---
slug: "termiagent-mcp服务器"
source_type: "readme"
source_url: "https://cdn.jsdelivr.net/gh/ccq1/termiAgent@master/README.md"
repo: "https://github.com/ccq1/termiAgent"
source_file: "README.md"
branch: "master"
---
# termi

**A terminal Agent that works beside you — and keeps working when you leave.**

termi is a terminal Agent built into your real Shell. When you do not know a
command, ask termi immediately. It understands what you just did, can explain
the next step, or execute the command for you without taking you out of the
terminal.

termi also runs as a persistent background daemon. It can monitor services and
long-running tasks around the clock, even after you detach. For larger goals,
start a termi Team and let multiple Agents coordinate the work across your
terminal sessions.

![termi Shell-Agent workspace](https://github.com/ccq1/termiAgent/raw/HEAD/docs/images/termi-workspace.svg)

## What makes it different

- **Ask without leaving the terminal.** If you do not know the command, ask
  termi in the workspace or use `termi -q` from your current Shell.
- **The Shell stays a Shell.** Completion, history, SSH, tmux, REPLs, and
  interactive programs work as usual.
- **The Agent shares the scene.** It follows the current cwd, commands, output,
  remote session, and task instead of starting from an unrelated chat.
- **Monitoring never clocks out.** The daemon keeps sessions, monitors, and
  scheduled checks running 24/7 and reports their status on a local Web board.
- **Teams take on bigger goals.** Activate a termi Team to coordinate multiple
  Agents around one terminal task goal.
- **Automation remains supervised.** Auto Review, sandboxing, approvals, and
  audit logs keep authority with the user.

## Get started

termi currently supports Linux and requires Go 1.26+ and
[`bubblewrap`](https://github.com/containers/bubblewrap).

```bash
git clone https://github.com/ccq1/termiAgent.git
cd termiAgent
make build
./bin/termi
```

On the first run, termi opens a guided command-line setup for the Provider,
model name, API key, context window, and output limit. DeepSeek, OpenAI,
Anthropic, and OpenAI-compatible endpoints are supported. Run `termi setup`
whenever you want to configure the model again, or press `F2` inside the TUI.

## Use termi

### Work beside the Agent

Run commands on the left and talk to the Agent on the right. Press `Shift+Tab`
to switch focus.

```text
What did I just do?
Why did this service fail?
Monitor this service every five minutes.
Fix the configuration, but do not restart production.
```

If the main Shell is busy, the Agent can use a private sub-Shell for independent
inspection. If you are connected through SSH, it works in that remote Shell.

### Call it from any terminal

```bash
termi -q "What is using port 4000?"
termi -q "Clean safe caches in this directory"
termi -q --new "Start a fresh context"
```

Quick Mode follows the parent Shell's cwd and recent activity. For reliable
Bash history capture:

```bash
eval "$(termi shell-hook bash)"
```

### Monitor services around the clock

Press `Ctrl+B`, then `D`, to detach without stopping your sessions. Open
[`http://127.0.0.1:7331`](http://127.0.0.1:7331) for the local operations board.

![termi operations board](https://github.com/ccq1/termiAgent/raw/HEAD/docs/images/termi-board.svg)

The daemon continues running managed monitors and scheduled Agent checks after
you detach. The board shows active work, service health, idle sessions, and
operations waiting for approval.

### Activate a termi Team

Create a Goal, activate a Team, and assign multiple Agents to investigate,
execute, verify, and monitor different parts of the same terminal task. Team
progress remains visible from the operations board.

## Highlights

- Real PTY Shell-Agent sessions
- Persistent tabs, task queues, monitoring, and soft resume
- Quick AI from an ordinary Shell
- Operations memory for hosts, services, ports, topology, and runbooks
- Goals and multi-Agent Teams
- Web search and fetch
- Auto Review and Full Access modes
- Automatic model context compression
- Chinese and English UI/output

## Essential keys

| Key | Action |
| --- | --- |
| `Shift+Tab` | Switch Shell / Agent |
| `Alt+T` | New session |
| `Alt+←` / `Alt+→` | Switch session |
| `Alt+P` | Auto Review / Full Access |
| `Ctrl+G` | Stop Agent, keep Shell running |
| `Ctrl+B`, then `D` | Detach |
| `F2` | Settings |

## Learn more

- [Architecture](https://github.com/ccq1/termiAgent/blob/HEAD/ARCHITECTURE.md)
- [Quick Mode](https://github.com/ccq1/termiAgent/blob/HEAD/docs/PRD_QUICK_MODE.md)
- [Context compression](https://github.com/ccq1/termiAgent/blob/HEAD/docs/PRD_CONTEXT_COMPACTION.md)
- [Agent Teams and Goals](https://github.com/ccq1/termiAgent/blob/HEAD/docs/PRD_AGENT_TEAMS.md)
- [Security](https://github.com/ccq1/termiAgent/blob/HEAD/SECURITY.md)

```bash
make check
make install
```

termi is under active development. Release builds currently target Linux
`amd64` and `arm64`.