Summary

Pulled the branch and walked api/config.py plus the three test files. The shape is right: ask Hermes CLI's provider_model_ids() first, then fall back to WebUI's static _PROVIDER_MODELS, with explicit user-configured models allowlists still winning. This is one source-of-truth slice for #1240 — the agent owns the live registry, WebUI follows.

Code reference

The new priority order on api/config.py:2980-2998:

provider_cfg = cfg.get("providers", {}).get(pid, {})
raw_models = []

# User-configured model allowlists are explicit local
# source-of-truth and should still beat auto-discovery.
# Otherwise, ask Hermes CLI first so WebUI tracks the same
# live catalog as the agent/CLI picker; WebUI's static
# _PROVIDER_MODELS table is now a fallback only (#1240).
if isinstance(provider_cfg, dict) and "models" in provider_cfg:
    cfg_models = provider_cfg["models"]
    if isinstance(cfg_models, dict):
        raw_models = [{"id": k, "label": k} for k in cfg_models.keys()]
    elif isinstance(cfg_models, list):
        raw_models = [{"id": k, "label": k} for k in cfg_models]

if not raw_models:
    raw_models = _models_from_live_provider_ids(
        pid,
        _read_live_provider_model_ids(pid),
    )

if not raw_models:
    raw_models = copy.deepcopy(_PROVIDER_MODELS.get(pid, []))

That four-tier resolution (config allowlist → CLI live → static fallback → auto-detected) reads cleanly against what each layer is authoritative for.

Cross-repo agent contract

The CLI side at ~/.hermes/hermes-agent/hermes_cli/models.py:1905 is exactly the right entry point — provider_model_ids() already special-cases the live-fetch providers internally:

def provider_model_ids(provider: Optional[str], *, force_refresh: bool = False) -> list[str]:
    normalized = normalize_provider(provider)
    if normalized == "openrouter":
        return model_ids(force_refresh=force_refresh)
    if normalized == "openai-codex":
        # ...uses live ChatGPT catalog endpoint with OAuth token...
        return get_codex_model_ids(access_token=access_token)
    if normalized in {"copilot", "copilot-acp"}:
        try:
            live = _fetch_github_models(_resolve_copilot_catalog_api_key())
            if live:
                return live
        except Exception:
            pass
    if normalized == "nous":
        # ...live Nous Portal /models endpoint...

Calling that means WebUI now sees catalog updates for openrouter, codex, copilot, nous, stepfun, etc. without a WebUI release. Good cross-repo factoring.

Diagnosis

Three small observations:

1. Alias handling at api/config.py:1985-1990. _resolve_provider_alias(pid) is tried as a secondary lookup, and the seen dedup happens across both candidates. That handles cases like WebUI's google / x-ai display IDs that Hermes CLI normalizes to google-gemini-cli / xai. Confirmed _resolve_provider_alias is defined at api/config.py:661.

2. Failure shape. _read_live_provider_model_ids() returns [] on any exception — no logging at info/warn level, only logger.debug("Failed to load %s models from hermes_cli", candidate). That is the correct level for a normal-path fallback (e.g. older agent installs without provider_model_ids); a noisier logger would spam logs for users on stale agents. Static fallback then takes over.

3. Ollama label formatter routing. _models_from_live_provider_ids at api/config.py:2007-2020 selects _format_ollama_label for ollama and ollama-cloud. That preserves the existing label shape (e.g. Llama 3.1 70B) for live-discovered Ollama tags rather than falling back to the generic _get_label_for_model. Worth keeping in mind that any future Ollama label refinement needs to land in both the live and static paths now.

One question

The provider_cfg allowlist branch builds raw_models from raw IDs but does not apply any label transform. Is that intentional? For ollama, a configured allowlist will get [{"id": "llama3:70b", "label": "llama3:70b"}], while the CLI-live path will go through _format_ollama_label and produce [{"id": "llama3:70b", "label": "Llama 3 70B"}]. Probably already true on master, so not a regression — but if you wanted parity, plumbing the formatter into the allowlist branch is a one-liner. Not blocking.

Verification

The three test files cover the right cases: test_issue1240_generic_cli_catalog_sync.py for the new CLI-first ordering, test_issue644.py for prefix handling, and test_model_resolver.py adjustments for the new resolution order. CI is green on 3.11/3.12/3.13 at commit 91890ab. Looks ready to merge.

Thanks @Michaelyklam — this shipped in v0.51.8 (commit 85d0279) as part of a 7-PR full-sweep batch release. Stage rebased your branch onto current master, ran the full pre-release gate (4584 pytest, browser tests, Opus advisor verdict SHIP), and merged via release PR #1737.

GitHub didn't auto-close because the merge commit only references the squash-merged stage branch, not your fork's commit directly — closing manually for hygiene.

Live now on https://get-hermes.ai/ and on existing installs after git pull + restart.

Release notes: https://github.com/nesquena/hermes-webui/releases/tag/v0.51.8