Skip to content

Paseru/sinew

Repository files navigation

Sinew — the coding harness you shape

Release License Build Downloads Discord

Sinew is a desktop AI coding harness you can actually reshape.
Rewrite the description of every tool, turn the ones you don't need off,
and the assistant only sees what you keep.

tauri 2 · react · rust · monaco · xterm · mcp


Why Sinew

Most AI coding tools ship a fixed harness: a hard-coded prompt, a hard-coded set of tools, a hard-coded loop. You get whatever the vendor decided is "right".

Sinew flips that. The harness is the surface area you control.

  • Every tool description is editable. Rephrase it, scope it down, change the contract.
  • Every tool is toggleable. Run minimal like Pi, or unlock the full set.
  • Every provider is pluggable. Same agent loop across Anthropic, OpenAI, Google, Kimi, OpenRouter.
  • Agents can run in swarms. Multiple sub-agents, one shared task board, message passing.
  • It's a real IDE. Monaco editor and xterm terminal, not a chat box with a file picker.

Shape the harness

Minimal harness vs full harness

The same agent loop can behave like a barebones pair-programmer or a full autonomous coding crew. You choose. Tool descriptions live in settings, versioned with your config, and the assistant gets exactly the surface area you decide on.

Minimal           →   read · patch · bash
Search-capable    →   + grep · glob
Web-aware         →   + WebSearch · WebFetch
Full              →   + image · mcp · todo · question · skill · subagent · team

No magic. No hidden tools. No prompt you can't see.


Agent swarm

Agent swarm with shared task board

Launch a team of sub-agents from a single prompt. Each agent gets its own context, its own role, its own tool subset. They coordinate through a shared task board and SendMessage. The main agent stays in the loop, can watch the board update live, stop any teammate, or hand work off.

  • Roles. Architect, frontend, backend, tester — or whatever you define.
  • Task board. Real dependencies, real ownership, real status.
  • Messaging. Direct (@name) or broadcast (*).
  • Live view. Streamed into the chat as cards, not buried in logs.

Multi-provider, one loop

Anthropic OpenAI Google Kimi OpenRouter

OAuth where possible, API keys where it makes sense. Switch provider mid-project without changing your tools, your prompt, or your workflow.


Built-in tools

Tool Purpose
read Read text files or attach images visually
apply_patch Create, update, delete, rename files via a strict patch format
grep Workspace-wide text / regex search
Glob Find files by glob pattern
bash Run shell commands in a managed session
WebSearch / WebFetch Hit the web, read the page
CreateImage Generate images (OpenAI Image, Nano Banana)
ToDoList Maintain a structured task list across turns
Question Ask the user single / multiple-choice questions
mcp Bind any Model Context Protocol server as tools
LoadMcpTool Lazy-load MCP tools on demand
Skill Long-form, on-disk skills the agent can invoke
subagent_* Delegate to a configured sub-agent
TeamRun / TeamStatus / TeamStop Drive the agent swarm

All of them are listed in Settings, every description is a textarea, every row has an on/off toggle.


Screenshot

Sinew IDE


Architecture

Sinew architecture

  • src/ — React UI (Monaco, xterm, chat, settings, file tree).
  • src-tauri/ — Tauri 2 shell, IPC commands, workspace I/O, conversation store.
  • crates/sinew-core — Provider-agnostic types: messages, tools, streams.
  • crates/sinew-app — Agent loop, tool implementations, swarm, MCP, compaction.
  • crates/sinew-{anthropic,openai,google,kimi,openrouter} — Provider adapters (auth, wire, streaming).

How it compares

Sinew Cursor Claude Code Aider Zed AI
Native desktop app — (CLI) — (CLI)
Open source
Multi-provider
Editable tool descriptions
Toggle individual tools partial
Agent swarm + task board
MCP servers partial
Embedded terminal n/a n/a

Install

Grab the latest build for your OS from the releases page.

  • macOS.dmg
  • Windows.msi / .exe
  • Linux.AppImage / .deb

The app self-updates from GitHub releases.


Build from source

# requires Rust 1.80+ and Node 20+
# see https://tauri.app/start/prerequisites/ for platform deps

npm install
npm run tauri dev      # development
npm run tauri build    # release bundle

The repo is a Cargo workspace (crates/* + src-tauri/) plus a Vite + React frontend (src/).


OAuth credentials

Provider OAuth client IDs (and Google's client secret) are embedded in the source. This follows the standard practice for "installed applications" — the same approach used by tools like gcloud. These credentials are not treated as secret in this context.


Community

  • Discord — chat, support, share your harness configs
  • Issues — bugs and feature requests
  • Discussions — design, providers, MCP

License

MIT

Built with Tauri, Rust, and a stubborn refusal to ship a black-box harness.

About

Desktop IDE with built-in AI coding agents

Resources

License

Stars

Watchers

Forks

Packages

 
 
 

Contributors