aichs 0.4.3

A minimal visual harness for agentic coding.

aichs

aichs is a minimal visual harness for agentic coding. Adapt it to your workflows, not the other way around.

Open a repository, chat with a model, and let the agent work with your code through approved tools, git context, file references, skills, and extensions. aichs treats the agent as something you can shape: prompts, models, UI behavior, workflow defaults, skills, extensions, tools, hooks, badges, and panels are all meant to be customized.

Status

aichs is early software. It is intended for developers who are comfortable running a local app that can read a workspace and, with approval, edit files or run shell commands.

Why aichs exists

Most coding agents give you a workflow. aichs gives you a workbench. It is for developers who want to tune the prompts, tools, context, approvals, memory, and UI around how they already work.

Install From PyPI

Install the published command with pipx:

pipx install aichs

Start it from a repository:

cd /path/to/your/repo
aichs

Or pass the workspace explicitly:

aichs /path/to/your/repo

Run

By default, aichs honors the directory it was started from. Use aichs --last-workspace only when you want to reopen the previously saved workspace.

Conversation history is stored in user data, not in your repository. Each workspace gets a stable entry in AICHS_HOME/workspaces.json (default ~/.aichs/workspaces.json) and its chats are saved under AICHS_HOME/<workspace_id>/conversations/.

API keys can be configured in Settings -> Models or through environment variables. Requires Python 3.11+.

What It Does

Open a workspace folder, pick a model, and work in one window:

• Agentic coding: ask the agent to inspect files, explain code, search the repo, make edits, and run approved shell commands
• Approval-gated tools: read, edit, search, git context, shell commands, and extension tools with workspace-scoped paths
• Coding workspace: file tree, syntax-highlighted tabs, git status, diffs, file references, and conversation history
• Context management: usage view, auto-compaction, compacted summaries, and decision memory for long sessions
• Customization: configurable prompts and workflow defaults, slash-command skills, command palette, project/user extensions, custom tools, hooks, badges, and panels
• Conversation flow: streaming Markdown, vision-capable models, file mentions, edit/resend, queued messages, pinned chats, search, and export

Shell commands and extensions run as the current user; only enable extensions you trust.

Contributing

Contributor setup, source installs, tests, packaging, and release notes live in CONTRIBUTING.md.

Documentation

Topic
Configuration	docs/configuration.md
Custom model providers	docs/custom-models.md
Extensions and custom tools	docs/extensions.md
Slash-command skills	docs/skills.md
Compaction and decision memory	docs/compact.md
Performance north star	docs/performance-north-star.md
YUK user kits	docs/yuk.md

FAQ

What does aichs stand for?

It is a play on Hebrew "ichs" (איכס), from Arabic إخسا, roughly "yuck" or "ew." The name is a little cursed on purpose.

Why use aichs when Cursor, Claude Code, Codex, and other agent tools exist?

You may not need it. Those tools are wonderful. aichs is for when you want a small, local, hackable agentic coding workspace that is shaped around your own habits.

Is aichs trying to replace my IDE?

No. It is a companion workbench: one window for the agent conversation, repo context, files, diffs, approvals, custom prompts, and extension tools.

Can I contribute something?

Very much. Small fixes, weird ideas, extensions, docs, and sharp opinions are all welcome. See CONTRIBUTING.md.

Does the S feel forced?

Yes. It is.

License

MIT License. See LICENSE.