Group: Technical Capabilities · Status: Demonstrated
Claude Code, Ollama, Postgres, and n8n composed into custom render pipelines. AI tooling stays in the loop without surrendering data to third-party services.
The shape of the capability
- Claude Code as the IDE. All scripting and content authoring happens inside Claude Code with the codebase as context. No copy-paste round-trips to a chat window.
- Ollama for inference.
qwen3-a3b-32kfor the RWP digest, summary generation, and tag inference. Runs at127.0.0.1:11434— meeting content, financials, and personal data never leave the machine. - Postgres + n8n for personalOS. Local Postgres at
127.0.0.1:5432, n8n workflows in Docker. Webhooks land locally, workflows run locally, notifications fire locally. - Custom generators over off-the-shelf tools.
portal_index.py,render_section_hub.py,gen_state_index.py,gen_digest.pyare all ~200-400 lines of plain Python. Each does one thing the way the engagement needs, not the way a SaaS thinks an engagement should work.
Canonical sources
📎 RWP repo → RWP/CLAUDE.md → Infrastructure
📎 RWP repo → 02_active/personalos/CLAUDE.md
Where to look in the repo
RWP/tools/sync_portal.py— sync daemon (watchdog + debounce + atomic writes)02_active/rwp/workspaces/gen_digest.py— Ollama digest generatorRWP/fathom_archive.py— nightly Fathom archive (launchd)