ingest(xpost): Hermes Mixture of Agents (MoA) — multi-model fusion feature
- raw: raw/xpost/2026-06-28-hermes-moa-vaibhavsisinty.md (NEW) - wiki: tools/hermes-desktop.md (updated — new MoA section: architecture, performance, composability, criticism, OpenRouter Fusion comparison) - wiki: index.md (updated — 38th update, Hermes Desktop row updated, new raw source) - wiki: log.md (new entry) Source: @VaibhavSisinty X post, shared by Pit Weber in OME Topic 3770 (Hermès Agents), 2026-06-28
This commit is contained in:
Hector
parent
07bb7e67fc
commit
c35d8e0736
4 changed files with 120 additions and 5 deletions
49
raw/xpost/2026-06-28-hermes-moa-vaibhavsisinty.md
Normal file
49
raw/xpost/2026-06-28-hermes-moa-vaibhavsisinty.md
Normal file
|
|
@ -0,0 +1,49 @@
|
|||
---
|
||||
type: xpost
|
||||
source_url: https://x.com/vaibhavsisinty/status/2070741416649850898
|
||||
retrieved: 2026-06-28
|
||||
author: "@VaibhavSisinty"
|
||||
is_thread: false
|
||||
tags: [hermes, mixture-of-agents, moa, model-fusion, ensemble, nous-research, teknium, lambdua]
|
||||
---
|
||||
|
||||
# Hermes "Mixture of Agents" (MoA) — Merge Any Models Into One Virtual Model
|
||||
|
||||
**Geteilt von:** Pit Weber in OME-Gruppe Topic "Hermès" Agents (Topic 3770), 2026-06-28
|
||||
|
||||
## Source: X Post
|
||||
|
||||
**Author:** Vaibhav Sisinty (@VaibhavSisinty)
|
||||
**Posted:** 2026-06-28
|
||||
**URL:** https://x.com/vaibhavsisinty/status/2070741416649850898
|
||||
|
||||
## Content Summary
|
||||
|
||||
Hermes "Mixture of Agents" (MoA) feature:
|
||||
|
||||
- **Merge any two (or more) AI models into one virtual model**
|
||||
- One model runs as **reference**, one as **aggregator**
|
||||
- Both run **in parallel** per task; aggregator synthesizes final output and handles tool calls
|
||||
- Virtual model appears as a **single selectable model** in the model picker
|
||||
- **Results:** 8% above Opus 4.8 solo, 11% above GPT-5.5 solo on hard agentic tasks
|
||||
- Full Hermes features work **untouched**: Memory, tool use, skills, long sessions, cross-channel messaging
|
||||
- **Any provider mix works:** OpenAI, Anthropic, OpenRouter, local models
|
||||
|
||||
### Community Clarifications
|
||||
|
||||
- **@Teknium** clarified: any number of models, not just two — even multiples of the same model
|
||||
- **Criticism:** increased token costs; @lambdua called it "toy stage"
|
||||
|
||||
## Key Points
|
||||
|
||||
1. **Architectural approach:** Reference model generates draft responses; aggregator model synthesizes final output from reference + its own reasoning. Tool calls handled by aggregator.
|
||||
2. **Performance gains:** Double-digit improvements on hard agentic tasks over solo frontier models (Opus 4.8, GPT-5.5)
|
||||
3. **Transparency:** Virtual model is selectable in picker like any single model — no workflow change needed
|
||||
4. **Composability:** Any provider combination (OpenAI + Anthropic, OpenRouter + local, etc.)
|
||||
5. **Scalability:** Not limited to 2 models — N models supported, including same-model multiples (self-fusion pattern)
|
||||
|
||||
## Cross-References
|
||||
|
||||
- [[../../wiki/tools/hermes-desktop.md]] — Hermes Desktop main page
|
||||
- [[../../wiki/concepts/llm/llm-model-fusion-ensembles.md]] — OpenRouter Fusion / DRACO-Benchmark (related ensemble approach)
|
||||
- [[../../wiki/concepts/llm/ai-intelligence-commoditization-thesis.md]] — Intelligence commoditization context
|
||||
|
|
@ -2,7 +2,7 @@
|
|||
|
||||
*Auto-generated: 2026-06-23*
|
||||
|
||||
*Letzte Aktualisierung: 2026-06-26 (37. Update — Gemini WM 2026 Spielplan-Demo + Meta-Tutorial "KI Lernen mit Praxisbeispielen". Neue Konzept-Seite `concepts/tutorials/gemini-wm2026-spielplan-demo.md`. Neue Tutorial-Übersicht `tutorials/ki-lernen-mit-praxisbeispielen.md`. Raw-Datei `other/2026-06-26_gemini-wm2026-spielplan-ausfuellung.md`.)*
|
||||
*Letzte Aktualisierung: 2026-06-28 (38. Update — Hermes "Mixture of Agents" (MoA) Feature: Merge any N models into one virtual model. Reference + Aggregator Pattern, +8% über Opus 4.8 solo. Neue Sektion in `tools/hermes-desktop.md`. Raw-Datei `xpost/2026-06-28-hermes-moa-vaibhavsisinty.md`.)*
|
||||
|
||||
## Architecture
|
||||
|
||||
|
|
@ -22,7 +22,7 @@
|
|||
| Seite | Beschreibung | Quellen |
|
||||
|-------|-------------|---------|
|
||||
| [OpenClaw](tools/openclaw.md) | Plattform-Referenz: v2026.6.8 stable (16.06.2026), Architektur-Übersicht, plattformrelevante Release-Changes, Update-Plan | other/2026-06-16_openclaw-releases-v2026.6.8.md |
|
||||
| [Hermes Desktop](tools/hermes-desktop.md) | Offizielle native GUI für Hermes Agent. Electron/React/Python. Update v0.17.0 ("Reach") bringt native iMessage-Integration via Photon, asynchrone Subagents mit Watch-Windows, Image-to-Image Bearbeitung, Automation Blueprints und Telegram Rich Text Bot API 10.1. OpenClaw-Migrationstool `hermes claw migrate`. | youtube/2026-06-22_jonas-keil-hermes-desktop.md + youtube/2026-06-24_alex-finn-hermes-agent-v0170-reach.md + youtube/2026-06-24_peter-yang-hermes-full-course.md |
|
||||
| [Hermes Desktop](tools/hermes-desktop.md) | Offizielle native GUI für Hermes Agent. Electron/React/Python. Update v0.17.0 ("Reach") bringt native iMessage-Integration via Photon, asynchrone Subagents mit Watch-Windows, Image-to-Image Bearbeitung, Automation Blueprints und Telegram Rich Text Bot API 10.1. **Mixture of Agents (MoA)** (28.06.2026): Merge any N models into one virtual model (Reference + Aggregator), +8% über Opus 4.8 solo. OpenClaw-Migrationstool `hermes claw migrate`. | youtube/2026-06-22_jonas-keil-hermes-desktop.md + youtube/2026-06-24_alex-finn-hermes-agent-v0170-reach.md + youtube/2026-06-24_peter-yang-hermes-full-course.md + xpost/2026-06-28-hermes-moa-vaibhavsisinty.md |
|
||||
| [Anthropic Claude](tools/anthropic-claude.md) | Claude Opus 4.7, Mythos, Business-Kennzahlen, Pentagon-Streit, Amazon-Jailbreak, Branchen-Fallout | 5 |
|
||||
| [OpenAI GPT Models](tools/openai-gpt.md) | GPT-5.5, GPT-5.5 Instant, Pricing | 1 |
|
||||
| [Ecosystem Tools (April 2026)](tools/ecosystem-tools-april-2026.md) | GBrain, SearXNG, Hermes Atlas, X-API, Claude Shortcuts | context-tree |
|
||||
|
|
@ -236,3 +236,4 @@
|
|||
| `raw/blog/2026-06-25_binaryverseai-qwen-agentworld.md` | blog | BinaryVerse AI: Qwen AgentWorld Explained — Predict Before Agents Act (AgentWorldBench scores, CPT/SFT/RL pipeline, 7 domains, GUI text representation) |
|
||||
| `raw/youtube/2026-06-26_karpathy-how-i-use-llms.md` | youtube | Andrej Karpathy: How I use LLMs (2:11:11, ~2.5M views, English) — Praxis-Crashkurs: LLM Fundamentals, Tool Integration, Multimodal, Custom GPTs |
|
||||
| `raw/other/2026-06-26_gemini-wm2026-spielplan-ausfuellung.md` | other | Gemini Demo: WM 2026 Spielplan ausfüllen per KI — Foto-Verarbeitung (cv2/PIL), Multimodalität, Limitationen (WM läuft noch) |
|
||||
| `raw/xpost/2026-06-28-hermes-moa-vaibhavsisinty.md` | xpost | Hermes "Mixture of Agents" (MoA): Merge any N models into one virtual model (Reference + Aggregator), +8%/+11% über Opus 4.8/GPT-5.5 solo. @Teknium: any number of models. @lambdua: "toy stage" |
|
||||
|
|
|
|||
12
wiki/log.md
12
wiki/log.md
|
|
@ -2,6 +2,18 @@
|
|||
|
||||
*Append-only changelog. Start: 2026-06-05*
|
||||
|
||||
## [2026-06-28] Ingest | Hermes "Mixture of Agents" (MoA) — Multi-Model Fusion Feature
|
||||
**Type:** ingest | **Scope:** raw/xpost, wiki/tools
|
||||
**Source:** X Post — https://x.com/vaibhavsisinty/status/2070741416649850898
|
||||
**Trigger:** Pit Weber shared the X post by @VaibhavSisinty in OME-Gruppe Topic "Hermès" Agents (Topic 3770) on 2026-06-28.
|
||||
**Actions:**
|
||||
- raw: `raw/xpost/2026-06-28-hermes-moa-vaibhavsisinty.md` (created — 2.4 KB; Frontmatter [type: xpost, tags: hermes, mixture-of-agents, moa, model-fusion, ensemble], content summary with MoA architecture, performance data, @Teknium clarification on N-model support, @lambdua criticism, cross-refs to hermes-desktop + llm-model-fusion-ensembles + ai-intelligence-commoditization-thesis)
|
||||
- wiki: `tools/hermes-desktop.md` (updated — new "🧬 Mixture of Agents (MoA)" section before Architecture; Frontmatter sources + tags updated; Quellen-Block um X-Post ergänzt. Section covers: Funktionsweise [Reference/Aggregator table], Performance [+8%/+11% table], Composability [any provider mix, N models, self-fusion], Kritik [token costs, @lambdua "toy stage"], Einordnung [comparison table OpenRouter Fusion vs. Hermes MoA])
|
||||
- wiki: `index.md` (updated — Header auf "38. Update", Hermes Desktop Tabellenzeile aktualisiert mit MoA-Beschreibung + neuer Source, neuer Raw-Sources-Eintrag)
|
||||
- log: this entry
|
||||
**Hector-Hauptthese:** Hermes MoA integriert Model-Fusion in den Agent-Core (mit Tool-Call-Support und Memory-Zugriff für alle beteiligten Modelle) — das ist der entscheidende Unterschied zu reinen API-Fusion-Ansätzen wie OpenRouter Fusion. MoA macht Ensemble-Fusion zu einer First-Class Citizen im Agent-Workflow. Die +8%/+11% Zahlen auf hard agentic tasks sind signifikant und bestätigen das DRACO-Muster aus der OpenRouter-Fusion-Analyse: die Synthese-Schicht ist mindestens so wichtig wie das Modell. Die N-Model-Erweiterung (inkl. Self-Fusion) durch @Teknium ist bemerkenswert — sie bestätigt das Self-Fusion-Pattern (+6.7 DRACO-Punkte) aus unserer Wiki-Seite. Die Kritik ("toy stage", token costs) ist real: Multi-Model-Parallel-Ausführung vervielfacht die API-Kosten. Für OpenClaw ist MoA ein Architectural-Pattern-Vorbild für eine mögliche integrierte Fusion-Schicht im Model-Routing.
|
||||
**Subagent-Modell:** ollama/glm-5.2:cloud
|
||||
|
||||
## [2026-06-26] Ingest | Gemini WM 2026 Spielplan-Demo + Meta-Tutorial "KI Lernen mit Praxisbeispielen"
|
||||
**Type:** ingest | **Scope:** raw/other, wiki/concepts/tutorials (NEW), wiki/tutorials, wiki/index
|
||||
**Source:** Gemini Share Link — https://gemini.google.com/share/57024de890a3
|
||||
|
|
|
|||
|
|
@ -1,8 +1,8 @@
|
|||
---
|
||||
created: 2026-06-22
|
||||
updated: 2026-06-25
|
||||
sources: [youtube/2026-06-22_jonas-keil-hermes-desktop.md, youtube/2026-06-24_alex-finn-hermes-agent-v0170-reach.md, youtube/2026-06-24_peter-yang-hermes-full-course.md]
|
||||
tags: [tools, hermes-agent, hermes-desktop, nous-research, desktop-app, electron, react, gui, openclaw-migration, self-improving, v0.17.0, reach]
|
||||
updated: 2026-06-28
|
||||
sources: [youtube/2026-06-22_jonas-keil-hermes-desktop.md, youtube/2026-06-24_alex-finn-hermes-agent-v0170-reach.md, youtube/2026-06-24_peter-yang-hermes-full-course.md, xpost/2026-06-28-hermes-moa-vaibhavsisinty.md]
|
||||
tags: [tools, hermes-agent, hermes-desktop, nous-research, desktop-app, electron, react, gui, openclaw-migration, self-improving, v0.17.0, reach, mixture-of-agents, moa, model-fusion, ensemble]
|
||||
---
|
||||
|
||||
# Hermes Desktop — Native GUI für Hermes Agent
|
||||
|
|
@ -11,6 +11,7 @@ tags: [tools, hermes-agent, hermes-desktop, nous-research, desktop-app, electron
|
|||
> - YouTube-Video "Hermes DESKTOP ist der WAHNSINN!!" von [Jonas Keil](https://www.youtube.com/@JonasKeil) (AI Trainer), 22.06.2026 → [Raw-Datei](../../raw/youtube/2026-06-22_jonas-keil-hermes-desktop.md)
|
||||
> - YouTube-Video "The new Hermes Agent update has me speechless...." von [Alex Finn](https://www.youtube.com/@AlexFinnOfficial), 24.06.2026 → [Raw-Datei](../../raw/youtube/2026-06-24_alex-finn-hermes-agent-v0170-reach.md)
|
||||
> - YouTube-Video "Hermes Full Course: Build Your 24/7 AI Chief of Staff in 45 Minutes" von Peter Yang, 24.06.2026 → [Raw-Datei](../../raw/youtube/2026-06-24_peter-yang-hermes-full-course.md)
|
||||
> - X-Post von @VaibhavSisinty zu "Mixture of Agents" (MoA), 28.06.2026 → [Raw-Datei](../../raw/xpost/2026-06-28-hermes-moa-vaibhavsisinty.md)
|
||||
> **Offizielle Doku:** [hermes-agent.nousresearch.com/docs/user-guide/desktop](https://hermes-agent.nousresearch.com/docs/user-guide/desktop)
|
||||
> **Background:** [Ewan Mak, Medium Juni 2026](https://medium.com/@tentenco/hermes-agent-desktop-app-everything-you-need-to-know-about-nous-researchs-self-improving-ai-agent-3cb59bd31e5f)
|
||||
|
||||
|
|
@ -66,6 +67,58 @@ Am **19. Juni 2026** veröffentlichte Nous Research das Major-Update **v0.17.0**
|
|||
|
||||
---
|
||||
|
||||
## 🧬 Mixture of Agents (MoA) — Multi-Model Fusion (28. Juni 2026)
|
||||
|
||||
Am **28. Juni 2026** wurde eine neue Hermes-Funktion bekannt: **Mixture of Agents (MoA)**. Diese erlaubt es, beliebig viele AI-Modelle zu einem einzigen **virtuellen Modell** zu verschmelzen, das im Model-Picker als reguläres Modell ausgewählt werden kann.
|
||||
|
||||
> **Quelle:** X-Post von [@VaibhavSisinty](https://x.com/vaibhavsisinty/status/2070741416649850898), geteilt von Pit Weber im OME-Topic "Hermès" Agents (Topic 3770) am 28.06.2026 → [Raw-Datei](../../raw/xpost/2026-06-28-hermes-moa-vaibhavsisinty.md)
|
||||
|
||||
### Funktionsweise
|
||||
|
||||
| Komponente | Rolle |
|
||||
|-----------|------|
|
||||
| **Reference Model(s)** | Generieren parallele Draft-Responses pro Task |
|
||||
| **Aggregator Model** | Synthetisiert finales Output aus Reference-Outputs + eigenem Reasoning; handles Tool Calls |
|
||||
| **Virtuelles Modell** | Erscheint als einzelnes auswählbares Modell im Model-Picker |
|
||||
|
||||
- Beide (oder alle) Modelle laufen **parallel** pro Task
|
||||
- Der Aggregator produziert das finale Output und manages Tool Calls
|
||||
- Alle Hermes-Features (Memory, Tool Use, Skills, Long Sessions, Cross-Channel Messaging) funktionieren **unberührt**
|
||||
|
||||
### Performance
|
||||
|
||||
| Benchmark | MoA vs. Solo |
|
||||
|-----------|-------------|
|
||||
| vs. Opus 4.8 solo | **+8%** auf hard agentic tasks |
|
||||
| vs. GPT-5.5 solo | **+11%** auf hard agentic tasks |
|
||||
|
||||
### Composability
|
||||
|
||||
- **Any provider mix:** OpenAI + Anthropic, OpenRouter + local, beliebige Kombinationen
|
||||
- **Any number of models:** @Teknium klärte, dass nicht nur zwei Modelle unterstützt werden — beliebig viele, inklusive **Multiplikation desselben Modells** (Self-Fusion-Pattern)
|
||||
- **Same-model multiples:** Z.B. 3× GLM-5.2 als Reference + 1× Opus 4.8 als Aggregator
|
||||
|
||||
### Kritik & Limitationen
|
||||
|
||||
- **Erhöhte Token-Kosten:** Mehrere Modelle laufen parallel → höhere API-Kosten pro Task
|
||||
- **@lambdua** bezeichnete es als **"toy stage"** — noch nicht produktionsreif für kostensensitive Workflows
|
||||
|
||||
### Einordnung
|
||||
|
||||
MoA ist eng verwandt mit dem [OpenRouter Fusion / DRACO-Benchmark-Pattern](../concepts/llm/llm-model-fusion-ensembles.md), hat aber entscheidende Unterschiede:
|
||||
|
||||
| Dimension | OpenRouter Fusion | Hermes MoA |
|
||||
|-----------|-------------------|-----------|
|
||||
| **Architektur** | Panel + externer Judge | Reference + Aggregator (integriert) |
|
||||
| **Integration** | API-Layer | Agent-Core (Memory, Tools, Skills) |
|
||||
| **Modell-Anzahl** | 2+ | N (inkl. Self-Fusion) |
|
||||
| **Tool Calls** | Extern | Aggregator handles nativ |
|
||||
| **Kosten** | Pay-per-model | Pay-per-model (höher bei N Modellen) |
|
||||
|
||||
Die Integration in den Agent-Core (mit Tool-Call-Support und Memory-Zugriff für alle beteiligten Modelle) ist der wesentliche Unterschied zu reinen API-Fusion-Ansätzen. MoA macht Ensemble-Fusion zu einer **First-Class Citizen** im Agent-Workflow.
|
||||
|
||||
---
|
||||
|
||||
## Architektur
|
||||
|
||||
### Tech-Stack
|
||||
|
|
|
|||
Loading…
Add table
Reference in a new issue