This commit is contained in:
Hermes Agent
@@ -2,6 +2,9 @@
|
||||
|
||||
## [Unreleased]
|
||||
|
||||
### Fixed
|
||||
- **Local model setup no longer fails mid-conversation with `LOCAL_API_KEY` error** — when `model.base_url` pointed at an OpenAI-compatible loopback endpoint that didn't match the `ollama`/`localhost`/`lmstudio` keyword classifier (e.g. `http://192.168.1.10:8080/v1`, llama.cpp on `127.0.0.1:8080`, vLLM, TabbyAPI, custom proxies), `_build_available_models_uncached` auto-detected the provider as `"local"` and persisted that into `config.yaml`. Inference worked initially because the main agent has its own direct path that uses the explicit `base_url + api_key`. But once the conversation grew enough to trip auto-compression — or when vision / web extraction / skills-hub fired — the agent's auxiliary client routed through `resolve_provider_client("local", …)`, fell through every branch, and raised `Provider 'local' is set in config.yaml but no API key was found. Set the LOCAL_API_KEY environment variable, or switch to a different provider with hermes model.` Three-layer fix: (1) the auto-detect block now writes `provider: "custom"` instead of `provider: "local"` for unknown loopback hosts — `custom` is the canonical OpenAI-compat fall-through and the agent's auxiliary client takes the `no-key-required` path for it; (2) `resolve_model_provider()` rewrites legacy `"local"` to `"custom"` at read time so existing broken configs heal automatically without requiring the user to edit `config.yaml` by hand; (3) `set_hermes_default_model()` refuses to persist `"local"` going forward, and `_PROVIDER_ALIASES` gets a `"local" → "custom"` entry for any consumer that normalises through the alias table. 9 regression tests in `test_issue1384_local_provider.py` covering the source-code invariant, both YAML cases, the migration path, and the alias table. (`api/config.py`, `tests/test_issue1384_local_provider.py`) Closes #1384
|
||||
|
||||
## [v0.50.252] — 2026-05-01
|
||||
|
||||
### Fixed
|
||||
@@ -17,7 +20,6 @@
|
||||
### Changed
|
||||
- **API credential redaction now uses `force=True`** — `_combined_redact` (introduced by #1379) now passes `force=True` to `redact_sensitive_text` so the agent's broader patterns (Stripe `sk_live_`, Google `AIza…`, JWT `eyJ…`, DB connection strings, Telegram bot tokens) run regardless of the user's `HERMES_REDACT_SECRETS` opt-in. The local fallback then handles the short-prefix shapes the agent omits (`ghp_`, `sk-`, `hf_`, `AKIA`). WebUI API responses are a hard safety boundary — no opt-in should be required. (`api/helpers.py`) — Opus pre-release follow-up
|
||||
- **`_active_profile_for_live_models_cache` logs the fallback path** — when `get_active_profile_name()` raises (transient state, mid-switch, etc.) the live-models cache (#1378) falls back to `"default"`, mis-scoping the cache for up to 60s. Now logs at debug so we can detect this in production logs without changing the blast radius (TTL still caps the bad-cache window). (`api/routes.py`) — Opus pre-release follow-up
|
||||
|
||||
## [v0.50.251] — 2026-04-30
|
||||
|
||||
### Fixed
|
||||
|
||||
@@ -614,6 +614,12 @@ _PROVIDER_ALIASES = {
|
||||
"nvidia-nim": "nvidia",
|
||||
"build-nvidia": "nvidia",
|
||||
"nemotron": "nvidia",
|
||||
# Legacy alias — earlier WebUI builds wrote ``provider: local`` for unknown
|
||||
# loopback endpoints, but ``local`` is not registered in
|
||||
# ``hermes_cli.auth.PROVIDER_REGISTRY``. Routing it through ``custom``
|
||||
# lets the agent's auxiliary client take the ``no-key-required``
|
||||
# OpenAI-compat path. See #1384.
|
||||
"local": "custom",
|
||||
}
|
||||
|
||||
|
||||
@@ -971,6 +977,16 @@ def resolve_model_provider(model_id: str) -> tuple:
|
||||
config_provider = model_cfg.get("provider")
|
||||
config_base_url = model_cfg.get("base_url")
|
||||
|
||||
# Heal legacy ``provider: local`` entries (written by WebUI < v0.50.252)
|
||||
# at read time. ``local`` is not a registered provider, so passing it
|
||||
# downstream raises a ``LOCAL_API_KEY`` error from the auxiliary client
|
||||
# mid-conversation when compression/vision/web-extract fires. Route
|
||||
# through ``custom`` instead — it takes the ``no-key-required``
|
||||
# OpenAI-compat path that local servers (Ollama, LM Studio, llama.cpp,
|
||||
# vLLM, TabbyAPI) actually use. See #1384.
|
||||
if isinstance(config_provider, str) and config_provider.strip().lower() == "local":
|
||||
config_provider = "custom"
|
||||
|
||||
model_id = (model_id or "").strip()
|
||||
if not model_id:
|
||||
return model_id, config_provider, config_base_url
|
||||
@@ -1185,6 +1201,13 @@ def set_hermes_default_model(model_id: str) -> dict:
|
||||
# matcher — see `static/panels.js` (#895).
|
||||
persisted_model = str(resolved_model or selected_model).strip()
|
||||
persisted_provider = str(resolved_provider or previous_provider or "").strip()
|
||||
# Never persist the bogus ``local`` value — see #1384. The auto-detect
|
||||
# block in ``_build_available_models_uncached`` was rewriting unknown
|
||||
# loopback hosts to ``provider: "local"``, which is not registered and
|
||||
# broke compression/vision mid-conversation. Route through ``custom``
|
||||
# so the agent's auxiliary client uses the ``no-key-required`` path.
|
||||
if persisted_provider.lower() == "local":
|
||||
persisted_provider = "custom"
|
||||
|
||||
model_cfg["default"] = persisted_model
|
||||
if persisted_provider:
|
||||
@@ -1745,7 +1768,16 @@ def get_available_models() -> dict:
|
||||
elif "lmstudio" in host or "lm-studio" in host:
|
||||
provider = "lmstudio"
|
||||
else:
|
||||
provider = "local"
|
||||
# Unknown loopback/private endpoint: route through
|
||||
# the generic ``custom`` provider so the agent's
|
||||
# auxiliary client (compression, vision, web
|
||||
# extraction) takes the OpenAI-compat custom path
|
||||
# with ``no-key-required`` semantics. Writing
|
||||
# ``provider: local`` here used to break
|
||||
# compression mid-conversation because ``local``
|
||||
# is not a registered provider in
|
||||
# ``hermes_cli.auth.PROVIDER_REGISTRY`` — see #1384.
|
||||
provider = "custom"
|
||||
except ValueError:
|
||||
pass
|
||||
|
||||
|
||||
194
tests/test_issue1384_local_provider.py
Normal file
194
tests/test_issue1384_local_provider.py
Normal file
@@ -0,0 +1,194 @@
|
||||
"""Regression tests for #1384 — ``provider: "local"`` mid-conversation crash.
|
||||
|
||||
Earlier WebUI builds auto-detected unknown loopback hosts and persisted
|
||||
``provider: "local"`` to ``config.yaml``. That value is not a registered
|
||||
provider in ``hermes_cli.auth.PROVIDER_REGISTRY``, so the agent's auxiliary
|
||||
client (compression, vision, web extraction) raised
|
||||
``"Provider 'local' is set in config.yaml but no API key was found"``
|
||||
mid-conversation when the context-compression threshold was hit.
|
||||
|
||||
The fix is three layers deep so a user with an already-broken config gets
|
||||
healed automatically:
|
||||
|
||||
1. The auto-detect block now writes ``provider: "custom"`` for unknown
|
||||
loopback hosts (``custom`` is the canonical generic-OpenAI-compat
|
||||
provider, and the agent's auxiliary client takes the
|
||||
``no-key-required`` path for it).
|
||||
2. ``resolve_model_provider()`` rewrites legacy ``"local"`` to ``"custom"``
|
||||
at read time, so existing configs route correctly without requiring the
|
||||
user to edit ``config.yaml`` by hand.
|
||||
3. ``set_hermes_default_model()`` refuses to persist ``"local"`` going
|
||||
forward, so any other code path that tries to write it is also healed.
|
||||
4. The local alias table has ``"local" → "custom"`` for any consumer that
|
||||
normalises through ``_resolve_provider_alias``.
|
||||
"""
|
||||
|
||||
import re
|
||||
from pathlib import Path
|
||||
|
||||
import pytest
|
||||
|
||||
import api.config as cfg
|
||||
|
||||
|
||||
# ── 1. Auto-detect block writes ``custom``, not ``local`` ────────────────
|
||||
|
||||
|
||||
class TestAutoDetectWritesCustom:
|
||||
"""The auto-detect branch in ``_build_available_models_uncached`` must
|
||||
never write ``provider = "local"`` because that value breaks the
|
||||
auxiliary client mid-conversation."""
|
||||
|
||||
def test_source_code_no_local_assignment(self):
|
||||
"""The string ``provider = "local"`` must not appear in api/config.py."""
|
||||
src = Path(cfg.__file__).read_text(encoding="utf-8")
|
||||
assert 'provider = "local"' not in src, (
|
||||
'api/config.py must not assign provider = "local" — see #1384. '
|
||||
"Use ``custom`` instead so the agent's auxiliary client takes the "
|
||||
"``no-key-required`` OpenAI-compat path."
|
||||
)
|
||||
|
||||
def test_auto_detect_branch_uses_custom(self):
|
||||
"""The else-branch in the auto-detect block resolves to ``custom``."""
|
||||
src = Path(cfg.__file__).read_text(encoding="utf-8")
|
||||
# Find the auto-detect block (host-keyword classifier).
|
||||
m = re.search(
|
||||
r'if "ollama" in host or "127\.0\.0\.1" in host or "localhost" in host:\s*\n'
|
||||
r'\s*provider = "ollama"\s*\n'
|
||||
r'\s*elif "lmstudio" in host or "lm-studio" in host:\s*\n'
|
||||
r'\s*provider = "lmstudio"\s*\n'
|
||||
r'\s*else:',
|
||||
src,
|
||||
)
|
||||
assert m, "Auto-detect host-classifier block not found in api/config.py"
|
||||
# Find the next provider assignment after the else.
|
||||
tail = src[m.end() : m.end() + 1500]
|
||||
provider_assign = re.search(r'provider = "([a-z-]+)"', tail)
|
||||
assert provider_assign, "No provider assignment found after auto-detect else"
|
||||
assert provider_assign.group(1) == "custom", (
|
||||
f"Auto-detect else branch must assign provider = \"custom\", "
|
||||
f"got {provider_assign.group(1)!r}"
|
||||
)
|
||||
|
||||
|
||||
# ── 2. resolve_model_provider() heals legacy ``local`` configs ───────────
|
||||
|
||||
|
||||
class TestResolveModelProviderHealsLegacyLocal:
|
||||
"""Existing ``config.yaml`` files with ``provider: local`` (written by
|
||||
earlier WebUI builds) must be normalised to ``custom`` at read time so
|
||||
downstream agent calls take the working path."""
|
||||
|
||||
def test_provider_local_normalised_to_custom(self, tmp_path, monkeypatch):
|
||||
cfgfile = tmp_path / "config.yaml"
|
||||
cfgfile.write_text(
|
||||
"model:\n"
|
||||
" default: qwen2.5-coder:14b\n"
|
||||
" provider: local\n"
|
||||
" base_url: http://127.0.0.1:11434/v1\n",
|
||||
encoding="utf-8",
|
||||
)
|
||||
monkeypatch.setattr(cfg, "_get_config_path", lambda: cfgfile)
|
||||
cfg.reload_config()
|
||||
try:
|
||||
model_id, provider, base_url = cfg.resolve_model_provider("qwen2.5-coder:14b")
|
||||
assert provider == "custom", (
|
||||
f"resolve_model_provider must rewrite legacy 'local' to 'custom', "
|
||||
f"got {provider!r}"
|
||||
)
|
||||
assert base_url == "http://127.0.0.1:11434/v1"
|
||||
assert model_id == "qwen2.5-coder:14b"
|
||||
finally:
|
||||
cfg.reload_config()
|
||||
|
||||
def test_provider_local_uppercase_also_normalised(self, tmp_path, monkeypatch):
|
||||
"""Case-insensitive match — YAML may have ``Local`` or ``LOCAL``."""
|
||||
cfgfile = tmp_path / "config.yaml"
|
||||
cfgfile.write_text(
|
||||
"model:\n"
|
||||
" default: my-model\n"
|
||||
" provider: Local\n"
|
||||
" base_url: http://192.168.1.10:8000/v1\n",
|
||||
encoding="utf-8",
|
||||
)
|
||||
monkeypatch.setattr(cfg, "_get_config_path", lambda: cfgfile)
|
||||
cfg.reload_config()
|
||||
try:
|
||||
_, provider, _ = cfg.resolve_model_provider("my-model")
|
||||
assert provider == "custom"
|
||||
finally:
|
||||
cfg.reload_config()
|
||||
|
||||
def test_other_providers_pass_through_unchanged(self, tmp_path, monkeypatch):
|
||||
"""The migration must not touch any other provider name."""
|
||||
cfgfile = tmp_path / "config.yaml"
|
||||
cfgfile.write_text(
|
||||
"model:\n"
|
||||
" default: claude-sonnet-4.6\n"
|
||||
" provider: anthropic\n",
|
||||
encoding="utf-8",
|
||||
)
|
||||
monkeypatch.setattr(cfg, "_get_config_path", lambda: cfgfile)
|
||||
cfg.reload_config()
|
||||
try:
|
||||
_, provider, _ = cfg.resolve_model_provider("claude-sonnet-4.6")
|
||||
assert provider == "anthropic"
|
||||
finally:
|
||||
cfg.reload_config()
|
||||
|
||||
|
||||
# ── 3. set_hermes_default_model never persists 'local' ───────────────────
|
||||
|
||||
|
||||
class TestSetHermesDefaultModelNeverPersistsLocal:
|
||||
"""Even if a caller (or a stale resolver) hands us ``provider='local'``,
|
||||
we must not write that value back to ``config.yaml``."""
|
||||
|
||||
def test_existing_local_provider_is_replaced_on_save(self, tmp_path, monkeypatch):
|
||||
"""Writing a new default model when previous_provider is 'local'
|
||||
must persist 'custom' instead, because previous_provider falls
|
||||
through into persisted_provider when resolve_model_provider doesn't
|
||||
return a new provider hint."""
|
||||
import yaml
|
||||
|
||||
cfgfile = tmp_path / "config.yaml"
|
||||
cfgfile.write_text(
|
||||
"model:\n"
|
||||
" default: old-model\n"
|
||||
" provider: local\n"
|
||||
" base_url: http://127.0.0.1:11434/v1\n",
|
||||
encoding="utf-8",
|
||||
)
|
||||
monkeypatch.setattr(cfg, "_get_config_path", lambda: cfgfile)
|
||||
cfg.reload_config()
|
||||
try:
|
||||
cfg.set_hermes_default_model("qwen2.5-coder:14b")
|
||||
saved = yaml.safe_load(cfgfile.read_text(encoding="utf-8"))
|
||||
persisted = saved.get("model", {}).get("provider", "")
|
||||
assert persisted != "local", (
|
||||
f"set_hermes_default_model must rewrite 'local' on save — "
|
||||
f"got {persisted!r}"
|
||||
)
|
||||
assert persisted == "custom"
|
||||
finally:
|
||||
cfg.reload_config()
|
||||
|
||||
|
||||
# ── 4. Alias table has the entry ─────────────────────────────────────────
|
||||
|
||||
|
||||
class TestAliasTableHasLocalEntry:
|
||||
"""``_resolve_provider_alias`` must rewrite ``local`` → ``custom`` for
|
||||
any other code path that normalises through the alias table."""
|
||||
|
||||
def test_local_alias_resolves_to_custom(self):
|
||||
assert cfg._resolve_provider_alias("local") == "custom"
|
||||
|
||||
def test_local_alias_case_insensitive(self):
|
||||
assert cfg._resolve_provider_alias("LOCAL") == "custom"
|
||||
assert cfg._resolve_provider_alias("Local") == "custom"
|
||||
|
||||
def test_alias_table_contains_local_entry(self):
|
||||
assert cfg._PROVIDER_ALIASES.get("local") == "custom", (
|
||||
"_PROVIDER_ALIASES must map 'local' → 'custom' for #1384"
|
||||
)
|
||||
Reference in New Issue
Block a user