# Tool relationships and working methods

A point-in-time map of what each tool changes in the workflow.

## Research and synthesis

**Status:** Active

I use broad tools to get oriented, then return to primary evidence.

- Genspark
- Web research
- Markdown notes

## Model routing and comparison

**Status:** Experimenting

Different models are collaborators with different strengths, not a permanent winner.

- OpenRouter
- DeepSeek
- Gemini models
- NVIDIA Nemotron

## Coding and building

**Status:** Active

I move between agents to compare their planning, implementation, and review instincts.

- Codex
- Cursor
- Claude Code
- Gemini CLI
- Pi

## Agent orchestration

**Status:** Personal experiments

The experiment is the team shape: roles, hand-offs, tools, memory, and permission boundaries.

- Hermes
- OpenClaw
- Sarah agent
- Reusable skills

## Local inference

**Status:** Prototyping

I test which useful workflows can stay on-device without pretending every task belongs there.

- Ollama
- MLX
- oMLX
- Local model workspaces

## Image and motion

**Status:** Experimenting

Images are evidence and narrative material, so generation stays connected to a real build story.

- OpenAI image models
- Flux
- Kling through Fal

## Terminal and connection

**Status:** Active

Small interface choices make multi-agent work calmer and easier to inspect.

- Ghostty
- Herdr
- Tailscale
- Wispr Flow

## Shipping

**Status:** Active

A build only becomes useful when somebody else can open it and its limits are explicit.

- GitHub
- Vercel
- Railway
- Supabase
- GitHub Actions
