GRaPE Scribe
The terminal AI coding agent from Skinnertopia Lab for AI (SLAI).
Lighter than Claude Code. Faster than Codex. Built from the ground up for GRaPE models, with full tool access, streaming reasoning, and a TUI that stays out of your way.
In action
A terminal agent that actually works.
Model info, credit balance, context usage, and session stats in a compact header.
Parallel subagents, inline tool results, streaming reasoning, and a persistent input line.
From prompt to product
“Make a single-page website for yourself. Make it glorious.”
Scribe reads the project, writes the files, and streams the work as it goes. One instruction, a finished site.
Scribe reads the project, delegates to subagents, and writes index.html, styles.css, and main.js, then shows the result it shipped.
Delegator mode
One head agent. A team of subagents.
Scribe gives each subagent its own worktree, model, and task, then merges their work back into your project.
Audits changes for bugs and regressions
Implements the feature in isolation
Runs tests and validates output
Writes docs and summaries
Tab between agents, return to the head agent, or cancel a subagent with full visibility into every parallel task.
Make it yours
One command. Seven color schemes.
Scroll to sweep through every built-in theme.
Your models, your way
Your models. Your choice.
Run SLAI’s own GRaPE models in the cloud, download an open-source GRaPE build to run locally, sign in with ChatGPT (Codex), or point Scribe at any OpenAI- or Anthropic-compatible endpoint.
Connect ChatGPT (Codex), load a local GGUF, or add a custom API endpoint from one picker.
Choose a download that fits your hardware, including vision models. Already have a GGUF? Load it directly.
Why Scribe
Everything a coding agent needs.
GRaPE-first models
Purpose-built for GRaPE 2.5 models, with optimized context windows, prompting, and tool formats.
Full tool suite
bash, read_file, write_file, apply_diff, grep, list_directory, web_search, view_image, and a todo tracker built in.
Streaming thinking
Watch the model reason in real time. Toggle visibility with /show_thinking or set a thinking budget per session.
Permission modes
Lenient mode asks once. Turbo mode executes everything immediately. Fine-grained control over what Scribe can touch.
Conversation history
Every session is saved locally. Resume any past conversation with scribe -c or scribe --resume <id>.
One-shot mode
Pipe queries directly: scribe -p "fix the bug". Scriptable, CI-friendly, and supports JSON output format.
Auto-compact
When context fills up, Scribe summarizes the conversation automatically so you can keep working.
Slash commands
/model, /thinking_mode, /scheme, /compact, /history, /undo, /commit. A full command palette in your terminal.
Lightweight
A single statically linked binary. No Node, no Python runtime, no Docker; just download and code.
Install
Put Scribe in your terminal.
Choose your operating system, copy the command, and the installer will detect the right CPU architecture, download the latest published build, verify its SHA256, and run Scribe when it is finished.
curl -fsSL https://skinnertopia.com/install.sh | shThe installer chooses the matching platform and CPU architecture, verifies the published SHA256, and stops if the downloaded file is stale or altered.