fix(picker): remove max_models=50 cap in interactive model pickers

[Elshayib]

Summary

The interactive model pickers (Desktop REST API, TUI model.options, CLI /model) were hard-capped at max_models=50, which truncated large provider catalogs like Kilo Gateway (336 models) to just 50 entries. This made most models undiscoverable via the picker search box — searching for a known model returned zero results even though it existed in the local cache.

The total_models field was already present in the response payload but was never rendered by the Desktop UI, so users had no indication that the cap existed.

Root Cause

build_models_payload() defaulted to max_models=50 and list_authenticated_providers() defaulted to max_models=8. All 5 interactive picker callers passed max_models=50 explicitly, truncating the model list before the picker ever saw the data.

Changes

• build_models_payload(): default changed from max_models=50 to max_models=None (unlimited)
• list_authenticated_providers(): default changed from max_models=8 to max_models=None
• list_picker_providers(): default changed from max_models=8 to max_models=None
• All [:max_models] slicing updated to [:max_models] if max_models else model_ids to handle None as "no limit"
• Removed max_models=50 from 5 interactive picker callers:
  • hermes_cli/web_server.py:get_model_options — Desktop /api/model/options
  • hermes_cli/web_server.py:get_recommended_default_model — Desktop recommended model endpoint
  • hermes_cli/model_switch.py:prewarm_picker_cache_async — picker cache prewarm
  • tui_gateway/server.py:model.options — TUI JSON-RPC handler
  • cli.py:HermesCLI — CLI /model picker
• Telegram/Discord inline keyboard picker (gateway/slash_commands.py) still passes max_models=50 explicitly — unchanged behavior (inline keyboards need a cap)
• Added regression test test_build_models_payload_no_max_models_returns_full_list

Testing

• All existing tests pass: test_inventory.py (31 tests), test_list_picker_providers.py (10 tests), test_model_catalog.py (28 tests)
• New test verifies that omitting max_models returns the full 100-model list

What platforms did you test on?

Linux (Ubuntu 26.04). The changes are in shared Python backend code — no platform-specific behavior.

Fixes #48279

[alt-glitch]

Duplicate of #35447 — identical mechanism. Both change build_models_payload() default to max_models=None (unlimited), make list_authenticated_providers()/list_picker_providers() skip the [:max_models] slice when None, and remove the max_models=50 cap from the same set of interactive picker callers (cli.py, model_switch.py, web_server.py, tui_gateway/server.py). #35447 is the earlier open PR using this remove-the-cap-entirely approach.

Related: #48279 (the Kilo Gateway issue this fixes), #40747 (open competing fix that bumps the cap 50→200 instead of removing it), #40601/#42270/#35443 (issue cluster for the same picker cap).

[Elshayib]

Appreciate the heads-up. Yes, #35447 and this PR address the same root cause (the hard-coded max_models=50 cap). Here's how they compare:

• fix(model-picker): load all models in interactive pickers instead of capping at 50 #35447 (liuhao1024, May 30): changes the default to max_models=None in build_models_payload() and skips the slice when None. Covers 4 slicing sites.
• fix(picker): remove max_models=50 cap in interactive model pickers #48297 (this PR, Jun 18): same mechanism, but also changes the defaults in list_authenticated_providers() and list_picker_providers() from max_models=8 to None, covers all 5 interactive picker callers (including web_server.py:get_recommended_default_model and model_switch.py:prewarm_picker_cache_async which fix(model-picker): load all models in interactive pickers instead of capping at 50 #35447 misses), and adds a regression test.

This PR is a strict superset — it fixes every interactive picker path, not just the ones in #35447. I'd suggest we merge this one and close #35447. If maintainers disagree I'm happy to rebase on top.

[alaamohanad169-ship-it]

Acknowledged: noted competing PR(s) #35447, #35447, #48279, #40747, #40601, #42270, #35443. My approach differs and is complementary, not mutually exclusive. Open to consolidating if reviewers prefer one direction.

[teknium1]

Merged to main via PR #48611 (rebase-merge, commit 3042045) — your commit was cherry-picked with authorship preserved in git history. Thank you!

We added one follow-up on top: tightened the slicing guard from if max_models to if max_models is not None so max_models=0 (used by the slug-only callers that want an empty model list) stays distinct from None/unlimited, plus a regression test locking the three-way cap contract (None=full, 0=empty, N=first N).

#48611