Remove unused backend LLM setting

Author: beacoder

python_agent_harness/agent.py

@@ -452,7 +452,7 @@ def safe_delta(text: str) -> None:
             )
             # sub-agent runs use their own LLM when one is configured
             # (a per-invocation clone of session.subagent_client,
-            # mirroring gptel-agent-harness-subagent-model/-backend);
+            # mirroring gptel-agent-harness-subagent-model);
             # everything unset inherits the main agent's settings, so
             # the sub-agent path is identical when no separate LLM is
             # configured

python_agent_harness/cli.py

@@ -54,8 +54,8 @@ def make_session(
         config_path=config_path,
     )
     # A separate client for sub-agent requests only when a different
-    # LLM is configured (mirrors gptel-agent-harness-subagent-model /
-    # -backend); otherwise the sub-agent shares the main client.
+    # LLM is configured (mirrors gptel-agent-harness-subagent-model);
+    # otherwise the sub-agent shares the main client.
     subagent_client = None
     if any(
         subagent_settings[k] != settings[k] for k in ("base_url", "api_key", "model", "timeout")
@@ -101,7 +101,6 @@ def make_session(
         project_dir=abs_project,
         client=client,
         model=model,
-        backend=settings["backend"],
         system_prompt=system_prompt,
         subagent_system_prompt=subagent_system_prompt,
         temperature=settings["temperature"],
@@ -176,7 +175,7 @@ def cmd_config(args: argparse.Namespace) -> int:
     print(f"config file: {path}")
     if not path.exists():
         print("(file does not exist yet — run `python-agent-harness config --init` to create it)")
-    for key in ("base_url", "model", "backend"):
+    for key in ("base_url", "model"):
         print(f"{key}: {settings[key]}")
     print(f"api_key: {config.mask_secret(settings['api_key'])}")
     print(f"temperature: {settings['temperature']}")

python_agent_harness/config.py

@@ -168,7 +168,6 @@
     "base_url": "https://api.openai.com/v1",
     "api_key": None,
     "model": "gpt-5-mini",
-    "backend": "OpenAI-compatible",
     "temperature": TEMPERATURE,
     "max_tokens": MAX_TOKENS,
     "timeout": 600.0,
@@ -182,8 +181,7 @@
 }
 
 # Sub-agent LLM overrides: every key defaults to None, meaning "inherit
-# the main LLM setting" (mirrors gptel-agent-harness-subagent-model /
-# -backend).  Only the keys the user actually sets differ from the main
+# the main LLM setting" (mirrors gptel-agent-harness-subagent-model).
 # agent's LLM.  ``profile`` references a named profile from the
 # ``models`` section: its settings are applied on top of any explicit
 # subagent_llm keys (profile wins), and unset keys still inherit the
@@ -193,7 +191,6 @@
     "base_url": None,
     "api_key": None,
     "model": None,
-    "backend": None,
     "temperature": None,
     "max_tokens": None,
     "timeout": None,
@@ -264,14 +261,12 @@
     "base_url": "OPENAI_BASE_URL",
     "api_key": "OPENAI_API_KEY",
     "model": "OPENAI_MODEL",
-    "backend": "OPENAI_BACKEND",
 }
 
 _SUBAGENT_ENV_OVERRIDES = {
     "base_url": "OPENAI_SUBAGENT_BASE_URL",
     "api_key": "OPENAI_SUBAGENT_API_KEY",
     "model": "OPENAI_SUBAGENT_MODEL",
-    "backend": "OPENAI_SUBAGENT_BACKEND",
 }
 
 
@@ -320,7 +315,6 @@ def load_llm_config(path: str | os.PathLike | None = None) -> dict:
         "base_url",
         "api_key",
         "model",
-        "backend",
         "temperature",
         "max_tokens",
         "timeout",
@@ -342,7 +336,7 @@ def load_subagent_llm_config(
 ) -> dict:
     """Resolve sub-agent LLM settings; unset keys inherit ``main``.
 
-    Mirrors gptel-agent-harness-subagent-model/-backend: sub-agents
+    Mirrors gptel-agent-harness-subagent-model: sub-agents
     (the Agent tool) use their own LLM when configured, otherwise the
     main agent's.  Precedence: ``main`` settings < config file
     ``subagent_llm`` object < referenced ``models`` profile (when
@@ -379,7 +373,6 @@ def load_subagent_llm_config(
             "base_url",
             "api_key",
             "model",
-            "backend",
             "temperature",
             "max_tokens",
             "timeout",

python_agent_harness/persistence.py

@@ -107,7 +107,6 @@ def __init__(
         self,
         project_dir: str,
         model: str,
-        backend: str,
         system_prompt: str | None = None,
         temperature: float | None = None,
         max_tokens: int | None = None,
@@ -116,7 +115,6 @@ def __init__(
     ) -> None:
         self.project_dir = project_dir
         self.model = model
-        self.backend = backend
         self.system_prompt = system_prompt
         self.temperature = temperature
         self.max_tokens = max_tokens
@@ -173,7 +171,6 @@ def metadata_block(self) -> str:
         pairs = [
             ("python-agent-harness--project-dir", self.project_dir),
             ("gptel-model", self.model),
-            ("gptel--backend-name", self.backend),
             ("gptel-system-prompt", self.system_prompt),
             ("gptel-temperature", self.temperature),
             ("gptel-max-tokens", self.max_tokens),

python_agent_harness/session.py

@@ -71,7 +71,6 @@ def __init__(
         project_dir: str,
         client: Client,
         model: str,
-        backend: str = "OpenAI-compatible",
         system_prompt: str | None = None,
         subagent_system_prompt: str | None = None,
         temperature: float = config.TEMPERATURE,
@@ -96,7 +95,6 @@ def __init__(
         self.project_dir = project_dir
         self.client = client
         self.model = model
-        self.backend = backend
         self.system_prompt = system_prompt
         self.subagent_system_prompt = subagent_system_prompt
         self.temperature = temperature
@@ -113,7 +111,7 @@ def __init__(
         # Sub-agent LLM: a dedicated client (base_url/api_key/model/
         # timeout) and per-request options when a different LLM is
         # configured for sub-agents (mirrors gptel-agent-harness-
-        # subagent-model/-backend); every unset option inherits the
+        # subagent-model); every unset option inherits the
         # main agent's value.  The sub-agent loop never uses this
         # client directly — each Agent tool invocation clones it
         # (see run_subagent) so concurrent sub-agents never share a
@@ -161,7 +159,6 @@ def __init__(
         self.store = SessionPersistence(
             project_dir=project_dir,
             model=model,
-            backend=backend,
             system_prompt=system_prompt,
             temperature=temperature,
             max_tokens=max_tokens,
@@ -659,7 +656,6 @@ def switch_model(self, name: str) -> tuple[bool, str]:
             "base_url": self.client.base_url,
             "api_key": self.client.api_key,
             "model": self.model,
-            "backend": self.backend,
             "temperature": self.temperature,
             "max_tokens": self.max_tokens,
             "timeout": self.client.timeout,
@@ -673,7 +669,6 @@ def switch_model(self, name: str) -> tuple[bool, str]:
                 "base_url",
                 "api_key",
                 "model",
-                "backend",
                 "temperature",
                 "max_tokens",
                 "timeout",
@@ -690,8 +685,6 @@ def switch_model(self, name: str) -> tuple[bool, str]:
         # the calibration factor is tokenizer-specific: a factor tuned
         # to the previous model must not skew estimates for the new one
         self.calibrator.reset()
-        self.backend = merged["backend"]
-        self.store.backend = merged["backend"]
         self.temperature = merged["temperature"]
         self.max_tokens = merged["max_tokens"]
         if hasattr(self.client, "set_timeout"):

python_agent_harness/tools/agent_tool.py

@@ -6,7 +6,7 @@
 long-running sub-agent never blocks the parent's sequential tool loop.
 
 Sub-agents run the same agent loop with a fresh loop instance; their
-backend/model can be overridden (see config).  Results flow back to the
+model can be overridden (see config).  Results flow back to the
 parent as a single tool result string.  Errors are contained: an
 unexpected sub-agent response becomes an error string fed to the parent,
 never a crash.

tests/agent/agent_test_utils.py

@@ -110,7 +110,6 @@ def __init__(self, project_dir="/tmp/fakeproj", model_profiles=None, llm_setting
         self.store = SessionPersistence(
             project_dir=project_dir,
             model=self.model,
-            backend=self.backend,
             system_prompt=self.system_prompt,
             temperature=self.temperature,
             max_tokens=self.max_tokens,

tests/test_cli.py

@@ -20,11 +20,9 @@ class TestMakeSessionPromptDefaults(unittest.TestCase):
         "OPENAI_BASE_URL",
         "OPENAI_API_KEY",
         "OPENAI_MODEL",
-        "OPENAI_BACKEND",
         "OPENAI_SUBAGENT_BASE_URL",
         "OPENAI_SUBAGENT_API_KEY",
         "OPENAI_SUBAGENT_MODEL",
-        "OPENAI_SUBAGENT_BACKEND",
     ]
 
     def setUp(self):
@@ -178,7 +176,7 @@ def test_subagent_stream_inherits_main_without_cli_flag(self):
 
     def test_subagent_llm_inherits_main_client_by_default(self):
         """Without subagent_llm overrides, the sub-agent shares the
-        main client (mirrors gptel-agent-harness: nil backend/model
+        main client (mirrors gptel-agent-harness: nil model overrides
         inherit the main agent's)."""
         session = cli.make_session(self._tmp.name, config_path=self._config_path)
         try:

tests/test_config.py

@@ -11,12 +11,10 @@
     "OPENAI_BASE_URL",
     "OPENAI_API_KEY",
     "OPENAI_MODEL",
-    "OPENAI_BACKEND",
     "PYTHON_AGENT_HARNESS_CONFIG",
     "OPENAI_SUBAGENT_BASE_URL",
     "OPENAI_SUBAGENT_API_KEY",
     "OPENAI_SUBAGENT_MODEL",
-    "OPENAI_SUBAGENT_BACKEND",
 ]
 
 
@@ -49,7 +47,6 @@ def test_file_overrides_defaults(self):
                 '"base_url": "https://api.example.com/v1", '
                 '"api_key": "sk-test", '
                 '"model": "custom-model", '
-                '"backend": "Example", '
                 '"temperature": 0.2, '
                 '"max_tokens": 2048, '
                 '"timeout": 30.0, '
@@ -60,7 +57,6 @@ def test_file_overrides_defaults(self):
             self.assertEqual(settings["base_url"], "https://api.example.com/v1")
             self.assertEqual(settings["api_key"], "sk-test")
             self.assertEqual(settings["model"], "custom-model")
-            self.assertEqual(settings["backend"], "Example")
             self.assertEqual(settings["temperature"], 0.2)
             self.assertEqual(settings["max_tokens"], 2048)
             self.assertEqual(settings["timeout"], 30.0)
@@ -327,7 +323,7 @@ def test_get_context_window_no_file(self):
 
 class TestSubagentLlmConfig(unittest.TestCase):
     """Sub-agent LLM settings: unset keys inherit the main settings
-    (mirrors gptel-agent-harness-subagent-model/-backend)."""
+    (mirrors gptel-agent-harness-subagent-model)."""
 
     def setUp(self):
         self._saved = {k: os.environ.get(k) for k in ENV_KEYS}
@@ -345,7 +341,6 @@ def tearDown(self):
         "base_url": "https://api.main.example/v1",
         "api_key": "sk-main",
         "model": "big-model",
-        "backend": "Main",
         "temperature": 0.0,
         "max_tokens": None,
         "timeout": 600.0,

tests/test_persistence.py

@@ -28,7 +28,6 @@ def test_metadata_roundtrip(self):
         store = SessionPersistence(
             project_dir="/tmp/proj",
             model="deepseek-v4",
-            backend="DeepSeek",
             system_prompt="be helpful",
             temperature=0.7,
             max_tokens=8192,
@@ -47,7 +46,6 @@ def test_round_times_persisted_in_metadata(self):
         store = SessionPersistence(
             project_dir="/tmp/proj",
             model="deepseek-v4",
-            backend="DeepSeek",
             round_times=[1700000000.0, 1700000100.5],
         )
         meta = store.metadata_block()
@@ -59,7 +57,7 @@ def test_round_times_persisted_in_metadata(self):
         self.assertEqual(times, [1700000000.0, 1700000100.5])
 
     def test_round_times_absent_when_empty(self):
-        store = SessionPersistence(project_dir="/tmp/proj", model="m", backend="b")
+        store = SessionPersistence(project_dir="/tmp/proj", model="m")
         self.assertNotIn("round-times", store.metadata_block())
 
     def test_save_and_restore_flow(self):
@@ -70,7 +68,6 @@ def test_save_and_restore_flow(self):
             store = SessionPersistence(
                 project_dir=d,
                 model="m",
-                backend="b",
             )
             path = store.save("hello world")
             self.assertTrue(path)
@@ -88,7 +85,7 @@ def test_apply_title_renames(self):
             old_dir = config.SESSION_DIR
             config.SESSION_DIR = __import__("pathlib").Path(d)
             try:
-                store = SessionPersistence(project_dir="/tmp/proj", model="m", backend="b")
+                store = SessionPersistence(project_dir="/tmp/proj", model="m")
                 store.save("x")
                 store.apply_title("My Great Session")
                 self.assertTrue(os.path.exists(store.file_path))
@@ -110,9 +107,9 @@ def test_same_second_save_collision_keeps_both(self):
             old_dir = config.SESSION_DIR
             config.SESSION_DIR = __import__("pathlib").Path(d)
             try:
-                s1 = SessionPersistence(project_dir="/tmp/proj", model="m", backend="b")
+                s1 = SessionPersistence(project_dir="/tmp/proj", model="m")
                 p1 = s1.save("one")
-                s2 = SessionPersistence(project_dir="/tmp/proj", model="m", backend="b")
+                s2 = SessionPersistence(project_dir="/tmp/proj", model="m")
                 p2 = s2.save("two")
                 self.assertNotEqual(p1, p2)
                 self.assertTrue(os.path.exists(p1))
@@ -136,11 +133,11 @@ def test_title_collision_keeps_both_files(self):
             old_dir = config.SESSION_DIR
             config.SESSION_DIR = __import__("pathlib").Path(d)
             try:
-                s1 = SessionPersistence(project_dir="/tmp/proj", model="m", backend="b")
+                s1 = SessionPersistence(project_dir="/tmp/proj", model="m")
                 s1.save("one")
                 s1.apply_title("Same Title")
                 t1 = s1.file_path
-                s2 = SessionPersistence(project_dir="/tmp/proj", model="m", backend="b")
+                s2 = SessionPersistence(project_dir="/tmp/proj", model="m")
                 s2.save("two")
                 s2.apply_title("Same Title")
                 t2 = s2.file_path
@@ -155,14 +152,14 @@ def test_save_returns_none_without_file_path(self):
         must return None instead of writing anywhere."""
         import unittest.mock as mock
 
-        store = SessionPersistence(project_dir="/tmp/proj", model="m", backend="b")
+        store = SessionPersistence(project_dir="/tmp/proj", model="m")
         with mock.patch.object(SessionPersistence, "session_file", return_value=None):
             self.assertIsNone(store.save("hello"))
 
     def test_apply_title_requires_nonempty_sanitized_title(self):
         """A title that sanitizes to '' must not rename anything and must
         not be recorded."""
-        store = SessionPersistence(project_dir="/tmp/proj", model="m", backend="b")
+        store = SessionPersistence(project_dir="/tmp/proj", model="m")
         store.apply_title("   ---   ")  # sanitizes to ""
         self.assertIsNone(store.title)
         self.assertIsNone(store.file_path)
@@ -180,7 +177,7 @@ def test_apply_title_oserror_keeps_original_file(self):
             old_dir = config.SESSION_DIR
             config.SESSION_DIR = __import__("pathlib").Path(d)
             try:
-                store = SessionPersistence(project_dir="/tmp/proj", model="m", backend="b")
+                store = SessionPersistence(project_dir="/tmp/proj", model="m")
                 store.save("x")
                 before = store.file_path
                 with mock.patch(
@@ -211,7 +208,7 @@ def test_concurrent_save_and_apply_title_do_not_split(self):
             old_dir = config.SESSION_DIR
             config.SESSION_DIR = __import__("pathlib").Path(d)
             try:
-                store = SessionPersistence(project_dir="/tmp/proj", model="m", backend="b")
+                store = SessionPersistence(project_dir="/tmp/proj", model="m")
                 store.save("old content")
                 before = store.file_path
 
@@ -298,9 +295,9 @@ def test_list_sessions_newest_first(self):
             old_dir = config.SESSION_DIR
             config.SESSION_DIR = Path(d)
             try:
-                s1 = SessionPersistence(project_dir="/tmp/proj", model="m", backend="b")
+                s1 = SessionPersistence(project_dir="/tmp/proj", model="m")
                 p1 = s1.save("one")
-                s2 = SessionPersistence(project_dir="/tmp/proj", model="m", backend="b")
+                s2 = SessionPersistence(project_dir="/tmp/proj", model="m")
                 p2 = s2.save("two")
                 os.utime(p1, (1_000_000, 1_000_000))  # make p1 the older file
                 sessions = SessionPersistence.list_sessions()