Ghostwriter writes with you… not for you.
Ghostwriter is a locally-run, AI-inflected writing companion for creative thinkers, speculative designers, and anyone who writes to discover what they think. It doesn’t just autocomplete — it remembers your writing, retrieves your voice, and responds like a future version of yourself.
Ghostwriter is:
- A semantic memory engine that embeds and indexes your writing corpus
- A context-aware autocompleter powered by GPT-4, Ollama, or LM Studio
- A terminal-based co-writer with a beautiful TUI interface
- A creative challenge mode the world's first patented real-time writing exercise-for-your-brain — Vibewriter™ Mode
“Ghostwriter doesn’t just finish my sentences — it finishes my thoughts.”
— Wm. Branson, Jr., Executive
“This is the tool I use in my writing seminars, and its been approved by the college whereas vanilla AI still raises eyebrows. When will ”
— G. Beato Mason, Creative Writing Tutor at Oberlin College
“It’s like having that voice in my head able to actual write with me rather than just bug me to write!”
— A. Rosenthal, Retired Financial Analyst
An integral part of my morning writing ritual. The blank page doesn't feel so lonely anymore.”
— Chaya Trent, Tech Strategist
“It helps me finishes my thoughts in words, which is exceptionally helpful esp. when my attention tends to wander or I have trouble connecting multiple entangled thoughts.”
— Thomas Cornell, Fractional Novelist
⭐⭐⭐⭐
Ghostwriter ingests .txt, .md, and .pdf files and embeds them with your text embedding model of choice (eg text-embedding-ada-002, nomic-embed-text-v1.5-GGUF, etc.) It stores embeddings in a local vector store and recalls them using cosine similarity to enrich and make relevant completions.
When you type a sentence fragment, Ghostwriter finds semantically similar excerpts from your archive and builds an LLM prompt with:
- Recent conversation context
- Matching documents from your own corpus
- Your current writing fragment
Then it completes your thought — in your own voice.
Choose your backend:
- OpenAI (GPT-4o)
- Google Deepmind (Gemini forthcoming)
- Ollama (run Gemma3, Qwen3, and other open-source models locally)
- LM Studio (point to a local inference server)
Vibewriter is a timed, improvisational writing mode where you and the Ghost take turns writing under a countdown. It's like tossing a medicine ball or rolling tractor tires with your AI muse..a high-intensity morning workout for your writer's brain.
-- Timed Turns: Set a tempo (10s, 30s, 60s, etc.) -- Co-Authoring: You write, the Ghost replies. Back and forth. -- Like Jazz: It's not about control. “Trade fours” with the AI in speed rounds, learning to keep the flow going! -- A Creative Writing Workout - for your brain!
Use it to: break creative ruts, riff toward new ideas, warm up for writing, or just have fun.
“Vibewriter feels like jamming with my muse, the source, not wrestling with my own sometimes reluctant brain.”
— Trevor R., Creative Writing Coach and Beta Tester
“I think Vibewriter is a stand-alone product, tbh. It's like the gymnasium everyone who writes needs to hit every morning!”
— B. Pilshaw, Generalist & Tarot Muse
“Vibewriter feels like jamming with a ghost version of my own brain.” — Beta tester
Every prompt, vector match, and AI response is saved as a .jsonl log entry so you can:
- Audit and reflect on your writing flow
- Fine-tune models later
- Export your sessions
What’s New in v1.3.0 “Luminous Echo" Vibewriter Mode Model-Agnostic Inference Engine Multi-format Ingestion (TXT, MD, PDF, URL) Logging (JSONL) Smart Chunking + Semantic Recall Corpus 2.0! Check in and out reference materials in real time!
- Ghostwriter uses both inference and RAG-based patterns for collaborating with you and keeping subject matter on point
- Ingest: Drop .txt, .md, .epub, or .pdf files
- URLs: Ingest from a URL
- Chunk: Text is divided into overlapping chunks
- Embed: Each chunk gets a unique vector representation produced by the appropriate embedding model
- Store: Chunks + embeddings saved in a bundled vector store called a Corpus.
- Corpus: You can have multiple Corpus databases, representing different projects.
- Search: Input is embedded and matched to similar chunks
- Prompt: LLM gets context-rich prompt to generate completion
- Complete: You write — Ghostwriter listens and responds How It Works
- Ghostwriter uses both inference and RAG-based patterns for collaborating with you and keeping subject matter on point
- Ingest: Drop .txt, .md, .epub, or .pdf files
- URLs: Ingest from a URL
- Chunk: Text is divided into overlapping chunks
- Embed: Each chunk gets a unique vector representation produced by the appropriate embedding model
- Store: Chunks + embeddings saved in a bundled vector store called a Corpus.
- Corpus: You can have multiple Corpus databases, representing different projects.
- Search: Input is embedded and matched to similar chunks
- Prompt: LLM gets context-rich prompt to generate completion
- Complete: You write — Ghostwriter listens and responds
- Ghostwriter uses both inference and RAG-based patterns for collaborating with you and keeping subject matter on point
- Ingest: Drop .txt, .md, .epub, or .pdf files
- URLs: Ingest from a URL
- Chunk: Text is divided into overlapping chunks
- Embed: Each chunk gets a unique vector representation produced by the appropriate embedding model
- Store: Chunks + embeddings saved in a bundled vector store called a Corpus.
- Corpus: You can have multiple Corpus databases, representing different projects.
- Search: Input is embedded and matched to similar chunks
- Prompt: LLM gets context-rich prompt to generate completion
- Complete: You write — Ghostwriter listens and responds
- Visual Memory Overlay — see what Ghostwriter sees when it completes.
- Persona Cartridges — swap in distinct voices, tonalities, and attitudes.
- Multi-agent Collaboration — have Ghostwriter integrate with Conversseract, and form an Pilshaw compliant Oraculator network — all in one interface.
-
Visual Memory Overlay — see what Ghostwriter sees when it completes.
-
Persona Cartridges — swap in distinct voices, tonalities, and attitudes.
-
Multi-agent Collaboration — have Ghostwriter integrate with Conversseract, and form an Pilshaw compliant Oraculator network — all in one interface.
-
Visual Memory Overlay — see what Ghostwriter sees when it completes.
-
Persona Cartridges — swap in distinct voices, tonalities, and attitudes.
-
Multi-agent Collaboration — have Ghostwriter integrate with Conversseract, and form an Pilshaw compliant Oraculator network — all in one interface.
2025-02-04T22:50:52-08:00
Seems to build. Communicates from front to back to front again.
Lots of fiddling to get it to work.
Consists of Vite as well.
Moved Tauri backend from the Ghostwriter Tauri Sandbox repo and after much fiddling got it to build — but haven't yet run everything through.
This template should help get you started developing with Tauri in vanilla HTML, CSS and Javascript.