What Are Kantai Agents?
Kantai agents are long-running AI processes deployed as Kubernetes pods. Each agent has a defined role, a set of tools, persistent memory, and connections to your messaging channels. They’re not chatbots — they’re crew members.
Every agent runs on the OpenClaw runtime, which provides:
Workspaces
Each agent gets an isolated filesystem workspace. Files, notes, projects — persisted across sessions and restarts.
Memory
Short-term (conversation context) and long-term (curated knowledge) memory. Agents remember what matters and forget what doesn’t.
Tools
Agents call tools — calendar, email, web search, shell commands, APIs, code execution. Tools are configurable per agent and per permission tier.
Sencho 🎌 — The Captain
Your personal AI assistant and fleet coordinator. Sencho is the agent you interact with most — your daily driver for information, organization, and decision-making.
Capabilities
- Natural conversation across Telegram, Discord, Signal, and web
- Calendar management — scheduling, reminders, conflict detection
- Note-taking and knowledge capture
- Daily briefings — weather, calendar, priorities, unread messages
- Web research and summarization
- File management in your workspace
- Delegation to Takumi and Bannin for specialized tasks
Example Workflows
Morning briefing:
“Good morning — here’s your day: 3 meetings, 2 PRs need review, weather is 12°C and clear. Your 10am with the design team was moved to 2pm.”
Research task:
You: “Research the latest K8s gateway API changes” Sencho: Searches, summarizes, saves findings to your workspace, and offers to brief you later.
Calendar management:
You: “Schedule a 1:1 with Sarah next week, 30 min, avoid mornings” Sencho: Checks availability, proposes slots, sends the invite on confirmation.
Use Cases
- Personal productivity and organization
- Information gathering and synthesis
- Team coordination and scheduling
- Daily standup summaries
- Decision support with context from your history
Takumi ⚒️ — The Craftsman
Your execution and automation agent. Takumi handles the technical heavy lifting — CI/CD, code review, infrastructure operations, and automated workflows.
Capabilities
- Code review with context-aware feedback
- CI/CD pipeline monitoring and triage
- Infrastructure automation (Terraform, Helm, kubectl)
- Shell command execution in sandboxed environments
- Git operations — branching, PRs, conflict resolution
- Log analysis and incident response
- Automated testing and quality checks
Example Workflows
CI failure triage:
Pipeline fails → Takumi analyzes logs → identifies the breaking change → opens an issue with fix suggestion → notifies you via Sencho.
Code review:
PR opened → Takumi reviews diff → checks for security issues, style violations, test coverage → posts inline comments → approves or requests changes.
Infrastructure task:
You: “Scale the staging cluster to 5 nodes” Takumi: Validates the request, runs the operation, confirms completion with metrics.
Use Cases
- Automated code review on every PR
- CI/CD pipeline management and failure analysis
- Infrastructure scaling and maintenance
- Security scanning and dependency updates
- Runbook automation for incident response
Bannin 🏯 — The Guardian
Your public-facing and security agent. Bannin handles external communications, customer support, security monitoring, and community management.
Capabilities
- Customer support across chat, email, and ticketing systems
- Security event monitoring and alerting
- Community moderation (Discord, forums)
- Public communications drafting
- Threat detection and incident escalation
- Knowledge base management for external docs
- Social media monitoring
Example Workflows
Customer support:
User asks a question in Discord → Bannin searches the knowledge base → provides an accurate answer with links → escalates to a human if uncertain.
Security monitoring:
Bannin detects unusual API traffic patterns → correlates with known threat signatures → creates an alert → notifies the team with recommended actions.
Public comms:
You: “Draft a changelog post for v1.2.0” Bannin: Pulls recent commits, generates a user-friendly changelog, submits for review.
Use Cases
- 24/7 customer support with knowledge base backing
- Security monitoring and automated incident response
- Community management and moderation
- Status page updates during incidents
- External documentation maintenance
Inter-Agent Communication
Agents don’t work in isolation — they communicate through a structured messaging system.
How It Works
- Agents send messages to each other via internal channels
- Sencho can delegate tasks to Takumi or Bannin
- Agents can report back with results, status updates, or escalations
- All inter-agent messages are logged and auditable
Example Flow
- You tell Sencho: “Deploy the latest build to staging”
- Sencho validates the request and delegates to Takumi
- Takumi runs the deployment, monitors rollout
- Takumi reports back: “Deployed v1.2.0 to staging, all health checks passing”
- Sencho confirms to you with a summary
Trust Model & Permission Tiers
Every agent operates within a defined permission boundary.
Tier 1 — Read Only
Browse files, search, read logs. No modifications. Safe for monitoring roles.
Tier 2 — Standard
Read + write to workspace, call approved tools, send messages. Default for most agents.
Tier 3 — Elevated
Full tool access including shell execution, infrastructure commands, and external API calls. Requires explicit approval in Governance.
Permissions are configured per-agent in the Governance tab and enforced by the OpenClaw runtime. Agents cannot escalate their own permissions.
Ready to deploy your fleet? Start with the Getting Started guide.