bonobo.sh
./terminal · 0.1.0-beta

Bonobo Terminal pairs an xterm.js emulator with a chat interface. Runs any .gguf model via ik_llama.cpp or llama.cpp, or talks to any OpenAI-compatible endpoint. Written in TypeScript on Bun. Linux-first; ships as a Docker Compose bundle.

View on GitHubSee features ↓
InterfaceDual-pane workspaceChat and an xterm.js terminal side by side, sharing one context.
ModelsBring your own LLMLocal .gguf via llama.cpp, or any OpenAI-compatible remote endpoint. Switch mid-chat.
ContextSee what the model seesA 3D memory graph and knowledge map show what's in context.
Tools66 built-in toolsShell, SSH, MCP, memory, and web tools share one chat.
Application Highlights and Features

Look around.

Chat, a live context graph, knowledge base, projects, models, and skills · all in one workspace. Want to see how it works? Jump to architecture →

Architecture

How it works

Two moving parts do the heavy lifting on every turn: a decision flow that assembles the right prompt and tools for your message, and adaptive context that keeps the conversation lean. Step through each below.

Tool decision flow
FEEDBACK01 · INPUTUser message02 · TRIGGERKeyword match03 · ASSEMBLEDynamic prompt04 · INFERModel05 · RISKApproval gate06 · RESULTSExecute & return07 · COMPACT3-layer trim
Smart context
3 layer trim
Everything in context

Full message history, every tool result, and recalled memories — together they overflow the token budget.

Messages kept12 / 12Memories2Context18.2k tok
Memory graph · in context
userassistanttoolmemorysystem
sequencesimilaritymemory link
open source · MIT

Open source

Bonobo Terminal is self-hosted, open-source, and transparent. Build on it, break it, send PRs.

GitHub →