HyperChat: local AI agents with MCP, CLI and a web UI
HyperChat
A local AI agent platform: agents, memory and MCP servers live in the project's .hyperchat folder, usable from the command line or the browser
- 708 ★ on the repository
- Last commit
Broad access
Why this level
- Agents call MCP tools and can work with local files and the terminal, and call confirmation is set by the `isConfirmCallTool` option
- Provider API keys live in `.env` files that easily end up in git along with the `.hyperchat` folder
- The web interface exposes agent control over the network: without a password and a local-only binding, anyone on the network gets access
Install
In your terminal, with SkillFoxx CLI
npx skillfoxx add cli/hyperchatDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Agents call MCP tools and can work with local files and the terminal, and call confirmation is set by the `isConfirmCallTool` option
- Provider API keys live in `.env` files that easily end up in git along with the `.hyperchat` folder
- The web interface exposes agent control over the network: without a password and a local-only binding, anyone on the network gets access
This is third-party code. Review the repository files before installing.
What it does
HyperChat describes AI helpers with files you can keep in git. A project's `.hyperchat` folder holds `mcp.json` with a pool of MCP servers, `ai_models.json` with models, `.env` and an `agents` directory. Each agent consists of an `agent.yaml` with a name, model, a list of allowed MCP servers and a prompt, a `memory.md` memory file, a conversation log and a `commands` folder with Markdown template commands. You can copy the folder to another project and the agent travels with it. From the command line an agent starts with no extra steps: `hyperchat "question"` calls the default model, `hyperchat agent name "task"` runs a specific agent and loads its MCP tools, and `hyperchat agent name chat` opens an interactive dialogue. A request can refer to several files through `@./path`, and a command like `/bug-fix @./src/login.ts description` drops the request into the `commands/bug-fix.md` template in place of `$ARG`. `hyperchat serve` starts a web interface on port 16100 with tabs for several workspaces, each with its own agents, MCP servers and history. Settings come from environment variables in five layers, from built-in defaults to command-line flags. There is an Electron app too, and models connect through OpenAI-compatible addresses and provider keys.
Who it is for. For people who want to keep a project's AI helpers as configuration in git and run them from a terminal, scripts and CI, with a web interface as an option.
Good fit when
- You need project agents with separate memory, prompt and MCP server set that a team stores and versions along with the code
- You need to run an agent from a script or CI with one command, for example `hyperchat agent test-runner "run all tests and make a report"`
- You prefer short Markdown template commands like `/review` and `/bug-fix` to long prompts every time
- You need to separate projects: each workspace has its own agents, MCP servers and history, and they do not mix
- You want to switch model providers through environment variables without changing the agents
Not a fit when
- You need a ready coding tool inside Claude Code or Cursor: HyperChat runs as a standalone client and does not act as a plugin for them
- You plan a multi-user service with registration and sign-in: the license forbids that use of the source without written permission
- You want to remove the logo and copyright information from the interface: the license forbids it
- You need documentation in English: the main README is written in Chinese
Example request
hyperchat agent coder "/review @./src/api/user.js"Limitations
The license is Apache 2.0 with additional conditions, and other commercial use is allowed.
The first version was written by hand and is being migrated to the second, so documentation lags behind the code and some descriptions are marked by the author as new features. The main README is in Chinese, and an English `README.1.md` covers the old branch. An API key is needed for the model provider you choose: Claude is recommended and Kimi K2 performs well.
The author disclaims responsibility for actions performed with the project, such as scraping. The web interface listens on a local address by default, and the password is set with the `HYPERCHAT_WEB_PASSWORD` variable.
How to disable. Remove the package with `npm uninstall -g @dadigua/hyperchat` and delete the `.hyperchat` folders in your projects and `~/Documents/HyperChat` by hand if the agents and history are no longer needed.
FAQ
How do I install HyperChat?
Globally with `npm install -g @dadigua/hyperchat` or with no install through `npx -y @dadigua/hyperchat`. Then set `HyperChat_API_KEY`, `HyperChat_API_URL`, `HyperChat_AI_Provider` and `HyperChat_AI_Model` and run `hyperchat "hello"`.
How do I create my own agent?
Run `hyperchat agent create name`. The command creates an agent folder with `agent.yaml`, memory and a `commands` directory. The model, the `allowMCPs` list and the prompt are set in `agent.yaml`.
How do I connect MCP servers?
Describe them in the workspace's `.hyperchat/mcp.json` or in the global `mcp.json` under `~/Documents/HyperChat/.hyperchat`. An agent takes only the servers listed in its `allowMCPs` and falls back to the workspace pool when it has none of its own.
How do I open the web interface?
Run `hyperchat serve` and open `http://localhost:16100`. Set the access password with the `HYPERCHAT_WEB_PASSWORD` variable or the `--password` flag.
Can I use HyperChat in a commercial project?
Yes, the Apache 2.0 license allows commercial use, including as a backend for other applications. The exceptions: a multi-user environment with registration needs written permission, and the logo and copyright information must stay.
Discussion
No questions or reviews yet.
Related
CLI toolsA desktop app with a local MCP server that lets ChatGPT work with files, commands, skills and other MCP servers
A tmux-based TUI that runs Claude Code, Codex and other agents in one list: status, quick prompts, worktrees and diff review
A local encrypted Slack and Discord alternative where AI agents work in channels alongside people through REST or MCP over a P2P mesh
Direct messages, groups and a durable inbox for AI agents on your own server, free within Cloudflare limits
A CLI that lets agents in different terminals message, watch and spawn each other
A CLI agent that works with project files and shell commands, lets you choose a model provider and connects external MCP tools.
README in short
The description presents the project as an "AI as code" platform: everything about an agent is set in files and versioned in git. It then compares two modes, the web one with several workspaces and the agent-first CLI, and walks through the configuration structure with `agent.yaml` and command examples.