原文
The Kubernetes for AI Agents
Deploy, orchestrate, and scale AI agents across your infrastructure.
One binary. No dependencies. From laptop to enterprise cluster.
Website • Quick Start • Deployment Modes • Features • Architecture
klaw is an open-source platform for deploying and managing AI agents at scale. Think of it as Kubernetes, but instead of containers, you're orchestrating intelligent agents that can code, research, communicate, and automate tasks.
# Interactive chat
klaw chat
# Start full platform (Slack bot + scheduler + agents)
klaw start# One-line install
curl -fsSL https://klaw.sh/install.sh | sh
# Or build from source
git clone https://github.com/klawsh/klaw.sh.git
cd klaw && make build
sudo mv bin/klaw /usr/local/bin/# Option 1: each::labs Router (300+ models, single API key)
export EACHLABS_API_KEY=your-key
# Option 2: OpenRouter (multi-provider gateway)
export OPENROUTER_API_KEY=your-key
# Option 3: Direct Anthropic
export ANTHROPIC_API_KEY=sk-ant-...# Interactive CLI chat
klaw chat
# Or start full platform with Slack integration
export SLACK_BOT_TOKEN=xoxb-...
export SLACK_APP_TOKEN=xapp-...
klaw startklaw supports multiple deployment modes to fit your needs:
Run everything on a single machine—perfect for development and small teams:
# Interactive chat
klaw chat
# Full platform (Slack + scheduler)
klaw start
# Run agent in container
klaw run coder --task "Fix the bug"Scale across multiple machines with controller-node architecture:
# Start the controller (central brain)
klaw controller start --port 9090
# Join worker nodes
klaw node join controller:9090 --token <token>
# Dispatch tasks across the cluster
klaw dispatch "analyze this codebase" --agent researcherRun agents in isolated containers for security and reproducibility:
# Build the container image
klaw build
# Run agent in container
klaw run coder --task "Review this PR"
klaw run coder --detach # Background
# Manage containers
klaw ps # List running
klaw logs coder # View logs
klaw stop coder # Stop containerUse any LLM through a single API with automatic provider selection:
klaw chat --model claude-sonnet-4-20250514
klaw chat --model gpt-4o
klaw chat --model gemini-proSupported Providers:
- each::labs Router - 300+ models through single API
- OpenRouter - Multi-provider gateway
- Anthropic - Direct Claude access
Organize agents with Kubernetes-style multi-tenancy:
# Create cluster
klaw create cluster production
# Create namespaces
klaw create namespace engineering --cluster production
klaw create namespace marketing --cluster production
# Switch context
klaw context use production/engineering
# Agents are scoped to namespaces
klaw create agent coder --namespace engineeringAgents come with powerful capabilities out of the box:
| Tool | Description |
|---|---|
bash |
Execute shell commands with 2-minute timeout |
read |
Read file contents |
write |
Create and write files |
edit |
Edit files with precise string replacement |
glob |
Find files by pattern |
grep |
Search file contents with regex |
web_fetch |
Fetch and process web content |
web_search |
Search the internet |
agent_spawn |
Create specialized sub-agents |
skill |
Install and manage agent skills |
cron |
Schedule recurring tasks |
Composable capability bundles for agents:
klaw skill list
klaw skill install browser
klaw create agent researcher --skills web-search,browser,code-execAutomate recurring tasks:
# Create a cron job
klaw cron create daily-report \
--schedule "0 9 * * *" \
--agent reporter \
--task "Generate daily status report"
# List jobs
klaw cron list
# Manage jobs
klaw cron enable daily-report
klaw cron disable daily-reportDeploy agents to multiple surfaces:
klaw chat # Interactive CLI
klaw chat --tui # Rich TUI mode
klaw start # Slack bot + scheduler┌─────────────────────────────────────────────────┐
│ CHANNELS │
│ (Slack, CLI, TUI, API, Telegram) │
└──────────────────┬──────────────────────────────┘
│
┌─────────▼──────────┐
│ ORCHESTRATOR │
│ (Message Routing) │
└─────────┬──────────┘
│
┌─────────▼──────────┐
│ AGENT │
│ (LLM + Tool Loop) │
└─────────┬──────────┘
│
┌──────┬───────┼───────┬──────┐
│ │ │ │ │
▼ ▼ ▼ ▼ ▼
BASH READ WRITE GREP WEB
┌─────────────────────────────────────┐
│ CONTROLLER │
│ ┌─────────────────────────────┐ │
│ │ Agent Registry │ │
│ │ Task Dispatcher │ │
│ │ State Manager │ │
│ └─────────────────────────────┘ │
│ │ │
│ │ TCP/JSON │
│ ▼ │
└──────────────┬──────────────────────┘
│
┌──────────────┼────────────────────────┐
│ │ │
▼ ▼ ▼
┌─────────────┐ ┌─────────────┐ ┌─────────────┐
│ NODE 1 │ │ NODE 2 │ │ NODE 3 │
│ ┌─────────┐ │ │ ┌─────────┐ │ │ ┌─────────┐ │
│ │ Agent A │ │ │ │ Agent C │ │ │ │ Agent E │ │
│ └─────────┘ │ │ └─────────┘ │ │ └─────────┘ │
└─────────────┘ └─────────────┘ └─────────────┘
Just as Kubernetes revolutionized container orchestration, klaw brings the same paradigm to AI agents:
┌─────────────────────────────────────────────────────────────────────────┐
│ KUBERNETES vs KLAW │
├─────────────────────────────────────────────────────────────────────────┤
│ │
│ KUBERNETES (Containers) KLAW (AI Agents) │
│ ════════════════════════ ═══════════════════ │
│ │
│ Container Image ────────► Agent Definition │
│ Pod ────────► Agent Instance │
│ Deployment ────────► AgentBinding │
│ Service ────────► Channel (Slack, CLI, API) │
│ Node ────────► Node (Worker Machine) │
│ Namespace ────────► Namespace (Isolation) │
│ ConfigMap ────────► SOUL.md / Config │
│ kubectl ────────► klaw CLI │
│ Scheduler ────────► Task Dispatcher │
│ CronJob ────────► klaw cron │
│ │
└─────────────────────────────────────────────────────────────────────────┘
The Problem klaw Solves:
| Challenge | Kubernetes Solution | klaw Solution |
|---|---|---|
| "Where does my workload run?" | Schedules pods to nodes | Schedules agents to nodes |
| "How do I scale?" | HPA/VPA for containers | Spawn agents on demand |
| "How do I isolate?" | Namespaces | Namespaces + workspaces |
| "How do I update?" | Rolling deployments | Hot-reload agent configs |
| "How do I communicate?" | Services + Ingress | Channels (Slack, API) |
| "How do I schedule?" | CronJobs | klaw cron |
Familiar Patterns:
# Kubernetes # klaw
kubectl get pods klaw get agents
kubectl create deployment klaw create agent
kubectl logs pod-name klaw logs agent-name
kubectl exec -it pod -- bash klaw attach agent-name
kubectl apply -f manifest.yaml klaw apply -f agent.toml
kubectl config use-context klaw context useklaw chat # Interactive terminal chat
klaw chat --tui # TUI mode with Bubble Tea
klaw start # Start platform (Slack + scheduler)klaw create agent <name> --model <model> --skills <skills>
klaw get agents
klaw describe agent <name>
klaw delete agent <name>klaw build # Build container image
klaw run <agent> --task "..." # Run in container
klaw run <agent> --detach # Run in background
klaw ps # List containers
klaw logs <container> # View logs
klaw stop <container> # Stop container
klaw attach <container> # Attach to containerklaw controller start # Start controller
klaw node join <addr> # Join a controller
klaw dispatch "task" # Dispatch task
klaw get nodes # List nodes
klaw get tasks # List tasksklaw create cluster <name>
klaw create namespace <name> --cluster <cluster>
klaw get clusters
klaw get namespaces
klaw context use <cluster>/<namespace>klaw skill list
klaw skill install <name>
klaw skill uninstall <name>
klaw skill show <name>klaw cron create <name> --schedule "..." --agent <agent> --task "..."
klaw cron list
klaw cron enable <name>
klaw cron disable <name>
klaw cron delete <name>~/.klaw/config.toml:
[defaults]
model = "claude-sonnet-4-20250514"
agent = "default"
[workspace]
path = "~/.klaw/workspace"
[provider.eachlabs]
api_key = "${EACHLABS_API_KEY}"
[provider.anthropic]
api_key = "${ANTHROPIC_API_KEY}"
model = "claude-sonnet-4-20250514"
[provider.openrouter]
api_key = "${OPENROUTER_API_KEY}"
[channel.slack]
enabled = true
bot_token = "${SLACK_BOT_TOKEN}"
app_token = "${SLACK_APP_TOKEN}"
[server]
port = 8080
host = "127.0.0.1"
[logging]
level = "info"~/.klaw/
├── config.toml # Main configuration
├── workspace/ # Agent workspace files
│ ├── SOUL.md
│ ├── AGENTS.md
│ └── TOOLS.md
├── agents/ # Agent definitions (TOML)
├── skills/ # Installed skills
├── sessions/ # Session history
└── logs/ # Log files
# Provider API Keys
ANTHROPIC_API_KEY # Anthropic Claude
OPENROUTER_API_KEY # OpenRouter gateway
EACHLABS_API_KEY # each::labs router
# Channel Tokens
SLACK_BOT_TOKEN # Slack bot
SLACK_APP_TOKEN # Slack app
# Overrides
KLAW_MODEL # Default model
KLAW_STATE_DIR # State directory| Feature | klaw | Claude Code | LangChain | AutoGPT |
|---|---|---|---|---|
| Open Source | Yes | No | Yes | Yes |
| Single Binary | Yes | No | No | No |
| No Dependencies | Yes | No | No | No |
| Distributed | Yes | No | No | No |
| Container Support | Yes (Podman) | No | No | No |
| Multi-Agent | Yes | No | Yes | Yes |
| Namespaces | Yes | No | No | No |
| Slack/Teams | Yes | No | Manual | No |
| 300+ Models | Yes | No | Yes | No |
| Cron Jobs | Yes | No | No | No |
| Production Ready | Yes | Yes | Partial | No |
git clone https://github.com/klawsh/klaw.sh.git
cd klaw
make build # Build binary
make test # Run tests
make fmt # Format code
make lint # Lint code
make cross # Cross-compile (Darwin, Linux, Windows)- Language: Go 1.24+ (no runtime dependencies)
- CLI Framework: Cobra
- TUI: Bubble Tea, Lipgloss, Bubbles
- LLM SDKs: anthropic-sdk-go, openai-go
- Config: TOML
- Containers: Podman
- IPC: gRPC, JSON over TCP
klaw/
├── cmd/klaw/
│ ├── main.go # Entry point
│ └── commands/ # CLI commands
├── internal/
│ ├── agent/ # Agent logic and definitions
│ ├── channel/ # Channel implementations
│ ├── config/ # Configuration management
│ ├── cluster/ # Namespaces and clusters
│ ├── controller/ # Distributed controller
│ ├── memory/ # Workspace memory system
│ ├── node/ # Worker node
│ ├── orchestrator/ # Message routing
│ ├── provider/ # LLM providers
│ ├── runtime/ # Podman runtime
│ ├── scheduler/ # Cron scheduling
│ ├── skill/ # Skills system
│ ├── tool/ # Tool implementations
│ └── tui/ # Bubble Tea TUI
└── proto/ # gRPC definitions
We welcome contributions! See CONTRIBUTING.md for guidelines.
git clone https://github.com/klawsh/klaw.sh.git
cd klaw
make build
make test
./bin/klaw chatklaw is source-available under the each::labs License.
Free to use for:
- Internal business purposes
- Personal projects
- Building your own AI applications
- Consulting and professional services
Requires license for:
- Multi-tenant SaaS offerings
- White-label/OEM distribution
See LICENSE for details. For enterprise licensing: [email protected]
klaw - The Kubernetes for AI Agents
Built by each::labs