False) -> str:
+        """Return the database URL, honouring *server_mode*.
+
+        In **server mode** the database URL must point to a PostgreSQL
+        instance.  If the caller has not provided an explicit PostgreSQL
+        URL via ``CLEVERAGENTS_DATABASE_URL``, a ``ValueError`` is
+        raised to prevent silent fallback to a possibly unreachable
+        default.
+
+        In **local mode** the existing SQLite URL is returned unchanged.
+
+        Args:
+            test: When True, derive or return the test database URL.
+
+        Returns:
+            A SQLAlchemy-compatible database URL string.
+
+        Raises:
+            ValueError: When *server_mode* is True but no explicit
+                PostgreSQL URL has been configured.
+        """
+        base_url = self.get_database_url(test=test)
+        if self.server_mode and base_url.startswith("sqlite"):
+            raise ValueError(
+                "server_mode=True requires an explicit database URL. "
+                "Set CLEVERAGENTS_DATABASE_URL to a PostgreSQL connection string."
+            )
+        return base_url
+
+    def is_postgresql(self) -> bool:
+        """Return True when the active database URL targets PostgreSQL."""
+        try:
+            return self.resolve_database_url().startswith("postgresql")
+        except ValueError:
+            return False
+
     def is_production(self) -> bool:
         """Whether the runtime matches production settings."""
         return (
diff --git a/src/cleveragents/infrastructure/database/migration_runner.py b/src/cleveragents/infrastructure/database/migration_runner.py
index 4f91f0477..c8205f668 100644
--- a/src/cleveragents/infrastructure/database/migration_runner.py
+++ b/src/cleveragents/infrastructure/database/migration_runner.py
@@ -167,9 +167,12 @@ class MigrationRunner:
             )
         else:
             engine = create_engine(self.database_url)
-        with engine.connect() as connection:
-            context = MigrationContext.configure(connection)
-            return context.get_current_revision()
+        try:
+            with engine.connect() as connection:
+                context = MigrationContext.configure(connection)
+                return context.get_current_revision()
+        finally:
+            engine.dispose()
 
     def get_pending_migrations(self) -> list[str]:
         """Get list of pending migrations.
@@ -241,7 +244,7 @@ class MigrationRunner:
                 )
 
         # For SQLite, ensure the parent directory exists
-        # Skip this for in-memory databases (:memory:)
+        # Skip this for in-memory databases (:memory:) and non-SQLite backends
         if (
             self.database_url.startswith("sqlite")
             and ":memory:" not in self.database_url
@@ -255,8 +258,10 @@ class MigrationRunner:
             db_file = Path(db_path)
             db_file.parent.mkdir(parents=True, exist_ok=True)
 
-        # Create or retrieve engine with proper settings for SQLite
-        # For in-memory databases, use cached engines to ensure schema persists
+        # Create or retrieve engine with proper settings.
+        # SQLite uses cached engines for in-memory databases; PostgreSQL
+        # (and other non-SQLite backends) use a simple disposable engine
+        # since the migration runner only needs a short-lived connection.
         if self.database_url.startswith("sqlite"):
             if self.database_url == "sqlite:///:memory:":
                 # For in-memory databases, reuse cached engine to ensure
@@ -276,6 +281,7 @@ class MigrationRunner:
                 )
                 should_dispose = True
         else:
+            # PostgreSQL / other RDBMS — no SQLite-specific connect_args
             engine = create_engine(self.database_url)
             should_dispose = True
 
diff --git a/src/cleveragents/infrastructure/database/migrations/env.py b/src/cleveragents/infrastructure/database/migrations/env.py
index 52bd13f0f..44cb4bee3 100644
--- a/src/cleveragents/infrastructure/database/migrations/env.py
+++ b/src/cleveragents/infrastructure/database/migrations/env.py
@@ -77,6 +77,7 @@ def run_migrations_online() -> None:
     In this scenario we need to create an Engine
     and associate a connection with the context.
 
+    Supports both SQLite (local mode) and PostgreSQL (server mode).
     """
     if config is None:
         msg = "Config is required for online migrations"
@@ -94,14 +95,22 @@ def run_migrations_online() -> None:
         )
         # We created the engine, so we should manage the connection lifecycle
         with connectable.connect() as connection:
-            context.configure(connection=connection, target_metadata=target_metadata)
+            context.configure(
+                connection=connection,
+                target_metadata=target_metadata,
+                compare_type=True,
+            )
 
             with context.begin_transaction():
                 context.run_migrations()
     else:
         # Connection was provided externally, use it but don't close it
         # (the caller is responsible for cleanup)
-        context.configure(connection=connectable, target_metadata=target_metadata)
+        context.configure(
+            connection=connectable,
+            target_metadata=target_metadata,
+            compare_type=True,
+        )
 
         with context.begin_transaction():
             context.run_migrations()
diff --git a/src/cleveragents/infrastructure/database/unit_of_work.py b/src/cleveragents/infrastructure/database/unit_of_work.py
index 357bd6570..c61cb24b7 100644
--- a/src/cleveragents/infrastructure/database/unit_of_work.py
+++ b/src/cleveragents/infrastructure/database/unit_of_work.py
@@ -43,6 +43,13 @@ class UnitOfWork:
         database_url: str,
         prompt_for_migration: Callable[[str], bool] | None = None,
         require_confirmation: bool = True,
+        # Pool defaults are intentionally duplicated from Settings.db_pool_*
+        # so that UnitOfWork remains usable without a Settings instance
+        # (e.g. in tests or standalone scripts).  When constructed via the
+        # DI container the caller should forward Settings values explicitly.
+        pool_size: int = 5,
+        max_overflow: int = 10,
+        pool_recycle: int = 1800,
     ) -> None:
         """Initialize Unit of Work with database connection.
 
@@ -53,13 +60,44 @@ class UnitOfWork:
             require_confirmation: When False, migrations are applied without
                 prompting the user.  Set to False when the caller has
                 already obtained blanket approval (e.g. ``--yes`` flag).
+            pool_size: Connection-pool size for PostgreSQL backends.
+            max_overflow: Extra connections above *pool_size* under load.
+            pool_recycle: Seconds before a pooled connection is recycled.
         """
+        # Validate arguments per CONTRIBUTING.md fail-fast principles
+        if not database_url:
+            raise ValueError("database_url must be a non-empty string")
+        if pool_size < 1:
+            raise ValueError(f"pool_size must be >= 1, got {pool_size}")
+        if max_overflow < 0:
+            raise ValueError(f"max_overflow must be >= 0, got {max_overflow}")
+        if pool_recycle < -1:
+            raise ValueError(f"pool_recycle must be >= -1, got {pool_recycle}")
+
         self.database_url = database_url
         self._engine: Engine | None = None
         self._session_factory: sessionmaker[Session] | None = None
         self._database_initialized = False
         self._prompt_for_migration = prompt_for_migration
         self._require_confirmation = require_confirmation
+        self._pool_size = pool_size
+        self._max_overflow = max_overflow
+        self._pool_recycle = pool_recycle
+
+    @property
+    def pool_size(self) -> int:
+        """Read-only access to the configured connection-pool size."""
+        return self._pool_size
+
+    @property
+    def max_overflow(self) -> int:
+        """Read-only access to the configured max overflow."""
+        return self._max_overflow
+
+    @property
+    def pool_recycle(self) -> int:
+        """Read-only access to the configured pool recycle interval."""
+        return self._pool_recycle
 
     @property
     def engine(self) -> Engine:
@@ -93,10 +131,29 @@ class UnitOfWork:
                         connect_args={"check_same_thread": False},
                     )
             else:
+                # PostgreSQL (or other non-SQLite): use connection pooling
+                # suitable for multi-user server mode.
+                # Use READ COMMITTED isolation level for concurrent multi-user access.
+                # This allows non-repeatable reads and phantom reads, which is
+                # acceptable for the concurrent plan execution use case where
+                # individual plan steps are atomic operations. For stronger
+                # consistency, callers can use explicit locking or application-
+                # level coordination.
+                try:
+                    import psycopg2  # noqa: F401
+                except ImportError as exc:
+                    raise ImportError(
+                        "PostgreSQL support requires the 'server' extra: "
+                        "pip install cleveragents[server]"
+                    ) from exc
                 self._engine = create_engine(
                     self.database_url,
                     echo=False,  # Set to True for SQL debugging
                     future=True,  # Use SQLAlchemy 2.0 style
+                    pool_size=self._pool_size,
+                    max_overflow=self._max_overflow,
+                    pool_recycle=self._pool_recycle,
+                    pool_pre_ping=True,
                 )
         return self._engine
 
@@ -136,6 +193,25 @@ class UnitOfWork:
         finally:
             session.close()
 
+    def dispose(self) -> None:
+        """Release all database connections and dispose the engine.
+
+        Must be called when the UnitOfWork is no longer needed, especially
+        for PostgreSQL backends where the connection pool holds TCP connections.
+        """
+        if self._engine is not None and self.database_url != "sqlite:///:memory:":
+            self._engine.dispose()
+            self._engine = None
+            self._session_factory = None
+
+    def __enter__(self) -> UnitOfWork:
+        """Enter context manager."""
+        return self
+
+    def __exit__(self, *args: object) -> None:
+        """Exit context manager and dispose resources."""
+        self.dispose()
+
     def _ensure_database_initialized(self, force: bool = False) -> None:
         """Run migrations if needed before database access."""
         if self._database_initialized and not force: