Extracting production-grade tools out of an enterprise AI platform — one repo at a time.
10 standalone tools carved out of AI Harbor, an in-house AI platform for finance & knowledge work. Each repo is self-contained: its own server, docs, tests where they exist, and a README that explains the mechanism, not the marketing.
| Repo | What it is |
|---|---|
| Computer-Use-Agent | ⭐ Teach once, replay forever — record browser flows as a taught SOP; a CDP driver + Hermes LLM agent replays them with self-healing, SSE task lifecycle, and full audit trails. |
| smart-marketing | Retail marketing workbench: audience selection → campaign orchestration → touchpoint execution → conversion attribution. |
| Intent-Knowledge-Builder | Rule-based intent knowledge authoring: DSL toolkit + workbench for NLU intent systems. |
| Repo | What it is |
|---|---|
| Ontology-builder | AI-native ontology modeling: AI extraction wizard → Protégé-style editing → SHACL/SKOS/SWRL → consistency reasoning → G6 graph browsing. |
| llm-wiki-builder | LLM-driven wiki: scan projects into wiki pages, [[bi-links]] graph, RAG/GraphRAG retrieval, zero-database single service. |
| rag-builder | RAG vector knowledge workbench: parse → chunk → embed → hybrid retrieval → QA. Single Node service. |
| graph-builder | Knowledge-graph building toolchain with dual RSI (Reasoning-Self-Improvement) evolution labs. |
| Repo | What it is |
|---|---|
| Skill-Hub | Agent skill lifecycle platform: author → validate → benchmark → evolve → publish → install → trial-run (OpenJiuWen DeepAgent). |
| Model-Builder | Model training & evaluation workbench: dataset synthesis, metric kernels, experiment reports. |
| ai-native-bank | AI-native mobile bank prototype: multi-intent conversational agent runtime + banking intent evaluation. |
- Want to see an agent do something? → Computer-Use-Agent (has a demo GIF in the README)
- Building knowledge systems? → Ontology-builder or rag-builder
- Working on agent skills / evals? → Skill-Hub + Model-Builder
Every repo was extracted from a production system (an enterprise AI platform serving finance & knowledge-work scenarios), then de-coupled: renamed namespaces, replaced platform config with env vars / local storage, added standalone entry points, and documented with mechanism-first READMEs — design trade-offs and known limitations stated honestly, benchmarks only where measured.