Launch an isolated Sea-Claw agent on our K3s cluster in under 10 seconds — or self-host on your own Kubernetes. 77 tools, 7 channels, arena memory, Grammar Shield. Your data stays in your pod.
The gateway handles all K8s complexity. You just pick a model and soul.
POST to /api/v1/agents/create with your username, LLM API key, preferred model, soul, and optional channel tokens. Everything else is automatic.
The gateway calls the K8s API to create an isolated pod, PVC, and service for your agent. Your API key is injected as a K8s env var — never baked into an image.
Your pod starts the Sea-Claw binary with 77 tools, your chosen soul, and all channels wired. Talk to it via Telegram, Discord, API, or the chat UI immediately.
The gateway UI shows all running pods, lets you pause/restart agents, update the model, toggle souls, and stream live logs — all without touching kubectl.
Coordinator agents can call swarm_spawn to create ephemeral worker pods. Workers receive tasks via swarm_relay and are cleaned up automatically when done.
The gateway has an HPA configured (1→5 replicas). PodDisruptionBudget ensures zero downtime during rolling updates. LimitRange prevents any single pod from hogging resources.
A single K8s namespace containing everything — from the gateway RBAC to per-user PVCs.
Built on K3s. Designed for edge, homelab, and production alike.
The Sea-Claw binary is ~300KB. Container startup is dominated by kubelet scheduling, not app init. Your agent is answering messages in under 10 seconds.
Each user gets their own K8s pod, PVC, and NetworkPolicy. No shared memory, no shared filesystem. Data is physically isolated at the container boundary.
File I/O, shell, web search, Git, project management, cron, DNS, SSL, JSON, CSV, math, voice transcription. All compiled into the binary — no plugins, no eval().
Eva (analyst), Alex (developer), Tom (creative), Sarah (communicator), Max (generalist). Each soul shapes how the agent responds, prioritizes, and communicates.
Telegram, Terminal, HTTP API, WebSocket, Slack, Discord, Voice (Whisper). Wire multiple channels to the same pod — same brain, same 77 tools, any interface.
Coordinator pods spawn ephemeral worker pods via swarm_spawn. Workers communicate via swarm_relay. The gateway manages the K8s lifecycle automatically.
The gateway Deployment scales from 1 to 5 replicas when CPU hits 70%. PodDisruptionBudget ensures at least 1 replica stays up during rolling updates.
Run a single K3s node or a full multi-node cluster with NFS RWX storage. The setup-node.sh script joins worker nodes with correct role labels in one command.
./deploy.sh deploy builds both Docker images, imports them into K3s containerd, applies all manifests in order, and waits for the gateway rollout. Done.
Configure channels at signup. All are optional except TUI — which always works.
Bot token + chat ID. Message your agent from anywhere.
30+ slash commands. Always available inside the pod.
REST POST /api/chat. Integrate with any app.
Real-time streaming for dashboards and live UIs.
Outbound webhook. Agent responses go to your workspace.
Native bot polling. Set token + channel ID at signup.
Transcribe audio via Groq API or local whisper.cpp.
Everything the UI does, you can do via API. Automate agent creation, teardown, and config from your own tooling.
| Method | Endpoint | Description |
|---|---|---|
| POST | /api/v1/agents/create | Spawn a new agent pod |
| GET | /api/v1/agents | List all running pods |
| GET | /api/v1/agents/{user} | Get pod status + info |
| DEL | /api/v1/agents/{user} | Delete pod + PVC |
| PATCH | /api/v1/agents/{user}/config | Update model, soul, swarm mode |
| POST | /api/v1/agents/{user}/restart | Restart pod |
| POST | /api/v1/swarm/workers | Spawn worker pod (swarm) |
| DEL | /api/v1/swarm/workers/{name} | Delete worker pod |
| POST | /api/v1/swarm/relay | Send message to worker agent |
| GET | /health | Gateway health check |
How does our K8s agent platform compare to other ways of deploying AI agents?
| Feature | SeaClaw Platform ✦ | Docker Compose | Raw K8s YAML | Managed AI (Cloud) |
|---|---|---|---|---|
| Deploy complexity | ✓ 1 command | docker compose up | 10+ manual steps | Web UI signup |
| Data sovereignty | ✓ 100% your cluster | ✓ Local | ✓ Local | ✗ Cloud vendor |
| Multi-user isolation | ✓ Pod per user | ~ Container per user | ✓ Pod per user | ~ Shared infra |
| Built-in tools | 77 | Depends on image | Depends on image | ~20 (provider) |
| Agent Swarm | ✓ K8s pods | ✗ | ~ Manual | ✗ |
| Autoscaling (gateway) | ✓ HPA built-in | ✗ | ~ Manual HPA | ✓ Vendor handles |
| Channel support | 7 channels | Same (self-config) | Same (self-config) | 1–3 channels |
| Edge / Raspberry Pi | ✓ K3s <512MB | ✓ | ✗ Too heavy | ✗ |
| Open source | ✓ MIT | ✓ | ✓ | ✗ |
Three commands. Fully self-hosted. Complete sovereignty over every pod.