Scribe — Universal local-LLM research & writing agent

What it does

Universal LLM adapter

Point it at any OpenAI-compatible endpoint — local llama.cpp, Ollama, LM Studio, or remote.

Rich TUI & web UI

A terminal interface (Rich + Textual) and a streaming FastAPI web chat with login.

Cross-session memory

SME (Semantic Memory Engine) recalls your last session so context survives restarts.

RAG over your docs

Local embeddings (multilingual-e5) + LanceDB for semantic search across your library.

Sandboxed file tools

The model gets write/read/list tools scoped strictly to a per-user workspace.

Modular skills

Deep-research, writer, and wiki-memory skills — drop in more as SKILL.md modules.

Quick start

# 1. Clone & install
git clone https://github.com/pedjaurosevic/scribe-ai.git
cd scribe-ai
./scripts/install.sh        # installs the package + config + ~/scribe-workspace

# 2. Start your llama-server (or use any OpenAI-compatible endpoint)
./scripts/start-server.sh

# 3. Chat in the terminal …
scribe chat

# … or launch the web UI
scribe web                  # http://localhost:8765

Configure the endpoint in ~/.config/scribe/config.toml or via SCRIBE_BASE_URL / SCRIBE_MODEL env vars.

Architecture

┌─────────────────────────────────────────┐ │ SCRIBE TUI / WEB │ │ Rich · Textual · FastAPI │ ├─────────────────────────────────────────┤ │ CORE KERNEL │ │ Session Manager · Skills · Config │ ├─────────────────────────────────────────┤ │ LLM ADAPTER LAYER │ │ OpenAI-compatible (llama.cpp) │ ├─────────────────────────────────────────┤ │ MEMORY LAYER │ │ SME (cross-session) · RAG (documents) │ ├─────────────────────────────────────────┤ │ TOOLS LAYER │ │ web_search · web_fetch · fs · bash │ └─────────────────────────────────────────┘

Philosophy

Wittgenstein — explicit command vocabularies so the model knows exactly what each action means.
Peirce — semiotic continuity: every response becomes the next sign in the interpretation chain, run as hidden reasoning.