Manage fleets of AI agents from your phone. One binary, full control.

xnc orchestrates nullclaw AI agents — each running isolated in hardened Docker containers with persistent memory. Talk to them from the terminal, pipe messages through scripts, or run your entire fleet from Telegram.

The killer feature. Run xnc mux and your Telegram bot becomes a full agent control plane:

xnc mux start              # start the mux daemon

From Telegram you can:

• Talk to any agent — the mux AI routes your messages, starts agents on demand, and manages conversations
• Send voice messages — transcribed via Whisper, agents can respond with TTS
• Manage your fleet — start, stop, clone, rename, snapshot agents — all through natural conversation
• Install skills — drop a .md or .zip file in the chat, the mux installs it to any agent or all of them
• Track costs — ask "how much did alice spend today?" and get real answers
• Send files — upload documents, the mux delivers them to the right agent's workspace

Everything the CLI can do, the mux can do from your phone. No SSH, no terminal, no VPN.

xnc mux stop               # stop the daemon
xnc mux logs -f            # follow logs
xnc mux status             # check if running

asdf plugin add xnc https://github.com/huru-io/asdf-xnc.git
asdf install xnc latest
asdf set --home xnc latest

Manages versions automatically. Update with asdf install xnc latest.

# macOS (Apple Silicon)
curl -fsSL https://github.com/huru-io/xnullclaw/releases/latest/download/xnc_$(curl -fsSL https://api.github.com/repos/huru-io/xnullclaw/releases/latest | grep tag_name | cut -d'"' -f4)_darwin_arm64.tar.gz | tar xz -C /usr/local/bin

# macOS (Intel)
curl -fsSL https://github.com/huru-io/xnullclaw/releases/latest/download/xnc_$(curl -fsSL https://api.github.com/repos/huru-io/xnullclaw/releases/latest | grep tag_name | cut -d'"' -f4)_darwin_amd64.tar.gz | tar xz -C /usr/local/bin

# Linux (x86_64)
curl -fsSL https://github.com/huru-io/xnullclaw/releases/latest/download/xnc_$(curl -fsSL https://api.github.com/repos/huru-io/xnullclaw/releases/latest | grep tag_name | cut -d'"' -f4)_linux_amd64.tar.gz | tar xz -C /usr/local/bin

# Linux (arm64)
curl -fsSL https://github.com/huru-io/xnullclaw/releases/latest/download/xnc_$(curl -fsSL https://api.github.com/repos/huru-io/xnullclaw/releases/latest | grep tag_name | cut -d'"' -f4)_linux_arm64.tar.gz | tar xz -C /usr/local/bin

Or grab the right archive from the Releases page, extract, and put xnc somewhere on your PATH.

git clone https://github.com/huru-io/xnullclaw.git
cd xnullclaw
make build        # -> ./build/xnc
make install      # -> /usr/local/bin/xnc

• Docker (running)
• nullclaw container image

# Interactive wizard — sets up API keys, agents, optional Telegram bot
xnc init

# Pull the nullclaw Docker image
xnc image build

# Start an agent
xnc start alice

# Talk to it
echo "Summarize the Go 1.24 release notes" | xnc send alice

# Interactive session
xnc cli alice

Non-interactive setup for automation:

xnc init --openai-key sk-... -n 3 --name alice --name bob --name carol

nullclaw is the AI agent runtime — it handles LLM calls, tool execution, memory, and skills inside a container.

xnc is the control plane on top — it creates, configures, starts, stops, clones, snapshots, and orchestrates those agents from the outside. Think docker to nullclaw's application.

Every agent runs locked down:

• Read-only root filesystem
• All Linux capabilities dropped
• No privilege escalation
• /tmp mounted noexec (64 MB)
• 128 MB memory, 0.25 CPU, 64 PIDs
• Runs as host user, never root
• Restart policy: unless-stopped

Configure any combination of providers per agent:

• OpenAI (openai/gpt-5-mini)
• Anthropic (anthropic/claude-sonnet-4)
• OpenRouter (openrouter/...)

Keys validated at setup time with a lightweight /models probe.

Skills are instruction sets that teach agents new capabilities. Install from directories, zip archives, or plain markdown files:

xnc skill install ./code-review/          # directory with SKILL.toml + SKILL.md
xnc skill install coding-standards.md     # single markdown (name from # heading)
xnc skill install skills.zip              # zip archive

xnc skill install ./my-skill --agent bob  # one agent
xnc skill install ./my-skill --all        # shared + sync to all agents
xnc skill list --all                      # see what's installed

Shared skills (~/.xnc/skills/) are auto-installed to new agents. Agent-local skills override shared ones.

xnc snapshot alice                    # backup full state
xnc restore alice-20240315 alice-v2   # restore as new agent
xnc clone alice bob --with-data       # duplicate with conversation history

Per-agent LLM cost tracking with budget enforcement:

xnc costs alice --today
xnc costs alice --month --json

Rename with full identity propagation — filesystem, config, system prompt, and an identity-change message sent to the agent:

xnc rename old-name new-name

xnc init [flags]                          Interactive setup wizard

xnc setup    <names...> [flags]           Create agent(s)
xnc start    <agents...> [--port N]       Start containers
xnc stop     <agents...> [--all]          Stop containers
xnc restart  <agents...> [--port N]       Restart containers
xnc destroy  <agents...> [--yes]          Delete permanently
xnc clone    <source> <new> [--with-data] Clone an agent
xnc rename   <old> <new>                  Rename an agent

xnc send     <agents...> [--all]          Pipe stdin to agent(s)
xnc cli      <agent>                      Interactive chat
xnc logs     <agent> [-f] [--tail N]      Container logs
xnc drain    <agent>                      Drain buffered output
xnc watch    <agent>                      Stream live output

xnc cp-to    <agent> <file> [dest]        Copy file into container
xnc cp-from  <agent> <path> [dest]        Copy file out

xnc config   get <agent> [key]            Read config
xnc config   set <agent> <key> <value>    Write config
xnc persona  <agent|mux> [--show] [--reset] [--preset NAME] [--list-presets] [--trait TEXT] [--warmth N] ...  Personality editor
xnc costs    <agent> [--today|--month]    Cost summary

xnc skill    list [--agent N] [--all]     List skills
xnc skill    install <src> [--agent N]    Install skill
xnc skill    remove <name> [--agent N]    Remove skill
xnc skill    info <name> [--agent N]      Skill details

xnc status   [agents...] [flags] [--json] Agent status (default: all)
xnc list                                  Alias for: status
xnc running                               Alias for: status --running

Status flags: --running, --stopped, --error. Combine with agent names to filter further.

xnc snapshot  <agent>                     Create snapshot
xnc restore   <snapshot> [name]           Restore from snapshot
xnc snapshots [--json]                    List snapshots
xnc snapshot-delete <snapshot>            Delete snapshot

xnc mux                                  Check mux status (same as mux status)
xnc mux      start [--foreground]        Start Telegram bot daemon
xnc mux      stop                        Stop daemon
xnc mux      status                      Check status
xnc mux      logs [-f]                   View logs

xnc image    build [--from-source]       Pull or build image
xnc image    update [--from-source]      Update image
xnc image    status                      Image info

--home <path>     Override XNC_HOME (default: ~/.xnc)
--image <name>    Override Docker image
--json            JSON output
--quiet           Suppress output

• Single static binary (~12 MB). Pure Go, CGO_ENABLED=0, SQLite via modernc.org/sqlite.
• Docker SDK behind an Ops interface — only internal/docker/ imports it.
• 14 internal packages: agent, cli, config, docker, llm, logging, loop, media, memory, mux, prompt, telegram, tools, voice.
• Concurrency: flock serialization inside containers for safe concurrent access.
• Data: ~/.xnc/ (override with XNC_HOME or --home).

MIT License. See LICENSE for details.