Snippbot

Your own AI agents that actually do the work.

Snippbot is a self-hosted autonomous AI agent platform. Install it on your own machine or server, connect your AI providers, and let agents research, build, automate, and execute — while you focus on strategy. Everything runs on your infrastructure; your prompts, data, and API keys never leave your control.

It ships as a single, self-contained daemon — a Starlette API server and a bundled React UI in one process on port 18781. The only runtime requirement is Python 3.11+: the compiled UI and the agent gateway are bundled inside the package, so there's no Node, pnpm, or repo checkout to manage.

Install

pipx install snippbot      # recommended — isolated environment, `snippbot` on PATH

pipx works the same on macOS, Linux, and Windows. Don't have it? Install it from pipx.pypa.io. Prefer pip? Use a virtualenv (modern macOS/Debian block a system pip install per PEP 668):

python3 -m venv ~/.snippbot-venv && ~/.snippbot-venv/bin/pip install snippbot

Optional extras — voice, browser automation, vector memory, and more:

pipx install "snippbot[voice]"     # or [all] for everything

Quick start

snippbot setup       # create your account + awaken your first agent (browser wizard)
snippbot start       # start the daemon (API + UI on :18781)

Then open http://localhost:18781/. A quick health check:

curl http://127.0.0.1:18781/health     # {"status": "ok", "service": "snippbot"}

snippbot start runs in the foreground. To launch on login and auto-restart on crash, register it as a user-level service (systemd on Linux, launchd on macOS, Task Scheduler on Windows) — no sudo required:

snippbot service install     # install + start
snippbot service status      # check it

To reach it from other machines (server/VM), bind to all interfaces and open ports 18781 (API + UI) and 8787 (agent gateway) in your firewall:

snippbot start --host 0.0.0.0

What Snippbot does

• Multi-agent orchestration — Architect → Executor → Reviewer team loop, each agent configurable with its own identity, memory, tools, and trust tier. Spawn sub-agents on demand for research, coding, or analysis.
• Projects & workflows — AI-generated plans with phases, tasks, and approval gates; a drag-and-drop DAG workflow builder with conditional branching and parallel execution.
• Scheduler — Cron, natural language (every day at 9am), one-time triggers, chains, conditions, and auto-pause on failure.
• Goal-seeking loops — Autonomous runs that iterate until a goal is met (verifiable shell check, LLM judge, or hybrid) or a budget rail stops them; manual / scheduled / event triggers, 7 built-in templates, chaining, and push/Slack notifications.
• Persistent memory — Layered context with episodic store, full-text + vector search, knowledge graph, automatic consolidation, and end-of-conversation reflection that updates the agent's notebook.
• Thinking engine & proactivity — Autonomous reasoning cycles that generate insights and track goals, with quiet hours and rate limits.
• Browser automation — Full Chrome DevTools control with recording/replay, stealth mode, and persistent sessions.
• Multi-channel messaging — Slack, Discord, Telegram, WhatsApp, Microsoft Teams, and Google Chat. Plus email and push notifications.
• Security & sandboxing — 5 permission tiers plus two isolation layers: a container engine (Docker/Podman) for untrusted code and an OS-level fence (macOS Seatbelt / Linux bubblewrap or firejail) for the agent's own commands. DLP scanning, encrypted secret storage, and a security scanner.
• Voice & devices — Local and cloud STT/TTS; pair and drive remote devices over a WebSocket protocol.
• Marketplace — The Singularity marketplace for installing skills, tools, and agent packages, buying packages, and funding bounties in USD via Stripe.

Coming soon: USDC payments + agent autonomy — bounded autonomous agent payments (USDC/x402 on Solana) with per-transaction/daily caps, allowlists, and escrow. In development.

Optional: Claude CLI agent backend

To run agents on your Claude Pro/Team account via the local Claude CLI, you also need Node.js 18+ and the @anthropic-ai/claude-code package:

npm install -g @anthropic-ai/claude-code   # global install needs sudo on a server
claude login                               # then authenticate

This is optional — you can skip it and run on an Anthropic (or other provider) API key instead.

Configuration & data

All state lives under ~/.snippbot/ (config, SQLite database, agent workspaces):

• ~/.snippbot/config.toml — main config
• SNIPPBOT_DATA_DIR — override the data directory
• SNIPPBOT_HOST / SNIPPBOT_PORT — override the bind address / port (default 127.0.0.1:18781)

Resolution order: environment variables → ~/.snippbot/config.toml → built-in defaults.

CLI

snippbot setup        # first-run wizard            snippbot start|stop|status
snippbot doctor       # diagnose system health      snippbot service install
snippbot config list  # view config                 snippbot secrets set KEY
snippbot project ...  # projects                    snippbot marketplace ...
snippbot --version    snippbot uninstall            snippbot --help

Full command reference: docs.snippbot.com/cli.

Upgrading

pipx upgrade snippbot

Links

• Home: https://snippbot.com
• Install guide: https://snippbot.com/install
• Documentation: https://docs.snippbot.com
• CLI reference: https://docs.snippbot.com/cli/

License

Free for personal, educational, and non-commercial use. Commercial use requires a license — see the Terms of Service.