LinuxCommandLibrary

zeroclaw

Autonomous AI agent runtime built in Rust

TLDR

Initialize configuration with interactive setup wizard

$ zeroclaw onboard --interactive
copy
Start an interactive CLI agent session
$ zeroclaw agent
copy
Send a single message without entering interactive mode
$ zeroclaw agent -m "[message]"
copy
Start the full autonomous daemon with channels and scheduler
$ zeroclaw daemon
copy
Check system health and run diagnostics
$ zeroclaw doctor
copy
List all supported LLM providers
$ zeroclaw providers
copy

SYNOPSIS

zeroclaw [command] [subcommand] [options]

DESCRIPTION

zeroclaw is a lightweight, security-first autonomous AI agent runtime built entirely in Rust. It serves as infrastructure for agentic workflows, abstracting models, tools, memory, and execution into a single binary that can be deployed across ARM, x86, and RISC-V architectures.
The runtime compiles to an approximately 9MB binary with cold start under 10 milliseconds and less than 5MB RAM usage at idle. It uses a trait-driven architecture where providers, channels, memory backends, and tools are swappable through configuration without code changes.
ZeroClaw includes a built-in hybrid search memory system combining vector embeddings with keyword search (SQLite-backed), requiring no external dependencies. It supports 28+ LLM providers including OpenRouter, Anthropic, OpenAI, Gemini, Ollama, and any OpenAI-compatible endpoint. Messaging integrations cover 70+ channels including Telegram, Discord, Slack, iMessage, Matrix, Signal, and WhatsApp.
The agent supports multi-turn conversations with context preservation, tool execution (shell, file, git, browser), scheduled tasks via cron expressions, and hardware peripheral control for IoT devices.

PARAMETERS

onboard

Initialize or reconfigure the workspace. Creates ~/.zeroclaw/config.toml and scaffold files.
onboard --interactive
Run the full 9-step configuration wizard.
onboard --api-key KEY --provider PROVIDER [--model MODEL]
Non-interactive setup with API key and provider in one command.
onboard --channels-only
Repair or reconfigure channels and allowlists only.
agent [-m MESSAGE]
Run interactive CLI agent or send a single message with -m.
agent --provider PROVIDER
Override the default provider for this session.
gateway [--port PORT]
Start the HTTP/WebSocket server for external integrations. Default port: 42617.
daemon
Start the full autonomous runtime with gateway, channels, heartbeat, and scheduler.
status
Display comprehensive system status including provider, memory, channels, and security.
doctor [models | traces]
Run system diagnostics. Optionally check model catalogs or inspect runtime traces.
service install | start | stop | status | restart
Manage zeroclaw as a background system service (systemd or OpenRC).
channel list | doctor | start
List, health-check, or start configured messaging channels.
channel add TYPE JSONCONFIG_
Add a new messaging channel configuration.
channel bind-telegram USERID_
Add a Telegram user to the allowlist.
auth login PROVIDER [PROFILE]
Authenticate via OAuth or device code flow.
auth status | logout PROFILEID_
Show or remove authentication profiles.
memory stats | list | search QUERY | delete KEY | prune
Manage the built-in memory system.
cron add NAME SCHEDULE MESSAGE [--tz TIMEZONE]
Schedule a recurring task with a cron expression.
cron list | remove NAME
List or remove scheduled tasks.
skills list | install NAME | remove NAME
Manage agent skills and extensions.
migrate openclaw [--dry-run]
Import memory from an OpenClaw runtime.
completions SHELL
Generate shell completions for bash, zsh, fish, powershell, or nushell.
providers
List all supported LLM providers and aliases.
estop [--resume]
Engage or resume from emergency stop.

CONFIGURATION

Configuration is stored in ~/.zeroclaw/config.toml. Run zeroclaw onboard --interactive for guided setup.

$ [providers]
default = "openrouter"

[providers.openrouter]
api_key = "sk-or-v1-xxx"
model = "anthropic/claude-opus-4-5"

[memory]
backend = "sqlite"

[runtime]
kind = "native"

[channels.telegram]
bot_token = "123:ABC..."
allowed_users = ["123456789"]
copy
Supported memory backends: sqlite (default, hybrid search), postgres, markdown, and none (stateless).
Environment variables override config values: ZEROCLAW_API_KEY, ZEROCLAW_PROVIDER, ZEROCLAW_MODEL, ZEROCLAW_WORKSPACE.

CAVEATS

ZeroClaw executes shell commands and file operations with your user permissions. The agent can modify files and run arbitrary commands when tool use is enabled. API keys are stored in config.toml and auth-profiles.json (encrypted at rest if secrets encryption is enabled). Channel integrations require the daemon to be running. Compilation from source requires at least 2GB RAM and 6GB disk space due to the Rust toolchain.

HISTORY

ZeroClaw was created by ZeroClaw Labs and first released in 2025 as a Rust-based alternative to existing AI agent runtimes. It was designed around the principle of zero overhead and zero lock-in, targeting deployment on resource-constrained hardware while maintaining production-grade extensibility. The project introduced a trait-driven architecture allowing providers, channels, and tools to be swapped without code changes, and gained attention for achieving sub-10ms cold start times with a single-binary deployment model.

SEE ALSO

nanobot(1), picoclaw(1), claude(1), openclaw(1)

> TERMINAL_GEAR

Curated for the Linux community

Copied to clipboard