feat(hermes model): add Configure auxiliary models UI to hermes model (#11891)

Previously users had to hand-edit config.yaml to route individual auxiliary
tasks (vision, compression, web_extract, etc.) to a specific provider+model.
Add a first-class picker reachable from the bottom of the existing `hermes
model` provider list.

Flow:
  hermes model
    → Configure auxiliary models...
      → <task picker: 9 tasks, shows current setting inline>
        → <provider picker: authenticated providers + auto + custom>
          → <model picker: curated list + live pricing>

The aux picker does NOT re-run credential/OAuth setup; users authenticate
providers through the normal `hermes model` flow, then route aux tasks to
them here.  `list_authenticated_providers()` gates the list to providers
the user has configured.

Also:
  - 'Cancel' entry relabeled 'Leave unchanged' (sentinel still 'cancel'
    internally, so dispatch logic is unchanged)
  - 'Reset all to auto' entry to bulk-clear aux overrides; preserves
    user-tuned timeout / download_timeout values
  - Adds `title_generation` task to DEFAULT_CONFIG.auxiliary — the task
    was called from agent/title_generator.py but was missing from defaults,
    so config-backed timeout overrides never worked for it

Co-authored-by: teknium1 <teknium@nousresearch.com>
This commit is contained in:
Teknium 2026-04-17 19:02:06 -07:00 committed by GitHub
parent bb85404b16
commit 8444f66890
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
3 changed files with 629 additions and 1 deletions

View file

@ -539,6 +539,13 @@ DEFAULT_CONFIG = {
"api_key": "",
"timeout": 30,
},
"title_generation": {
"provider": "auto",
"model": "",
"base_url": "",
"api_key": "",
"timeout": 30,
},
},
"display": {

View file

@ -1467,7 +1467,8 @@ def select_provider_and_model(args=None):
)
if _has_saved_custom_list:
ordered.append(("remove-custom", "Remove a saved custom provider"))
ordered.append(("cancel", "Cancel"))
ordered.append(("aux-config", "Configure auxiliary models..."))
ordered.append(("cancel", "Leave unchanged"))
provider_idx = _prompt_provider_choice(
[label for _, label in ordered],
@ -1479,6 +1480,10 @@ def select_provider_and_model(args=None):
selected_provider = ordered[provider_idx][0]
if selected_provider == "aux-config":
_aux_config_menu()
return
# Step 2: Provider-specific setup + model selection
if selected_provider == "openrouter":
_model_flow_openrouter(config, current_model)
@ -1579,6 +1584,328 @@ def _clear_stale_openai_base_url():
)
# ─────────────────────────────────────────────────────────────────────────────
# Auxiliary model configuration
#
# Hermes uses lightweight "auxiliary" models for side tasks (vision analysis,
# context compression, web extraction, session search, etc.). Each task has
# its own provider+model pair in config.yaml under `auxiliary.<task>`.
#
# The UI lives behind "Configure auxiliary models..." at the bottom of the
# `hermes model` provider picker. It does NOT re-run credential setup — it
# only routes already-authenticated providers to specific aux tasks. Users
# configure new providers through the normal `hermes model` flow first.
# ─────────────────────────────────────────────────────────────────────────────
# (task_key, display_name, short_description)
_AUX_TASKS: list[tuple[str, str, str]] = [
("vision", "Vision", "image/screenshot analysis"),
("compression", "Compression", "context summarization"),
("web_extract", "Web extract", "web page summarization"),
("session_search", "Session search", "past-conversation recall"),
("approval", "Approval", "smart command approval"),
("mcp", "MCP", "MCP tool reasoning"),
("flush_memories", "Flush memories", "memory consolidation"),
("title_generation", "Title generation", "session titles"),
("skills_hub", "Skills hub", "skills search/install"),
]
def _format_aux_current(task_cfg: dict) -> str:
"""Render the current aux config for display in the task menu."""
if not isinstance(task_cfg, dict):
return "auto"
base_url = str(task_cfg.get("base_url") or "").strip()
provider = str(task_cfg.get("provider") or "auto").strip() or "auto"
model = str(task_cfg.get("model") or "").strip()
if base_url:
short = base_url.replace("https://", "").replace("http://", "").rstrip("/")
return f"custom ({short})" + (f" · {model}" if model else "")
if provider == "auto":
return "auto" + (f" · {model}" if model else "")
if model:
return f"{provider} · {model}"
return provider
def _save_aux_choice(
task: str,
*,
provider: str,
model: str = "",
base_url: str = "",
api_key: str = "",
) -> None:
"""Persist an auxiliary task's provider/model to config.yaml.
Only writes the four routing fields timeout, download_timeout, and any
other task-specific settings are preserved untouched. The main model
config (``model.default``/``model.provider``) is never modified.
"""
from hermes_cli.config import load_config, save_config
cfg = load_config()
aux = cfg.setdefault("auxiliary", {})
if not isinstance(aux, dict):
aux = {}
cfg["auxiliary"] = aux
entry = aux.setdefault(task, {})
if not isinstance(entry, dict):
entry = {}
aux[task] = entry
entry["provider"] = provider
entry["model"] = model or ""
entry["base_url"] = base_url or ""
entry["api_key"] = api_key or ""
save_config(cfg)
def _reset_aux_to_auto() -> int:
"""Reset every known aux task back to auto/empty. Returns number reset."""
from hermes_cli.config import load_config, save_config
cfg = load_config()
aux = cfg.setdefault("auxiliary", {})
if not isinstance(aux, dict):
aux = {}
cfg["auxiliary"] = aux
count = 0
for task, _name, _desc in _AUX_TASKS:
entry = aux.setdefault(task, {})
if not isinstance(entry, dict):
entry = {}
aux[task] = entry
changed = False
if entry.get("provider") not in (None, "", "auto"):
entry["provider"] = "auto"
changed = True
for field in ("model", "base_url", "api_key"):
if entry.get(field):
entry[field] = ""
changed = True
# Preserve timeout/download_timeout — those are user-tuned, not routing
if changed:
count += 1
save_config(cfg)
return count
def _aux_config_menu() -> None:
"""Top-level auxiliary-model picker — choose a task to configure.
Loops until the user picks "Back" so multiple tasks can be configured
without returning to the main provider menu.
"""
from hermes_cli.config import load_config
while True:
cfg = load_config()
aux = cfg.get("auxiliary", {}) if isinstance(cfg.get("auxiliary"), dict) else {}
print()
print(" Auxiliary models — side-task routing")
print()
print(" Hermes uses small, fast models for vision, compression, web")
print(" extraction, and other side tasks. \"auto\" lets Hermes pick the")
print(" best available backend automatically (OpenRouter → Nous Portal")
print(" → your main provider). You rarely need to change these —")
print(" override only if you want a specific model for a task.")
print()
# Build the task menu with current settings inline
name_col = max(len(name) for _, name, _ in _AUX_TASKS) + 2
desc_col = max(len(desc) for _, _, desc in _AUX_TASKS) + 4
entries: list[tuple[str, str]] = []
for task_key, name, desc in _AUX_TASKS:
task_cfg = aux.get(task_key, {}) if isinstance(aux.get(task_key), dict) else {}
current = _format_aux_current(task_cfg)
label = f"{name.ljust(name_col)}{('(' + desc + ')').ljust(desc_col)}{current}"
entries.append((task_key, label))
entries.append(("__reset__", "Reset all to auto"))
entries.append(("__back__", "Back"))
idx = _prompt_provider_choice(
[label for _, label in entries], default=0,
)
if idx is None:
return
key = entries[idx][0]
if key == "__back__":
return
if key == "__reset__":
n = _reset_aux_to_auto()
if n:
print(f"Reset {n} auxiliary task(s) to auto.")
else:
print("All auxiliary tasks were already set to auto.")
print()
continue
# Otherwise configure the specific task
_aux_select_for_task(key)
def _aux_select_for_task(task: str) -> None:
"""Pick a provider + model for a single auxiliary task and persist it.
Uses ``list_authenticated_providers()`` to only show providers the user
has already configured. This avoids re-running OAuth/credential flows
inside the aux picker users set up new providers through the normal
``hermes model`` flow, then route aux tasks to them here.
"""
from hermes_cli.config import load_config
from hermes_cli.model_switch import list_authenticated_providers
cfg = load_config()
aux = cfg.get("auxiliary", {}) if isinstance(cfg.get("auxiliary"), dict) else {}
task_cfg = aux.get(task, {}) if isinstance(aux.get(task), dict) else {}
current_provider = str(task_cfg.get("provider") or "auto").strip() or "auto"
current_model = str(task_cfg.get("model") or "").strip()
current_base_url = str(task_cfg.get("base_url") or "").strip()
display_name = next((name for key, name, _ in _AUX_TASKS if key == task), task)
# Gather authenticated providers (has credentials + curated model list)
try:
providers = list_authenticated_providers(current_provider=current_provider)
except Exception as exc:
print(f"Could not detect authenticated providers: {exc}")
providers = []
entries: list[tuple[str, str, list[str]]] = [] # (slug, label, models)
# "auto" always first
auto_marker = " ← current" if current_provider == "auto" and not current_base_url else ""
entries.append(("__auto__", f"auto (recommended){auto_marker}", []))
for p in providers:
slug = p.get("slug", "")
name = p.get("name") or slug
total = p.get("total_models", 0)
models = p.get("models") or []
model_hint = f"{total} models" if total else ""
marker = " ← current" if slug == current_provider and not current_base_url else ""
entries.append((slug, f"{name}{model_hint}{marker}", list(models)))
# Custom endpoint (raw base_url)
custom_marker = " ← current" if current_base_url else ""
entries.append(("__custom__", f"Custom endpoint (direct URL){custom_marker}", []))
entries.append(("__back__", "Back", []))
print()
print(f" Configure {display_name} — current: {_format_aux_current(task_cfg)}")
print()
idx = _prompt_provider_choice([label for _, label, _ in entries], default=0)
if idx is None:
return
slug, _label, models = entries[idx]
if slug == "__back__":
return
if slug == "__auto__":
_save_aux_choice(task, provider="auto", model="", base_url="", api_key="")
print(f"{display_name}: reset to auto.")
return
if slug == "__custom__":
_aux_flow_custom_endpoint(task, task_cfg)
return
# Regular provider — pick a model from its curated list
_aux_flow_provider_model(task, slug, models, current_model)
def _aux_flow_provider_model(
task: str,
provider_slug: str,
curated_models: list,
current_model: str = "",
) -> None:
"""Prompt for a model under an already-authenticated provider, save to aux."""
from hermes_cli.auth import _prompt_model_selection
from hermes_cli.models import get_pricing_for_provider
display_name = next((name for key, name, _ in _AUX_TASKS if key == task), task)
# Fetch live pricing for this provider (non-blocking)
pricing: dict = {}
try:
pricing = get_pricing_for_provider(provider_slug) or {}
except Exception:
pricing = {}
model_list = list(curated_models)
# Let the user pick a model. _prompt_model_selection supports "Enter custom
# model name" and cancel. When there's no curated list (rare), fall back
# to a raw input prompt.
if not model_list:
print(f"No curated model list for {provider_slug}.")
print("Enter a model slug manually (blank = use provider default):")
try:
val = input("Model: ").strip()
except (KeyboardInterrupt, EOFError):
print()
return
selected = val or ""
else:
selected = _prompt_model_selection(
model_list, current_model=current_model, pricing=pricing,
)
if selected is None:
print("No change.")
return
_save_aux_choice(task, provider=provider_slug, model=selected or "",
base_url="", api_key="")
if selected:
print(f"{display_name}: {provider_slug} · {selected}")
else:
print(f"{display_name}: {provider_slug} (provider default model)")
def _aux_flow_custom_endpoint(task: str, task_cfg: dict) -> None:
"""Prompt for a direct OpenAI-compatible base_url + optional api_key/model."""
import getpass
display_name = next((name for key, name, _ in _AUX_TASKS if key == task), task)
current_base_url = str(task_cfg.get("base_url") or "").strip()
current_model = str(task_cfg.get("model") or "").strip()
print()
print(f" Custom endpoint for {display_name}")
print(" Provide an OpenAI-compatible base URL (e.g. http://localhost:11434/v1)")
print()
try:
url_prompt = f"Base URL [{current_base_url}]: " if current_base_url else "Base URL: "
url = input(url_prompt).strip()
except (KeyboardInterrupt, EOFError):
print()
return
url = url or current_base_url
if not url:
print("No URL provided. No change.")
return
try:
model_prompt = f"Model slug (optional) [{current_model}]: " if current_model else "Model slug (optional): "
model = input(model_prompt).strip()
except (KeyboardInterrupt, EOFError):
print()
return
model = model or current_model
try:
api_key = getpass.getpass("API key (optional, blank = use OPENAI_API_KEY): ").strip()
except (KeyboardInterrupt, EOFError):
print()
return
_save_aux_choice(
task, provider="custom", model=model, base_url=url, api_key=api_key,
)
short_url = url.replace("https://", "").replace("http://", "").rstrip("/")
print(f"{display_name}: custom ({short_url})" + (f" · {model}" if model else ""))
def _prompt_provider_choice(choices, *, default=0):
"""Show provider selection menu with curses arrow-key navigation.

View file

@ -0,0 +1,294 @@
"""Tests for the auxiliary-model configuration UI in ``hermes model``.
Covers the helper functions:
- ``_save_aux_choice`` writes to config.yaml without touching main model config
- ``_reset_aux_to_auto`` clears routing fields but preserves timeouts
- ``_format_aux_current`` renders current task config for the menu
- ``_AUX_TASKS`` stays in sync with ``DEFAULT_CONFIG["auxiliary"]``
These are pure-function tests the interactive menu loops are not covered
here (they're stdin-driven curses prompts).
"""
from __future__ import annotations
import pytest
from hermes_cli.config import DEFAULT_CONFIG, load_config
from hermes_cli.main import (
_AUX_TASKS,
_format_aux_current,
_reset_aux_to_auto,
_save_aux_choice,
)
# ── Default config ──────────────────────────────────────────────────────────
def test_title_generation_present_in_default_config():
"""`title_generation` task must be defined in DEFAULT_CONFIG.
Regression for an existing gap: title_generator.py calls
``call_llm(task="title_generation", ...)`` but the task was missing
from DEFAULT_CONFIG["auxiliary"], so the config-backed timeout/provider
overrides never worked for that task.
"""
assert "title_generation" in DEFAULT_CONFIG["auxiliary"]
tg = DEFAULT_CONFIG["auxiliary"]["title_generation"]
assert tg["provider"] == "auto"
assert tg["model"] == ""
assert tg["timeout"] > 0
def test_aux_tasks_keys_all_exist_in_default_config():
"""Every task the menu offers must be defined in DEFAULT_CONFIG."""
aux_keys = {k for k, _name, _desc in _AUX_TASKS}
default_keys = set(DEFAULT_CONFIG["auxiliary"].keys())
missing = aux_keys - default_keys
assert not missing, (
f"_AUX_TASKS references tasks not in DEFAULT_CONFIG.auxiliary: {missing}"
)
# ── _format_aux_current ─────────────────────────────────────────────────────
@pytest.mark.parametrize(
"task_cfg,expected",
[
({}, "auto"),
({"provider": "", "model": ""}, "auto"),
({"provider": "auto", "model": ""}, "auto"),
({"provider": "auto", "model": "gpt-4o"}, "auto · gpt-4o"),
({"provider": "openrouter", "model": ""}, "openrouter"),
(
{"provider": "openrouter", "model": "google/gemini-2.5-flash"},
"openrouter · google/gemini-2.5-flash",
),
({"provider": "nous", "model": "gemini-3-flash"}, "nous · gemini-3-flash"),
(
{"provider": "custom", "base_url": "http://localhost:11434/v1", "model": ""},
"custom (localhost:11434/v1)",
),
(
{
"provider": "custom",
"base_url": "http://localhost:11434/v1/",
"model": "qwen2.5:32b",
},
"custom (localhost:11434/v1) · qwen2.5:32b",
),
],
)
def test_format_aux_current(task_cfg, expected):
assert _format_aux_current(task_cfg) == expected
def test_format_aux_current_handles_non_dict():
assert _format_aux_current(None) == "auto"
assert _format_aux_current("string") == "auto"
# ── _save_aux_choice ────────────────────────────────────────────────────────
def test_save_aux_choice_persists_to_config_yaml(tmp_path, monkeypatch):
"""Saving a task writes provider/model/base_url/api_key to auxiliary.<task>."""
from pathlib import Path
monkeypatch.setenv("HERMES_HOME", str(tmp_path / ".hermes"))
monkeypatch.setattr(Path, "home", lambda: tmp_path)
(tmp_path / ".hermes").mkdir(exist_ok=True)
_save_aux_choice(
"vision", provider="openrouter", model="google/gemini-2.5-flash",
)
cfg = load_config()
v = cfg["auxiliary"]["vision"]
assert v["provider"] == "openrouter"
assert v["model"] == "google/gemini-2.5-flash"
assert v["base_url"] == ""
assert v["api_key"] == ""
def test_save_aux_choice_preserves_timeout(tmp_path, monkeypatch):
"""Saving must NOT clobber user-tuned timeout values."""
from pathlib import Path
monkeypatch.setenv("HERMES_HOME", str(tmp_path / ".hermes"))
monkeypatch.setattr(Path, "home", lambda: tmp_path)
(tmp_path / ".hermes").mkdir(exist_ok=True)
# Default vision timeout is 120
cfg_before = load_config()
default_timeout = cfg_before["auxiliary"]["vision"]["timeout"]
assert default_timeout == 120
_save_aux_choice("vision", provider="nous", model="gemini-3-flash")
cfg_after = load_config()
assert cfg_after["auxiliary"]["vision"]["timeout"] == default_timeout
# download_timeout also preserved for vision
assert cfg_after["auxiliary"]["vision"].get("download_timeout") == 30
def test_save_aux_choice_does_not_touch_main_model(tmp_path, monkeypatch):
"""Aux config must never mutate model.default / model.provider / model.base_url."""
from pathlib import Path
monkeypatch.setenv("HERMES_HOME", str(tmp_path / ".hermes"))
monkeypatch.setattr(Path, "home", lambda: tmp_path)
(tmp_path / ".hermes").mkdir(exist_ok=True)
# Simulate a configured main model
from hermes_cli.config import save_config
cfg = load_config()
cfg["model"] = {
"default": "claude-sonnet-4.6",
"provider": "anthropic",
"base_url": "",
}
save_config(cfg)
_save_aux_choice(
"compression", provider="custom",
base_url="http://localhost:11434/v1", model="qwen2.5:32b",
)
cfg = load_config()
# Main model untouched
assert cfg["model"]["default"] == "claude-sonnet-4.6"
assert cfg["model"]["provider"] == "anthropic"
# Aux saved correctly
c = cfg["auxiliary"]["compression"]
assert c["provider"] == "custom"
assert c["model"] == "qwen2.5:32b"
assert c["base_url"] == "http://localhost:11434/v1"
def test_save_aux_choice_creates_missing_task_entry(tmp_path, monkeypatch):
"""Saving a task that was wiped from config.yaml should recreate it."""
from pathlib import Path
monkeypatch.setenv("HERMES_HOME", str(tmp_path / ".hermes"))
monkeypatch.setattr(Path, "home", lambda: tmp_path)
(tmp_path / ".hermes").mkdir(exist_ok=True)
# Remove vision from config entirely
from hermes_cli.config import save_config
cfg = load_config()
cfg.setdefault("auxiliary", {}).pop("vision", None)
save_config(cfg)
_save_aux_choice("vision", provider="nous", model="gemini-3-flash")
cfg = load_config()
assert cfg["auxiliary"]["vision"]["provider"] == "nous"
assert cfg["auxiliary"]["vision"]["model"] == "gemini-3-flash"
# ── _reset_aux_to_auto ──────────────────────────────────────────────────────
def test_reset_aux_to_auto_clears_routing_preserves_timeouts(tmp_path, monkeypatch):
from pathlib import Path
monkeypatch.setenv("HERMES_HOME", str(tmp_path / ".hermes"))
monkeypatch.setattr(Path, "home", lambda: tmp_path)
(tmp_path / ".hermes").mkdir(exist_ok=True)
# Configure two tasks non-auto, and bump a timeout
_save_aux_choice("vision", provider="openrouter", model="gpt-4o")
_save_aux_choice("compression", provider="nous", model="gemini-3-flash")
from hermes_cli.config import save_config
cfg = load_config()
cfg["auxiliary"]["vision"]["timeout"] = 300 # user-tuned
save_config(cfg)
n = _reset_aux_to_auto()
assert n == 2 # both changed
cfg = load_config()
for task in ("vision", "compression"):
v = cfg["auxiliary"][task]
assert v["provider"] == "auto"
assert v["model"] == ""
assert v["base_url"] == ""
assert v["api_key"] == ""
# User-tuned timeout survives reset
assert cfg["auxiliary"]["vision"]["timeout"] == 300
# Default compression timeout preserved
assert cfg["auxiliary"]["compression"]["timeout"] == 120
def test_reset_aux_to_auto_idempotent(tmp_path, monkeypatch):
"""Second reset on already-auto config returns 0 without errors."""
from pathlib import Path
monkeypatch.setenv("HERMES_HOME", str(tmp_path / ".hermes"))
monkeypatch.setattr(Path, "home", lambda: tmp_path)
(tmp_path / ".hermes").mkdir(exist_ok=True)
assert _reset_aux_to_auto() == 0
_save_aux_choice("vision", provider="nous", model="gemini-3-flash")
assert _reset_aux_to_auto() == 1
assert _reset_aux_to_auto() == 0
# ── Menu dispatch ───────────────────────────────────────────────────────────
def test_select_provider_and_model_dispatches_to_aux_menu(tmp_path, monkeypatch):
"""Picking 'Configure auxiliary models...' in the provider list calls _aux_config_menu."""
from pathlib import Path
monkeypatch.setenv("HERMES_HOME", str(tmp_path / ".hermes"))
monkeypatch.setattr(Path, "home", lambda: tmp_path)
(tmp_path / ".hermes").mkdir(exist_ok=True)
from hermes_cli import main as main_mod
called = {"aux": 0, "flow": 0}
def fake_prompt(choices, *, default=0):
# Find the aux-config entry by its label text and return its index
for i, label in enumerate(choices):
if "Configure auxiliary models" in label:
return i
raise AssertionError("aux entry not in provider list")
monkeypatch.setattr(main_mod, "_prompt_provider_choice", fake_prompt)
monkeypatch.setattr(main_mod, "_aux_config_menu", lambda: called.__setitem__("aux", called["aux"] + 1))
# Guard against any main flow accidentally running
monkeypatch.setattr(main_mod, "_model_flow_openrouter",
lambda *a, **kw: called.__setitem__("flow", called["flow"] + 1))
main_mod.select_provider_and_model()
assert called["aux"] == 1, "aux menu not invoked"
assert called["flow"] == 0, "main provider flow should not run"
def test_leave_unchanged_replaces_cancel_label(tmp_path, monkeypatch):
"""The bottom cancel entry now reads 'Leave unchanged' (UX polish)."""
from pathlib import Path
monkeypatch.setenv("HERMES_HOME", str(tmp_path / ".hermes"))
monkeypatch.setattr(Path, "home", lambda: tmp_path)
(tmp_path / ".hermes").mkdir(exist_ok=True)
from hermes_cli import main as main_mod
captured: list[list[str]] = []
def fake_prompt(choices, *, default=0):
captured.append(list(choices))
# Pick 'Leave unchanged' (last item) to exit cleanly
for i, label in enumerate(choices):
if label == "Leave unchanged":
return i
raise AssertionError("Leave unchanged not in provider list")
monkeypatch.setattr(main_mod, "_prompt_provider_choice", fake_prompt)
main_mod.select_provider_and_model()
assert captured, "provider menu never rendered"
labels = captured[0]
assert "Leave unchanged" in labels
assert "Cancel" not in labels, "Cancel label should be replaced"
assert any("Configure auxiliary models" in label for label in labels)