Run it right now.

Use npx to try any command instantly. Or install globally for the short sisu alias.

Via npx (no install)

terminal
$ npx @sisu-ai/cli list tools
$ npx @sisu-ai/cli info vector-vectra
$ npx @sisu-ai/cli create chat-agent my-app
$ npx @sisu-ai/cli install skill
$ npx @sisu-ai/cli chat

After global install

terminal
$ npm install -g @sisu-ai/cli

# Then use the short alias:
$ sisu list tools
$ sisu chat
$ sisu --version

Everything you need, one command away.

Use the CLI as your configurable terminal agent: discover framework packages, scaffold agents, manage skills, and run chat with explicit control.

Discover

Explore every maintained Sisu package by category before writing a line of code.

$ sisu list tools
$ sisu list middleware
$ sisu list adapters
$ sisu list skills
$ sisu info mw-rag

Categories: libraries · middleware · tools · adapters · vector · skills · templates

Scaffold

Generate a maintained, working starter project in seconds. Three templates ready to go.

# create chat agent
$ sisu create chat-agent my-app

# single-shot CLI agent
$ sisu create cli-agent my-app

# local RAG with Vectra
$ sisu create rag-agent my-app

Install Skills

Install official or custom skills to your project or global profile.

$ sisu install skill
$ sisu install-skill @sisu-ai/skill-debug --project
$ sisu install-skill @sisu-ai/skill-repo-search --global
$ sisu list-official skills

Chat

Interactive AI chat in your terminal. Auto-detects local Ollama models — no API key needed.

$ sisu chat
$ sisu chat --prompt "show me git status"
$ sisu chat --session <session-id>
$ echo "hello" | sisu chat

A proper AI shell for daily work.

sisu chat is a first-class interactive chat mode for terminal-native agent workflows. Start it, and it auto-detects your local Ollama models — no API key, no config, no fuss.

sisu chat
✦ SISU Chat — session: abc123
Provider: ollama · Model: qwen3.5:9b
 
you show me git status
Running tool: terminal
✓ On branch main
nothing to commit, working tree clean
 
you /provider openai
→ Switched to openai · gpt-4o-mini
 
you /skills
enabled: skill-debug, skill-code-review

In-chat commands

/provider Switch AI provider interactively (ollama, openai, anthropic)
/model Pick a model from an interactive list
/tools List tools and their enabled state
/skills List skills and their enabled state
/enable <id> Enable a capability with session/project/global scope
/sessions List persisted sessions; resume or delete them
/branch <id> Fork a new session from any prior message
/search Search conversation history
/new · /exit Start a fresh session or close chat
⌨️

Ctrl+O

Open options menu: new session, switch session, branch, help, exit.

⌨️

Shift+S

Open settings: switch provider, model, or session.

⌨️

Shift+Enter

Insert a newline for multiline messages. Ctrl+J as terminal fallback.

⌨️

↑ / ↓ / Esc

Navigate and close menus. Markdown-aware assistant output rendering.

Policy-gated tool execution.

Every tool call is checked against a policy before execution. High-impact commands require confirmation by default. Denied and completed actions are persisted in session records.

allow

Command runs immediately, no prompt.

confirm

Explicit user approval is required before execution.

deny

Command is blocked with a reason. Logged to session record.

$ /allow-command git status session
$ /allow-command pnpm test project

Deterministic precedence.

1

Built-in defaults

Sensible starting point with balanced tool policy.

2

Global profile

~/.sisu/chat-profile.json — your personal defaults across all projects.

3

Project profile

<project>/.sisu/chat-profile.json — overrides global per project.

4

Session overrides

In-memory updates from interactive /provider, /model, and /enable commands.

Example chat profile.

.sisu/chat-profile.json
{
  "name": "default",
  "provider": "ollama",
  "model": "qwen3.5:9b",
  "theme": "auto",
  "toolPolicy": {
    "mode": "balanced",
    "requireConfirmationForHighImpact": true,
    "allowCommandPrefixes": ["echo", "ls", "git status", "pnpm test"]
  },
  "capabilities": {
    "tools": { "enabled": ["terminal"], "disabled": [] },
    "skills": {
      "directories": ["./.sisu/skills", "~/.sisu/skills"]
    },
    "middleware": {
      "pipeline": [
        { "id": "error-boundary", "enabled": true },
        { "id": "invariants", "enabled": true },
        { "id": "register-tools", "enabled": true },
        { "id": "tool-calling", "enabled": true },
        { "id": "conversation-buffer", "enabled": true },
        { "id": "skills", "enabled": true }
      ]
    }
  }
}

Provider notes

ollama — no API key, auto-detects local models.
openai — set OPENAI_API_KEY.
anthropic — set ANTHROPIC_API_KEY.

Ollama defaults

When no provider is configured, the CLI auto-detects local Ollama models.
Preferred: qwen3.5:9bllama3.1llama4qwen3.5:0.8b.

Open config

Edit profiles directly in your $EDITOR from inside chat:

/open-config project
/open-config global

Every conversation saved.
Resume anytime.

Sessions are persisted locally — messages, run state, tool lifecycle records, events. Resume a session, branch from any message, or search your history.

  • Deterministic restart and resume behavior
  • Session search and retrieval with /search
  • Branch-from-message lineage workflows with /branch
  • Persist command allow-list by scope: session, project, global
session management
# Resume a known session
$ sisu chat --session abc123

# Inside chat:
/sessions
  → abc123  2 hours ago
  → def456  yesterday

/resume def456
/branch msg-789
/search "git deploy"
/delete-session def456

Extend the CLI with skills.

Skills are reusable agent capabilities. Install official @sisu-ai/* skills or load your own from any directory.

🔎

skill-code-review

AI-powered code review with inline comments and suggestions.

🐛

skill-debug

Structured debugging: reproduce, trace, hypothesize, fix.

🚀

skill-deploy

Guided deployment workflows for common cloud targets.

💡

skill-explain

Explain any piece of code, error, or concept in plain language.

🔬

skill-repo-search

Semantic search across your codebase using local embeddings.

🧪

skill-test-gen

Generate unit and integration tests from function signatures.

terminal
# Install to project
$ sisu install-skill @sisu-ai/skill-debug --project

# Install globally
$ sisu install-skill @sisu-ai/skill-repo-search --global

# Install from local path
$ sisu install-skill ./my-custom-skill --dir ~/.sisu/skills

# Enforce official @sisu-ai/* namespace
$ sisu install-skill @sisu-ai/skill-debug --project --official

Why this exists.

Sisu already has a lot of maintained middleware, tools, adapters, and examples. The CLI is your configurable agent interface for real terminal work — with a direct path to discover framework packages before inventing custom code, plus built-in flows for skills, scaffolding, and chat with any provider.

View on GitHub Explore the Framework →