Ingest: Andrew Ng announces OpenWorker — open-source deliverable-focused AI agent

- raw/xpost/2026-07-23-andrewng-openworker.md (new)
- wiki/tools/openworker.md (new)
- 83rd wiki update

raw/xpost/2026-07-23-andrewng-openworker.md

@@ -0,0 +1,36 @@
+---
+source: xpost
+url: https://x.com/AndrewYNg/status/2080333504446108104
+author: Andrew Ng (@AndrewYNg)
+co-author: Rohit Prasad (@rohitcprasad)
+date: 2026-07-23
+ingested: 2026-07-24
+ingested_by: hector
+topic: ai-agent, open-source, local-first, model-agnostic
+---
+
+# Andrew Ng announces OpenWorker
+
+Andrew Ng and Rohit Prasad announce OpenWorker — an open-source AI agent
+that delivers completed work rather than just chat.
+
+## Key features
+
+- Produces finished deliverables: polished documents, Slack messages, calendar updates
+- Example tasks: customer briefs, calendar untangling, report drafts, Slack alert triage
+- Runs locally on Mac (Windows support coming soon)
+- Model-agnostic: GPT 5.6, Claude Fable, Gemini 3.6, Kimi, GLM, DeepSeek, Inkling, or Ollama for fully local/privacy-focused runs
+- Data stays on your machine except for chosen LLM providers and integrations
+- Open source, bring your own API key
+
+## Engagement
+
+- 7k+ likes, 5.5M+ views (as of 2026-07-24)
+- Replies compare OpenWorker to OpenClaw and Hermes
+
+## Relevance to OME
+
+- Overlap with OpenClaw (agent framework, local-first, model-agnostic) and Hermes (deliverable-focused)
+- Key difference: OpenWorker focuses on end-to-end deliverables (finished artifacts from a task),
+  while OpenClaw is a persistent agent framework (butler paradigm, memory, cron, group integration)
+- Andrew Ng entering the agent space validates the category

wiki/tools/openworker.md

@@ -0,0 +1,59 @@
+---
+sources:
+  - raw/xpost/2026-07-23-andrewng-openworker.md
+tags: [ai-agent, open-source, local-first, model-agnostic, deliverable-focused]
+people: [andrew-ng, rohit-prasad]
+updated: 2026-07-24
+---
+
+# OpenWorker — Open-Source AI Agent for Completed Work
+
+## Konzept
+
+OpenWorker ist ein Open-Source-KI-Agent, entwickelt von Andrew Ng und Rohit Prasad.
+Im Gegensatz zu Chat-basierten KI-Interfaces liefert OpenWorker **fertige Deliverables** —
+polierte Dokumente, Slack-Nachrichten, Kalender-Updates — statt nur Konversation.
+
+## Kernfeatures
+
+| Feature | Beschreibung |
+|---------|-------------|
+| Deliverable-Fokus | Produziert fertige Artefakte (Dokumente, Messages, Calendar Updates) |
+| Lokal-first | Runs on Mac (Windows coming soon). Daten bleiben auf der Maschine |
+| Model-agnostic | GPT 5.6, Claude Fable, Gemini 3.6, Kimi, GLM, DeepSeek, Inkling, Ollama |
+| Privacy | Daten lokal außer bei gewählten LLM-Providern und Integrationen |
+| Open Source | Bring your own API key, Source code öffentlich |
+
+## Beispiel-Tasks
+
+- Customer Briefs vorbereiten
+- Kalender-Konflikte auflösen
+- Reports entwerfen
+- Slack-Alerts triagieren
+
+## Einordnung im Agent-Landscape
+
+| Projekt | Ansatz | Schwerpunkt |
+|---------|--------|-------------|
+| **OpenWorker** | Deliverable-Generator | Fertige Artefakte aus einer Aufgabe |
+| **OpenClaw** | Persistent Agent Framework | Butler-Paradigma, Memory, Cron, Gruppen-Integration |
+| **Hermes** | Desktop Agent OS | MoA, GUI, iMessage, Browser Extension |
+| **LM Studio Bionic** | Local Agent Desktop | Sandbox, Checkpoints, Voice-Input |
+
+OpenWorker und OpenClaw überschneiden sich bei Local-First und Model-Agnostik,
+aber der architektonische Schwerpunkt unterscheidet sich: OpenWorker ist
+task-to-deliverable, OpenClaw ist persistent-butler. Beide könnten komplementär sein.
+
+## Bedeutung
+
+Andrew Ng betritt den Agent-Raum — das validiert die Kategorie.
+Mit 7k+ Likes und 5.5M+ Views ist das keine Niche-Meldung.
+Die Replies vergleichen OpenWorker direkt mit OpenClaw und Hermes.
+
+## Cross-References
+
+- [OpenClaw](tools/openclaw.md) — Platform-Referenz
+- [Hermes Desktop](tools/hermes-desktop.md) — Vergleichbarer Agent-Ansatz
+- [LM Studio Bionic](concepts/lm-studio-bionic.md) — Lokaler Agent, kommerzieller
+- [BMASS — Roemmele](concepts/bmass-roemmele.md) — Sovereign Edge Manifest
+- [Model Routing](tools/model-routing.md) — Cost-Saving Patterns für Multi-Model