engineering / agent

iCode

A terminal app for orchestrating agents and workflows with your own model profiles, with diff, rollback and trajectory views.

At a glance

Purpose
A terminal app for orchestrating agents and workflows with your own model profiles, with diff, rollback and trajectory views.
Runs in
Local agent with shell
Operating systems
macOS, Linux, Windows
Needs
Runtime: uv (provisions Python 3.14); Needs internet: No; Needs accounts or API keys beyond your agent: Not stated
Files and data
Reads files: Not stated; Writes files: Yes; Sends data to other services: Yes
Tool type
Agent
Delivery
From GitHub
Version
Latest on the main branch
Access
Code on GitHub
Compatibility
The README describes uv installation for Windows (x64), macOS on Apple silicon, and Linux (x86-64 or ARM64, glibc 2.27 or later). The README directs Intel Mac, Windows on Arm and older Linux users to the offline package. The README recommends a modern terminal such as Windows Terminal on Windows or Ghostty on macOS and Linux. The README describes configured model providers, MCP servers, skills, hooks and OpenTelemetry export. The README's model screenshot description shows a DeepSeek profile using the Responses API.
External costs
The repository is Apache-2.0-licensed and states no pricing. The model provider you connect may charge for usage. The README says iCode includes no AI model and that every model it uses is one you connect.
Permissions
The README shows source setup installing iCode and development tools into .venv, downloading vendored ripgrep, and provisioning Python through uv. The README also gives a uv tool installation and offline installers that execute downloaded scripts through sh or PowerShell iex. The README requires Xcode Command Line Tools for the Apple-silicon uv route. The README describes configurable agent tools, sub-agents, skills, MCP servers and memory, plus a terminal shell and file changes made by agents. The README says shell commands, skill scripts and workflows run with your network access. The README says web tools are off by default and can be enabled per agent. The README describes rollback controls for reverting file changes and discarding later conversation turns. The README shows an approval-mode indicator but does not explain the modes.
Usage terms
Apache License 2.0, as stated in the repository. See its LICENSE file for the exact terms. The README adds that iCode bundles third-party software, each part under its own license, and lists them in a NOTICE file.

Code on GitHub

openJiuwen-ai/iCode

Apache-2.0

Python347 stars59 forksLast pushed 12 Oct 2026 UTC

A lightweight, extensible, fully offline development platform and agent/workflow toolkit for the AI era, with an intuitive TUI, easy deployment, and complete control over your data, agents, and workflows.

Public GitHub data fetched · 0 days old

View on GitHub ↗

What it does

iCode is a terminal app for orchestrating agents and workflows. The README says you connect your own models and edit agent configuration, including instructions, tools, sub-agents, skills, MCP servers and memory.

The README describes Markdown chat, node-by-node workflow views, switchable model profiles, unified or split file diffs, rollback controls, time and token trajectories, and a terminal shell. The README says F8 opens the built-in user guide.

The README labels iCode a work in progress and describes source, uv installation and offline-package routes. The README links documentation in English and Simplified Chinese.

Who it is for

People who want to orchestrate agents and workflows from a terminal using model providers they connect themselves.

Example tasks

Press F4 on first launch to add a model profile.
Edit the built-in Code Agent's instructions, tools, skills and MCP servers with F2.
Run a workflow and watch it node by node.
Review the diff of every file the agent changed, in unified or split view.
Roll back to an earlier turn and review the file changes that would be reverted.

Getting started

The README gives three ways to run iCode. For source setup, the README gives uv sync, ./scripts/fetch_rg.sh (or the Windows variant), then uv run icode. For PyPI installation, the README gives uv tool install iCode-TUI, then icode. For the offline package, the README provides shell and Windows installer scripts that download the package for your platform. For a machine without internet access, the README says to download a release package elsewhere, unpack it, run chmod +x ./icode on non-Windows systems, then ./icode install (Windows: .\icode.exe install) and start icode in a new terminal. The README says to press F4 on first launch to add a model profile.

Limitations

The README labels iCode a work in progress and says interfaces, file formats and defaults still change between releases. The README says it ships no model profiles; you add one on first launch. The README assigns responsibility for applicable compliance obligations, including under the EU AI Act, to users integrating models for business or engineering use. The README recommends a modern terminal such as Windows Terminal on Windows or Ghostty on macOS and Linux. The README says source setup requires uv, which provisions Python 3.14. For uv installation, the README lists Windows (x64) and Linux (x86-64 or ARM64, glibc 2.27 or later), and says Apple-silicon Macs need xcode-select --install first. The README directs Intel Mac, Windows on Arm and older Linux users to the offline package. The README says web tools are off by default. The repository can change at any time; check its README before you install.

Data practices

The README's Privacy section says iCode has no telemetry, analytics or crash reporting. The README identifies configured model providers, MCP servers, hooks and OpenTelemetry export as destinations for application requests. The README also describes a daily startup update check against PyPI or GitHub for the newest version number, says the request carries no personal information, and gives Check for updates in Settings as its opt-out.

The README says web tools are off by default. The README says enabling them sends search queries to your configured provider, or Exa's public search service when none is configured, and requests pages from their websites. The README says shell commands, skill scripts and workflows have your network access and determine their own outbound data.

Community experience

Not rated yet

Sign in to share your experience.

Comments 0

No comments yet.

Sign in to comment

Report a listing concern