feat(moa): add reference model toggles by oppenheimor · Pull Request #59753 · NousResearch/hermes-agent

oppenheimor · 2026-07-06T16:34:42Z

What does this PR do?

Adds per-reference enable/disable support for MoA reference models.

Each MoA reference model now keeps an enabled flag, defaulting to true for existing configs. Disabled references stay in the preset, but are skipped during MoA inference so they do not make API calls or contribute to the aggregator context.

Related Issue

feat #59707

Type of Change

🐛 Bug fix (non-breaking change that fixes an issue)
✨ New feature (non-breaking change that adds functionality)
🔒 Security fix
📝 Documentation update
✅ Tests (adding or improving test coverage)
♻️ Refactor (no behavior change)
🎯 New skill (bundled or hub)

Changes Made

Added per-reference enabled normalization for MoA presets.
Skips disabled references in both persistent MoA provider runs and one-shot MoA context aggregation.
Added MoA reference toggles to the desktop and dashboard MoA settings UI.
Preserves reference enabled state when changing provider/model selections.
Added backend/runtime coverage for default-enabled references and disabled-reference skipping.
Added desktop settings coverage for saving disabled MoA references.

How to Test

Run MoA backend tests:

source .venv/bin/activate
pytest tests/hermes_cli/test_moa_config.py tests/run_agent/test_moa_loop_mode.py tests/cli/test_moa_command.py

Run dashboard typecheck:
```
npm --workspace web run typecheck
```
Manually verify in the dashboard:
```
hermes dashboard
```
Open Models -> Configure MoA presets, toggle a reference model off, save, and confirm the disabled reference stays in the preset but is skipped during MoA runs.

Checklist

Code

I've read the Contributing Guide
My commit messages follow Conventional Commits
I searched for existing PRs to make sure this isn't a duplicate
My PR contains only changes related to this feature
I've run pytest tests/ -q and all tests pass
I've added tests for my changes
I've tested on my platform: macOS

Documentation & Housekeeping

I've updated relevant documentation — N/A
I've updated cli-config.yaml.example if I added/changed config keys — N/A
I've updated CONTRIBUTING.md or AGENTS.md if I changed architecture or workflows — N/A
I've considered cross-platform impact — N/A
I've updated tool descriptions/schemas if I changed tool behavior — N/A

Screenshots / Logs

Manually verified through hermes dashboard.

alt-glitch · 2026-07-06T16:42:29Z

This was generated by AI during triage.

Related: implements the per-reference MoA enable/disable feature requested in #59707 (kept open as the spec thread). Distinct from #59743, which adds the per-preset enabled toggle to the desktop UI (#59723) -- this PR is per-reference-model granularity and touches the full stack (agent/moa_loop.py, moa_config, desktop + dashboard). Note: the body says 'Fixes #59744', but #59744 is about hiding unconfigured providers in the model picker and appears unrelated to this change.

oppenheimor · 2026-07-06T16:44:26Z

Thanks for catching that. I updated the PR body to reference #59707 instead of #59744, and used “Related to” since #59707 is being kept open as the spec thread. This PR is scoped to per-reference-model enable/disable behavior.

teknium1

Thanks for implementing the per-reference MoA toggle requested in #59707. The feature premise is still valid on current main: hermes_cli/moa_config.py:92-110 drops per-reference enabled, and both execution paths fan out every slot (agent/moa_loop.py:683-690, 906-908).

Problems

This branch predates current per-slot reasoning support. Current hermes_cli/moa_config.py:107-109 preserves reasoning_effort (commit 3dca75b45c2edf38089ef07c22023afe357aebc1); merge the enabled handling into that normalizer rather than replacing it.
The runtime test covers persistent MoAChatCompletions.create(), but the PR also changes one-shot aggregation. Add a disabled-reference test directly through aggregate_moa_context() (agent/conversation_loop.py:877-893).
Add an accessible name to the dashboard switch; nearby dashboard toggle controls provide aria-label (web/src/components/ToolsetConfigDrawer.tsx:252-256).

Suggested changes

Salvage against current MoA slot semantics and retain reasoning_effort, reference_max_tokens, and fanout behavior.
Cover both persistent and one-shot reference fan-out paths.

Automated hermes-sweeper review.

teknium1 · 2026-07-15T19:55:48Z

@@ -87,12 +106,15 @@ def _clean_slot(slot: Any) -> dict[str, str] | None:
    # an invalid slot is dropped, falling back to the preset's defaults.
    if provider.lower() == "moa":
        return None


Current main's _clean_slot preserves reasoning_effort (3dca75b45c2edf38089ef07c22023afe357aebc1, now hermes_cli/moa_config.py:107-109). When salvaging, compose enabled normalization with that behavior rather than replacing it, or normalization will discard existing per-slot reasoning settings.

teknium1 · 2026-07-15T19:55:48Z

@@ -594,6 +594,7 @@ def aggregate_moa_context(
    provider default applies — matching single-model agent behavior. Presets


Please add a direct aggregate_moa_context() test for this filter. The new runtime test covers persistent MoAChatCompletions.create(), but one-shot /moa reaches this separate path through agent/conversation_loop.py:877-893.

teknium1 · 2026-07-15T19:55:48Z

+                className={cn(
+                  "flex items-center gap-2 border border-border/50 bg-muted/20 px-3 py-2",
+                  slot.enabled === false && "opacity-60"
+                )}


Add a stateful accessible name such as Disable reference ${index + 1} / Enable reference ${index + 1}. This switch has no associated label; existing dashboard switches provide aria-label.

…abled flag The per-reference-model enabled toggle (#59753 salvage) intentionally adds 'enabled' to normalized slot dicts. The two endpoint tests asserted the exact key set {provider, model} — convert them to subset + round-trip contracts so optional slot keys (enabled, reasoning_effort, max_tokens) don't break them again.

Follow-up for salvaged PR #59753 rebased over the per-slot reasoning_effort feature: _clean_slot now round-trips reasoning_effort AND enabled together; add a normalize→normalize regression test, update the validate/normalize agreement contract for the canonical enabled default, restore the desktop per-slot toggle test on the current autosave editor, and map oppenheimor's contributor email.

…abled flag The per-reference-model enabled toggle (#59753 salvage) intentionally adds 'enabled' to normalized slot dicts. The two endpoint tests asserted the exact key set {provider, model} — convert them to subset + round-trip contracts so optional slot keys (enabled, reasoning_effort, max_tokens) don't break them again.

Follow-up for salvaged PR #59753 rebased over the per-slot reasoning_effort feature: _clean_slot now round-trips reasoning_effort AND enabled together; add a normalize→normalize regression test, update the validate/normalize agreement contract for the canonical enabled default, restore the desktop per-slot toggle test on the current autosave editor, and map oppenheimor's contributor email.

…abled flag The per-reference-model enabled toggle (#59753 salvage) intentionally adds 'enabled' to normalized slot dicts. The two endpoint tests asserted the exact key set {provider, model} — convert them to subset + round-trip contracts so optional slot keys (enabled, reasoning_effort, max_tokens) don't break them again.

Follow-up for salvaged PR #59753 rebased over the per-slot reasoning_effort feature: _clean_slot now round-trips reasoning_effort AND enabled together; add a normalize→normalize regression test, update the validate/normalize agreement contract for the canonical enabled default, restore the desktop per-slot toggle test on the current autosave editor, and map oppenheimor's contributor email.

…abled flag The per-reference-model enabled toggle (#59753 salvage) intentionally adds 'enabled' to normalized slot dicts. The two endpoint tests asserted the exact key set {provider, model} — convert them to subset + round-trip contracts so optional slot keys (enabled, reasoning_effort, max_tokens) don't break them again.

teknium1 · 2026-07-24T01:42:10Z

Merged via cluster PR #70283 (commit 385a065) — your commit cherry-picked with authorship preserved, rebased over the per-slot reasoning_effort feature that landed after you opened this; _clean_slot now round-trips reasoning_effort, max_tokens, AND enabled together with a round-trip regression test. Implements #59707. Thanks!

Follow-up for salvaged PR NousResearch#59753 rebased over the per-slot reasoning_effort feature: _clean_slot now round-trips reasoning_effort AND enabled together; add a normalize→normalize regression test, update the validate/normalize agreement contract for the canonical enabled default, restore the desktop per-slot toggle test on the current autosave editor, and map oppenheimor's contributor email.

…abled flag The per-reference-model enabled toggle (NousResearch#59753 salvage) intentionally adds 'enabled' to normalized slot dicts. The two endpoint tests asserted the exact key set {provider, model} — convert them to subset + round-trip contracts so optional slot keys (enabled, reasoning_effort, max_tokens) don't break them again.

feat(moa): add reference model toggles

28d6d70

teknium1 reviewed Jul 15, 2026

View reviewed changes

teknium1 added sweeper:risk-compatibility Sweeper risk: may break existing users, config, migrations, defaults, or upgrades sweeper:blast-moderate Sweeper blast radius: moderate — a subsystem or single platform labels Jul 15, 2026

teknium1 mentioned this pull request Jul 23, 2026

fix(moa): desktop/TUI advisor visibility, progress indicator, preset + per-advisor toggles (cluster salvage) #70283

Merged

teknium1 closed this Jul 24, 2026

teknium1 mentioned this pull request Jul 24, 2026

feat(moa): Add enable/disable toggle for individual reference models #59707

Closed

Navigation Menu

Uh oh!

feat(moa): add reference model toggles - #59753

feat(moa): add reference model toggles#59753
oppenheimor wants to merge 1 commit into
NousResearch:mainfrom
oppenheimor:feat/moa-reference-toggle-59744

oppenheimor commented Jul 6, 2026 •

edited

Loading

Uh oh!

alt-glitch commented Jul 6, 2026

Uh oh!

oppenheimor commented Jul 6, 2026

Uh oh!

teknium1 left a comment

Uh oh!

teknium1 Jul 15, 2026

Uh oh!

teknium1 Jul 15, 2026

Uh oh!

teknium1 Jul 15, 2026

Uh oh!

teknium1 commented Jul 24, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

		@@ -594,6 +594,7 @@ def aggregate_moa_context(
		provider default applies — matching single-model agent behavior. Presets

Uh oh!

Conversation

oppenheimor commented Jul 6, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What does this PR do?

Related Issue

Type of Change

Changes Made

How to Test

Checklist

Code

Documentation & Housekeeping

Screenshots / Logs

Uh oh!

alt-glitch commented Jul 6, 2026

Uh oh!

oppenheimor commented Jul 6, 2026

Uh oh!

teknium1 left a comment

Choose a reason for hiding this comment

Problems

Suggested changes

Uh oh!

teknium1 Jul 15, 2026

Choose a reason for hiding this comment

Uh oh!

teknium1 Jul 15, 2026

Choose a reason for hiding this comment

Uh oh!

teknium1 Jul 15, 2026

Choose a reason for hiding this comment

Uh oh!

teknium1 commented Jul 24, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

oppenheimor commented Jul 6, 2026 •

edited

Loading