fix: increase stream read timeout default to 120s, auto-raise for local LLMs (#6967)

Raise the default httpx stream read timeout from 60s to 120s for all
providers. Additionally, auto-detect local LLM endpoints (Ollama,
llama.cpp, vLLM) and raise the read timeout to HERMES_API_TIMEOUT
(1800s) since local models can take minutes for prefill on large
contexts before producing the first token.

The stale stream timeout already had this local auto-detection pattern;
the httpx read timeout was missing it — causing a hard 60s wall that
users couldn't find (HERMES_STREAM_READ_TIMEOUT was undocumented).

Changes:
- Default HERMES_STREAM_READ_TIMEOUT: 60s -> 120s
- Auto-detect local endpoints -> raise to 1800s (user override respected)
- Document HERMES_STREAM_READ_TIMEOUT and HERMES_STREAM_STALE_TIMEOUT
- Add 10 parametrized tests

Reported-by: Pavan Srinivas (@pavanandums)

run_agent.py

@@ -4433,7 +4433,17 @@ class AIAgent:
             """Stream a chat completions response."""
             import httpx as _httpx
             _base_timeout = float(os.getenv("HERMES_API_TIMEOUT", 1800.0))
-            _stream_read_timeout = float(os.getenv("HERMES_STREAM_READ_TIMEOUT", 60.0))
+            _stream_read_timeout = float(os.getenv("HERMES_STREAM_READ_TIMEOUT", 120.0))
+            # Local providers (Ollama, llama.cpp, vLLM) can take minutes for
+            # prefill on large contexts before producing the first token.
+            # Auto-increase the httpx read timeout unless the user explicitly
+            # overrode HERMES_STREAM_READ_TIMEOUT.
+            if _stream_read_timeout == 120.0 and self.base_url and is_local_endpoint(self.base_url):
+                _stream_read_timeout = _base_timeout
+                logger.debug(
+                    "Local provider detected (%s) — stream read timeout raised to %.0fs",
+                    self.base_url, _stream_read_timeout,
+                )
             stream_kwargs = {
                 **api_kwargs,
                 "stream": True,

tests/agent/test_local_stream_timeout.py

@@ -0,0 +1,70 @@
+"""Tests for local provider stream read timeout auto-detection.
+
+When a local LLM provider is detected (Ollama, llama.cpp, vLLM, etc.),
+the httpx stream read timeout should be automatically increased from the
+default 60s to HERMES_API_TIMEOUT (1800s) to avoid premature connection
+kills during long prefill phases.
+"""
+
+import os
+import pytest
+from unittest.mock import patch
+
+from agent.model_metadata import is_local_endpoint
+
+
+class TestLocalStreamReadTimeout:
+    """Verify stream read timeout auto-detection logic."""
+
+    @pytest.mark.parametrize("base_url", [
+        "http://localhost:11434",
+        "http://127.0.0.1:8080",
+        "http://0.0.0.0:5000",
+        "http://192.168.1.100:8000",
+        "http://10.0.0.5:1234",
+    ])
+    def test_local_endpoint_bumps_read_timeout(self, base_url):
+        """Local endpoint + default timeout -> bumps to base_timeout."""
+        with patch.dict(os.environ, {}, clear=False):
+            os.environ.pop("HERMES_STREAM_READ_TIMEOUT", None)
+            _base_timeout = float(os.getenv("HERMES_API_TIMEOUT", 1800.0))
+            _stream_read_timeout = float(os.getenv("HERMES_STREAM_READ_TIMEOUT", 120.0))
+            if _stream_read_timeout == 120.0 and base_url and is_local_endpoint(base_url):
+                _stream_read_timeout = _base_timeout
+            assert _stream_read_timeout == 1800.0
+
+    def test_user_override_respected_for_local(self):
+        """User sets HERMES_STREAM_READ_TIMEOUT -> keep their value even for local."""
+        with patch.dict(os.environ, {"HERMES_STREAM_READ_TIMEOUT": "300"}, clear=False):
+            _base_timeout = float(os.getenv("HERMES_API_TIMEOUT", 1800.0))
+            _stream_read_timeout = float(os.getenv("HERMES_STREAM_READ_TIMEOUT", 120.0))
+            base_url = "http://localhost:11434"
+            if _stream_read_timeout == 120.0 and base_url and is_local_endpoint(base_url):
+                _stream_read_timeout = _base_timeout
+            assert _stream_read_timeout == 300.0
+
+    @pytest.mark.parametrize("base_url", [
+        "https://api.openai.com",
+        "https://openrouter.ai/api",
+        "https://api.anthropic.com",
+    ])
+    def test_remote_endpoint_keeps_default(self, base_url):
+        """Remote endpoint -> keep 120s default."""
+        with patch.dict(os.environ, {}, clear=False):
+            os.environ.pop("HERMES_STREAM_READ_TIMEOUT", None)
+            _base_timeout = float(os.getenv("HERMES_API_TIMEOUT", 1800.0))
+            _stream_read_timeout = float(os.getenv("HERMES_STREAM_READ_TIMEOUT", 120.0))
+            if _stream_read_timeout == 120.0 and base_url and is_local_endpoint(base_url):
+                _stream_read_timeout = _base_timeout
+            assert _stream_read_timeout == 120.0
+
+    def test_empty_base_url_keeps_default(self):
+        """No base_url set -> keep 120s default."""
+        with patch.dict(os.environ, {}, clear=False):
+            os.environ.pop("HERMES_STREAM_READ_TIMEOUT", None)
+            _base_timeout = float(os.getenv("HERMES_API_TIMEOUT", 1800.0))
+            _stream_read_timeout = float(os.getenv("HERMES_STREAM_READ_TIMEOUT", 120.0))
+            base_url = ""
+            if _stream_read_timeout == 120.0 and base_url and is_local_endpoint(base_url):
+                _stream_read_timeout = _base_timeout
+            assert _stream_read_timeout == 120.0

website/docs/reference/environment-variables.md

@@ -278,6 +278,8 @@ For cloud sandbox backends, persistence is filesystem-oriented. `TERMINAL_LIFETI
 | `HERMES_HUMAN_DELAY_MAX_MS` | Custom delay range maximum (ms) |
 | `HERMES_QUIET` | Suppress non-essential output (`true`/`false`) |
 | `HERMES_API_TIMEOUT` | LLM API call timeout in seconds (default: `1800`) |
+| `HERMES_STREAM_READ_TIMEOUT` | Streaming socket read timeout in seconds (default: `120`). Auto-increased to `HERMES_API_TIMEOUT` for local providers. Increase if local LLMs time out during long code generation. |
+| `HERMES_STREAM_STALE_TIMEOUT` | Stale stream detection timeout in seconds (default: `180`). Auto-disabled for local providers. Triggers connection kill if no chunks arrive within this window. |
 | `HERMES_EXEC_ASK` | Enable execution approval prompts in gateway mode (`true`/`false`) |
 | `HERMES_ENABLE_PROJECT_PLUGINS` | Enable auto-discovery of repo-local plugins from `./.hermes/plugins/` (`true`/`false`, default: `false`) |
 | `HERMES_BACKGROUND_NOTIFICATIONS` | Background process notification mode in gateway: `all` (default), `result`, `error`, `off` |