K R Ü G  ··  spektorAI
the future belongs to the weird

spektorAI

talk your thoughts. get back a draft in your voice, with receipts.

spektorAI is the home of WhisperForge, the voice-to-knowledge workbench I use to turn spoken thoughts into publishable work without losing my own voice. I dictate or upload audio, it transcribes, pulls the wisdom, drafts an article, writes the social posts and image prompts, and ships the whole bundle to a structured Notion page. it runs local-first as a Streamlit app on my machine. the voice comes from a per-user prompt and knowledge base, so the output sounds like me and not like a content mill.

it is built for people who think out loud and hate doing the same cleanup-and-formatting grind twice. it is not a hosted SaaS. you run it from a checkout, your keys stay where you put them, and nothing leaves the machine unless you send it somewhere on purpose.

what it does

• take a Wispr Flow dictation, a pasted note, an upload, or a live browser recording, and turn it into one durable capture
• transcribe locally on Apple Silicon, on CPU, or in the cloud · your call, your tradeoff
• run the pipeline: clean the transcript, pull chapters and wisdom, draft an outline, write the article, spin up social posts and image prompts
• draft, critique, revise. a fact-check pass flags claims the transcript does not actually back up
• write in your voice via prompts + a knowledge base under prompts/<user>/, injected on every call
• review a draft next to its source receipts before anything ships. advisory scorecards, no gatekeeping
• hand off a capture or output as a GitHub or Linear issue draft. routing stays dry-run until you approve it
• ship a clean, color-coded Notion page with title, tags, summary, and run metrics
• pick your writing engine: OpenAI, Anthropic, or local Ollama. default is Claude Haiku 4.5, fast and cheap and in your voice

start here

needs Python 3.10+ and ffmpeg on your PATH (brew install ffmpeg on macOS).

git clone https://github.com/WalksWithASwagger/spektorAI.git
cd spektorAI
python3 -m venv venv && source venv/bin/activate
pip install -r requirements.txt
cp .env.example .env   # add OPENAI_API_KEY + NOTION keys, ANTHROPIC_API_KEY recommended
make app               # → http://localhost:8501

full setup, the provider matrix, and the run-recovery flow live in the engine handbook, WHISPERFORGE.md. roadmap: ROADMAP.md. current handoff state: STATUS.md.

these repos run on agents

built to be operated by humans and machines. see AGENTS.md and llms.txt.

---

i don't build systems that optimize humans. i craft spaces where humans can be gloriously inefficient, creative, and alive.

made by Kris Krüg · @WalksWithASwagger · kriskrug.co