sentoagent
Advanced tools
Agents sent to fight your battles. Self-improving AI agents powered by Claude Code.
Agents sent to fight your battles.
Self-improving AI agents that run 24/7. Controlled through Discord, Telegram, Slack, or iMessage. One command to set up. They learn, adapt, and get better over time.
Works with your Claude subscription (Pro, Max, Team, or Enterprise). No API keys, no per-token billing, no surprise costs. Other agent frameworks require pay-per-use API keys. Sentō runs on your existing Claude plan — flat rate, unlimited usage.
Built by Gabriel Gil. Powered by Claude Code.
npx sentoagent init
That's it. The interactive setup walks you through everything.
A 60-second walkthrough of init → first message → agent response is coming. In the meantime, the full flow is reproducible from the Quick Start: run npx sentoagent init, pick Discord, paste a bot token, and DM the agent hi — the onboarding conversation begins immediately.
After init, the sento command is available (run source ~/.bashrc or open a new terminal):
# Setup
npx sentoagent init # Set up a new agent (first time only)
sento update # Update Claude Code, plugins, CLI, and patches
# Monitor
sento status # Check if your agent is running and healthy
sento logs # View agent output (last 30 lines)
sento logs -n 100 # View more lines
sento logs --watchdog # View auto-restart history
# Configure
sento config # Change model, tokens, keys, settings
sento channels # Add/remove communication channels (Discord, Telegram, Slack)
# Skills
sento skills # List installed plugins and custom skills
sento skills install # Install a plugin from the marketplace
sento skills export # Export a custom skill to share
sento skills import # Import a skill from another agent
# Process Control
sento start # Start the agent
sento stop # Stop the agent
sento restart # Restart the agent
# Agent-to-Agent Communication
sento pair # Pair with another agent
sento agents # List your agent code and paired agents
# Diagnostics
sento doctor # Check for issues (12 checks)
sento doctor --fix # Auto-fix what's possible
sento init installs Claude Code, 17 plugins, persistent memory, and browser automationdocker compose up -d (see docker-compose.yml)# Clone the repo
git clone https://github.com/sentoagent/sento.git && cd sento
# Build the image
docker build -t sento .
# Run with Discord
docker run -d \
--name my-agent \
-e CLAUDE_TOKEN="sk-ant-oat01-..." \
-e AGENT_NAME="myagent" \
-e CHANNELS="discord" \
-e DISCORD_BOT_TOKEN="your-bot-token" \
-e SERVER_ID="your-server-id" \
-e AGENT_ROLE="General-purpose assistant" \
-e AGENT_PERSONALITY="Chill, helpful" \
-e AGENT_LANGUAGE="English" \
-e AGENT_CREATOR="Your Name" \
-e AGENT_TIMEZONE="America/New_York" \
sento
# Watch the setup (takes a few minutes on first run)
docker logs -f my-agent
# Once you see "Starting myagent (channels: discord)...", the bot is live!
Multi-channel:
-e CHANNELS="discord,telegram" \
-e DISCORD_BOT_TOKEN="..." \
-e TELEGRAM_BOT_TOKEN="..." \
-e TELEGRAM_CHAT_ID="your-chat-id" \
Telegram only:
-e CHANNELS="telegram" \
-e TELEGRAM_BOT_TOKEN="..." \
-e TELEGRAM_CHAT_ID="your-chat-id" \
About
CLAUDE_TOKEN: This is your Claude Code OAuth token (prefixsk-ant-oat01-), not an Anthropic API key. It authenticates Claude Code against your existing Claude subscription — no per-token billing. Get it by runningclaude setup-tokenlocally after signing into Claude Code, or copy it from~/.claude/.credentials.json. Treat it like a password.
First run takes a few minutes. Docker installs plugins, ClawMem, and generates embeddings. Watch progress with
docker logs -f my-agent. Subsequent starts are fast.
Multi-agent? Just add more services to docker-compose.yml. Each agent gets its own container, volumes, and port.
~/.bashrc, ~/workspace/start-agent.sh). Only run on a machine you control.--dangerously-skip-permissions, giving it full shell access. It can read, write, and delete any file your user account has access to.allowFrom config (see below) to restrict access.dmPolicy: "allowlist") to prevent exploitation.Sentō only makes network calls to:
api.anthropic.com) — Claude Code uses your subscriptionregistry.npmjs.org) — on first run, to install the 17 plugins listed in the architecture sectionapi.github.com) — only when you use git operations or install plugins from the marketplacecontext7.com) — only when the agent looks up library documentationGEMINI_KEY for memory embeddings (optional)No telemetry. No analytics. Sentō does not phone home. Guardian and Watchdog make zero network calls.
The agent reads an allow-list from ~/.claude/channels/<platform>/access.json. Example restricting a Discord channel to two user IDs:
{
"dmPolicy": "allowlist",
"allowFrom": [],
"groups": {
"123456789012345678": {
"requireMention": false,
"allowFrom": ["USER_ID_1", "USER_ID_2"]
}
}
}
An empty allowFrom: [] means "anyone with channel access." Run npx sentoagent channels to edit interactively.
sento update after Claude Code updates to re-apply patches.Your phone/PC (Discord, Telegram, Slack)
|
v
Claude Code (on your machine, in tmux)
|
+-- Messaging Plugin (Discord/Telegram/Slack/iMessage)
+-- ClawMem (persistent memory)
+-- Playwright (browser automation)
+-- Context7 (docs lookup)
+-- 17 Plugins (skills, code review, etc.)
+-- CLAUDE.md (agent identity + rules)
+-- ~/workspace/skills/custom/ (self-created skills)
+-- Cron jobs (scheduled tasks)
+-- Watchdog (auto-restart on stuck/errors every 5 min)
Sentō agents are designed to never go down. Two layers of protection:
Guardian Bot — A lightweight Node.js process that runs alongside the agent:
Watchdog — A bash cron job (every 5 minutes) as a backup:
To set up Guardian notifications, run sento config → Set Discord webhook.
Sentō agents can message each other securely — the first agent framework with this feature.
How it works:
SENTO-A3F8K2D1) during setupsento pair and enter your codeSecurity:
MIT
FAQs
Agents sent to fight your battles. Self-improving AI agents powered by Claude Code.
We found that sentoagent demonstrated a healthy version release cadence and project activity because the last version was released less than a year ago. It has 1 open source maintainer collaborating on the project.
Did you know?
Socket for GitHub automatically highlights issues in each pull request and monitors the health of all your open source dependencies. Discover the contents of your packages and block harmful activity before you install or update your dependencies.
Product
Socket Firewall blocks malicious VS Code and Open VSX extensions before install, protecting developers from compromised editor marketplaces.
Research
More than 140 Mastra npm packages were compromised in a supply chain attack that used a typosquatted dependency to deliver a cross-platform infostealer during installation.
Research
/Security News
A new npm package tests AI malware scanners with prompt injection, safety-triggering comments, context flooding, and obfuscated JavaScript.