fix: silent agent errors, stale model list, live model fetching (#377)
* fix: silent errors, stale models, live model fetching (#373, #374, #375) - api/streaming.py: detect empty agent response (_assistant_added check), emit apperror(type='no_response' or 'auth_mismatch') instead of silent done - api/streaming.py: add _token_sent flag so guard works for streaming agents - static/messages.js: done handler belt-and-suspenders guard for zero replies - static/messages.js: apperror handler labels 'no_response' type distinctly - api/config.py: remove gpt-4o and o3 from _FALLBACK_MODELS and _PROVIDER_MODELS['openai'] (superseded by gpt-5.4-mini and o4-mini) - api/routes.py: new /api/models/live?provider= endpoint, fetches /v1/models from provider API with B310 scheme check + SSRF guard - static/ui.js: _fetchLiveModels() background fetch after static list loads, appends new models to dropdown, caches per session, skips unsupported providers Other: - tests/test_issues_373_374_375.py: 25 new structural tests - tests/test_regressions.py: extend done-handler window 1500->2500 chars - CHANGELOG.md: v0.50.19 entry; 947 tests (up from 922) * fix: SSRF hostname bypass + auth detection operator precedence 1. routes.py: SSRF guard used substring matching (any(k in hostname)) which allows bypass via hostnames like evil-ollama.attacker.com. Changed to exact hostname matching against a fixed set of known local hostnames (localhost, 127.0.0.1, 0.0.0.0, ::1). 2. streaming.py: _is_auth detection had a Python operator precedence bug on the ternary expression. The line: 'AuthenticationError' in type(...).__name__ if _last_err else False parsed as the ternary absorbing the rest of the or-chain when _last_err was falsy. Fixed to: (_last_err and 'AuthenticationError' in ...) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> * docs: fix v0.50.20 CHANGELOG version number and test count (949 tests) --------- Co-authored-by: Nathan Esquenazi <nesquena@gmail.com> Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
nesquena-hermes
GitHub
@@ -406,8 +406,6 @@ CLI_TOOLSETS = get_config().get("platform_toolsets", {}).get("cli", _DEFAULT_TOO
|
||||
# Hardcoded fallback models (used when no config.yaml or agent is available)
|
||||
_FALLBACK_MODELS = [
|
||||
{"provider": "OpenAI", "id": "openai/gpt-5.4-mini", "label": "GPT-5.4 Mini"},
|
||||
{"provider": "OpenAI", "id": "openai/gpt-4o", "label": "GPT-4o"},
|
||||
{"provider": "OpenAI", "id": "openai/o3", "label": "o3"},
|
||||
{"provider": "OpenAI", "id": "openai/o4-mini", "label": "o4-mini"},
|
||||
{
|
||||
"provider": "Anthropic",
|
||||
@@ -463,8 +461,6 @@ _PROVIDER_MODELS = {
|
||||
],
|
||||
"openai": [
|
||||
{"id": "gpt-5.4-mini", "label": "GPT-5.4 Mini"},
|
||||
{"id": "gpt-4o", "label": "GPT-4o"},
|
||||
{"id": "o3", "label": "o3"},
|
||||
{"id": "o4-mini", "label": "o4-mini"},
|
||||
],
|
||||
"openai-codex": [
|
||||
|
||||
141
api/routes.py
141
api/routes.py
@@ -341,6 +341,9 @@ def handle_get(handler, parsed) -> bool:
|
||||
if parsed.path == "/api/models":
|
||||
return j(handler, get_available_models())
|
||||
|
||||
if parsed.path == "/api/models/live":
|
||||
return _handle_live_models(handler, parsed)
|
||||
|
||||
if parsed.path == "/api/settings":
|
||||
settings = load_settings()
|
||||
# Never expose the stored password hash to clients
|
||||
@@ -1410,6 +1413,144 @@ def _handle_approval_inject(handler, parsed):
|
||||
return j(handler, {"error": "session_id required"}, status=400)
|
||||
|
||||
|
||||
def _handle_live_models(handler, parsed):
|
||||
"""Fetch the live model list from a provider's /v1/models endpoint.
|
||||
|
||||
Returns the provider's actual model catalog so the UI can show all
|
||||
available models, not just the hardcoded fallback list.
|
||||
|
||||
Query params:
|
||||
provider (optional) — provider ID to fetch for; defaults to active
|
||||
base_url (optional) — override the base URL for the provider
|
||||
|
||||
Providers that don't expose a /v1/models endpoint (Anthropic) are not
|
||||
supported here — the caller should fall back to the static list.
|
||||
|
||||
Supported: openai, openrouter, custom (any OpenAI-compatible endpoint).
|
||||
"""
|
||||
import urllib.request as _ur
|
||||
import ipaddress as _ip
|
||||
import socket as _sock
|
||||
from urllib.parse import urlparse as _up
|
||||
|
||||
qs = parse_qs(parsed.query)
|
||||
provider = (qs.get("provider", [""])[0] or "").lower().strip()
|
||||
base_url_override = (qs.get("base_url", [""])[0] or "").strip()
|
||||
|
||||
try:
|
||||
from api.config import get_config as _gc, resolve_model_provider as _rmp
|
||||
cfg = _gc()
|
||||
active_provider = cfg.get("model", {}).get("provider") or ""
|
||||
if not provider:
|
||||
provider = active_provider
|
||||
|
||||
# Resolve API key and base URL for this provider
|
||||
api_key = None
|
||||
base_url = base_url_override or ""
|
||||
try:
|
||||
from hermes_cli.runtime_provider import resolve_runtime_provider
|
||||
rt = resolve_runtime_provider(requested=provider)
|
||||
api_key = rt.get("api_key")
|
||||
if not base_url:
|
||||
base_url = rt.get("base_url") or ""
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
# Determine the /v1/models endpoint URL
|
||||
if not base_url:
|
||||
if provider in ("openai", "openai-codex", "copilot"):
|
||||
base_url = "https://api.openai.com/v1"
|
||||
elif provider == "openrouter":
|
||||
base_url = "https://openrouter.ai/api/v1"
|
||||
elif provider in ("anthropic",):
|
||||
# Anthropic doesn't support /v1/models in a standard way
|
||||
return j(handler, {"error": "not_supported", "models": []})
|
||||
elif provider in ("google", "gemini"):
|
||||
return j(handler, {"error": "not_supported", "models": []})
|
||||
else:
|
||||
# Generic OpenAI-compatible — try common paths
|
||||
base_url = ""
|
||||
|
||||
if not base_url:
|
||||
return j(handler, {"error": "no_base_url", "models": []})
|
||||
|
||||
# Build URL safely
|
||||
base_url = base_url.rstrip("/")
|
||||
if base_url.endswith("/v1"):
|
||||
endpoint_url = base_url + "/models"
|
||||
elif "/v1" in base_url:
|
||||
endpoint_url = base_url.rstrip("/") + "/models"
|
||||
else:
|
||||
endpoint_url = base_url + "/v1/models"
|
||||
|
||||
# Validate scheme (B310 guard)
|
||||
parsed_ep = _up(endpoint_url)
|
||||
if parsed_ep.scheme not in ("http", "https"):
|
||||
return j(handler, {"error": "invalid_scheme", "models": []}, status=400)
|
||||
|
||||
# SSRF guard: block private IPs (allow known local provider hostnames).
|
||||
# Use exact hostname match — NOT substring — to prevent bypass via
|
||||
# hostnames like evil-ollama.attacker.com containing "ollama".
|
||||
_KNOWN_LOCAL_HOSTS = {"localhost", "127.0.0.1", "0.0.0.0", "::1"}
|
||||
if parsed_ep.hostname:
|
||||
hostname_lower = (parsed_ep.hostname or "").lower()
|
||||
try:
|
||||
for _, _, _, _, addr in _sock.getaddrinfo(parsed_ep.hostname, None):
|
||||
addr_obj = _ip.ip_address(addr[0])
|
||||
if addr_obj.is_private or addr_obj.is_loopback:
|
||||
if hostname_lower not in _KNOWN_LOCAL_HOSTS:
|
||||
return j(handler, {"error": "ssrf_blocked", "models": []}, status=400)
|
||||
except _sock.gaierror:
|
||||
pass
|
||||
|
||||
# Fetch models
|
||||
req = _ur.Request(endpoint_url, method="GET")
|
||||
req.add_header("User-Agent", "HermesWebUI/1.0")
|
||||
if api_key:
|
||||
req.add_header("Authorization", f"Bearer {api_key}")
|
||||
with _ur.urlopen(req, timeout=8) as resp: # nosec B310
|
||||
raw = resp.read().decode("utf-8")
|
||||
|
||||
import json as _json
|
||||
data = _json.loads(raw)
|
||||
raw_models = data.get("data") or data.get("models") or []
|
||||
|
||||
# Normalise to {id, label} list; filter to text-generation models
|
||||
models = []
|
||||
seen = set()
|
||||
for m in raw_models:
|
||||
if not isinstance(m, dict):
|
||||
continue
|
||||
mid = m.get("id") or m.get("name") or ""
|
||||
if not mid or mid in seen:
|
||||
continue
|
||||
# Skip embedding/image/audio models for direct providers
|
||||
obj_type = (m.get("object") or "").lower()
|
||||
if obj_type and obj_type not in ("model",):
|
||||
continue
|
||||
# Heuristic: skip obvious non-chat models
|
||||
if any(skip in mid.lower() for skip in ("embed", "tts", "whisper", "dall-e", "davinci-edit", "babbage", "ada", "curie")):
|
||||
continue
|
||||
seen.add(mid)
|
||||
label = m.get("name") or m.get("display_name") or mid
|
||||
# For OpenAI, the id IS the label — clean it up
|
||||
if label == mid:
|
||||
label = mid.replace("-", " ").replace(".", ".").title()
|
||||
# Restore original casing for well-known names
|
||||
for known in ("GPT", "o1", "o3", "o4", "gpt"):
|
||||
label = label.replace(known.title(), known)
|
||||
models.append({"id": mid, "label": label})
|
||||
|
||||
# Sort: newest (higher version numbers) first via lexicographic sort on reversed id
|
||||
models.sort(key=lambda m: m["id"], reverse=True)
|
||||
|
||||
return j(handler, {"provider": provider, "models": models, "count": len(models)})
|
||||
|
||||
except Exception as _e:
|
||||
logger.debug("Failed to fetch live models for %s: %s", provider, _e)
|
||||
return j(handler, {"error": str(_e), "models": []})
|
||||
|
||||
|
||||
def _handle_cron_output(handler, parsed):
|
||||
from cron.jobs import OUTPUT_DIR as CRON_OUT
|
||||
|
||||
|
||||
@@ -163,9 +163,13 @@ def _run_agent_streaming(session_id, msg_text, model, workspace, stream_id, atta
|
||||
logger.debug("Approval module not available, falling back to polling")
|
||||
|
||||
try:
|
||||
_token_sent = False # tracks whether any streamed tokens were sent
|
||||
|
||||
def on_token(text):
|
||||
nonlocal _token_sent
|
||||
if text is None:
|
||||
return # end-of-stream sentinel
|
||||
_token_sent = True
|
||||
put('token', {'text': text})
|
||||
|
||||
def on_tool(name, preview, args):
|
||||
@@ -308,6 +312,45 @@ def _run_agent_streaming(session_id, msg_text, model, workspace, stream_id, atta
|
||||
)
|
||||
s.messages = result.get('messages') or s.messages
|
||||
|
||||
# ── Detect silent agent failure (no assistant reply produced) ──
|
||||
# When the agent catches an auth/network error internally it may return
|
||||
# an empty final_response without raising — the stream would end with
|
||||
# a done event containing zero assistant messages, leaving the user with
|
||||
# no feedback. Emit an apperror so the client shows an inline error.
|
||||
_assistant_added = any(
|
||||
m.get('role') == 'assistant' and str(m.get('content') or '').strip()
|
||||
for m in (result.get('messages') or [])
|
||||
)
|
||||
# _token_sent tracks whether on_token() was called (any streamed text)
|
||||
if not _assistant_added and not _token_sent:
|
||||
_last_err = getattr(agent, '_last_error', None) or result.get('error') or ''
|
||||
_err_str = str(_last_err) if _last_err else ''
|
||||
_is_auth = (
|
||||
'401' in _err_str
|
||||
or (_last_err and 'AuthenticationError' in type(_last_err).__name__)
|
||||
or 'authentication' in _err_str.lower()
|
||||
or 'unauthorized' in _err_str.lower()
|
||||
or 'invalid api key' in _err_str.lower()
|
||||
or 'invalid_api_key' in _err_str.lower()
|
||||
)
|
||||
if _is_auth:
|
||||
put('apperror', {
|
||||
'message': _err_str or 'Authentication failed — check your API key.',
|
||||
'type': 'auth_mismatch',
|
||||
'hint': (
|
||||
'The selected model may not be supported by your configured provider or '
|
||||
'your API key is invalid. Run `hermes model` in your terminal to '
|
||||
'update credentials, then restart the WebUI.'
|
||||
),
|
||||
})
|
||||
else:
|
||||
put('apperror', {
|
||||
'message': _err_str or 'The agent returned no response. Check your API key and model selection.',
|
||||
'type': 'no_response',
|
||||
'hint': 'Verify your API key is valid and the selected model is available for your account.',
|
||||
})
|
||||
return # Don't emit done — the apperror already closes the stream on the client
|
||||
|
||||
# ── Handle context compression side effects ──
|
||||
# If compression fired inside run_conversation, the agent may have
|
||||
# rotated its session_id. Detect and fix the mismatch so the WebUI
|
||||
|
||||
Reference in New Issue
Block a user