Senter as Hermes Auxiliary¶

The integration pattern. See the full blog post: ../../blog/senter-as-hermes-auxiliary.md

Definition¶

Senter (any model in the Senter family) sits in front of Hermes Agent, doing the work that doesn't need a 70B-class brain. When the task gets hard, Senter hands a structured notebook to Hermes and gets a decision back.

The user doesn't see Hermes. They just see Senter being smart, fast, and remembering everything.

The integration point¶

hermes-agent/agent/auxiliary_client.py is the existing class. It extends to use Senter specifically: - Default auxiliary model: senter-ohm-moe-32a8b (4-bit GGUF, :11500). For lighter deployments, swap in omnisenter-12b. - Auxiliary tasks: vision, summarization, agentic routing, notebook management. - The main agent stays whatever the user has configured (Claude, Hermes-4, etc.)

The notebook-as-API pattern¶

The notebook is the API surface between Senter and Hermes:

Senter → Hermes (escalation):

notebook_handoff:
  schema_version: "1.0"
  session_summary: |
    User is working on a music video for their band.
    Previous turn: chose indie-pop, gave lyrics draft.
  recent_moments: [ ... 3 most recent ... ]
  question: "What's the best DaVinci Resolve workflow for syncing audio?"
  expected_response:
    format: "yaml"
    schema: "hermes_decision_v1"
    fields: { decision, steps, confidence, needs_clarification }
  constraints: { max_response_tokens: 500, must_include_sources: true }

Hermes → Senter (response):

hermes_response:
  decision: |
    1. Import the audio track
    2. Set the project frame rate
    3. Use Auto Sync
    4. Apply warp/elastic stretch
  steps: [ ... ]
  confidence: 0.92
  sources: [ ... ]
  senter_should:
    - action: "summarize_for_user"
    - action: "update_notebook"
      decision_record: true
      importance: 0.8

The escalation rules¶

Senter escalates to Hermes when: - Question requires deep reasoning ("why", "how", "best ... approach", "compare", etc.) - Question is a complex multi-step task (> 3 estimated steps) - User explicitly asks for the "smart agent" / "hermes"

Senter handles directly when: - Trivial (greeting, ack, "what time is it") - Plugin-friendly (image gen, music, search, weather) - Notebook query (recall by text/audio/image)

The cost model¶

Task type	Senter cost	Hermes cost
Trivial	~50ms, 100 tokens	$0
Plugin call	~500ms, 1K tokens	$0
Notebook query	~200ms, 500 tokens	$0
Reasoning	~2s, 2K tokens	~5s, 4K tokens
Multi-step	~3s, 3K tokens	~15s, 8K tokens

Estimated savings: 86% cost reduction for a typical session (20 turns, 12 trivial/plugin handled by Senter, 6 notebook queries, 2 escalations).

The deployment¶

# Terminal 1: Start the Senter server
python3 ohmd.py serve --model senter-ohm-moe-32a8b-q4_k_m.gguf \
    --notebook-path ~/.senter/notebook/ \
    --port 11500

# Terminal 2: Start Hermes with Senter as auxiliary
hermes --auxiliary-model senter-ohm-moe-32a8b \
       --auxiliary-endpoint http://localhost:11500/v1