Open Source · MIT LicensedStar on GitHub

Your AI, your server,
every channel.

Self-hosted AI ops platform. Connect any LLM to Telegram, WhatsApp, and the web — with built-in tools, persistent memory, and a full admin dashboard.

View on GitHub →
TypeScript monorepo
Multi-user RBAC
MIT Licensed
Anthropic
OpenAI
Ollama
Batteries IncludedMulti-ChannelMIT LicensedTypeScript

Building AI agents shouldn't be this painful

You just want an AI assistant that works across channels, uses your tools, and remembers context. Why does that require 6 different libraries?

01

Juggling multiple LLM providers

Anthropic, OpenAI, Ollama — different APIs, different SDKs, different rate limits. Switching providers means rewriting code.

02

No unified chat interface

You want AI on Telegram, WhatsApp, and the web. That's 3 separate integrations, 3 sets of auth, 3 message formats.

03

Tools and memory are DIY

Web search, file access, shell commands, persistent memory — every agent project rebuilds these from scratch.

Everything you need. One docker compose up.

Multi-Channel

Telegram, WhatsApp, Web Chat, and REST API — all sharing one agent, one memory, one session per user.

Multi-Provider LLM

Anthropic Claude, OpenAI GPT, Ollama local models. Automatic failover chain with exponential backoff and cost-aware routing.

Built-in Tools

Web search, file ops, shell exec, browser automation, PDF reader, email, sub-agents, Docker sandbox — batteries included.

Persistent Memory

Auto-extracts facts from conversations. Session notes, user profiles, two-stage context compaction. Remembers across sessions.

Plugin System

MCP servers, CLI wrappers, HTTP APIs, OpenAPI specs. No code required — hot-reload from the full admin dashboard.

Multi-User RBAC

Admin, Power User, User roles. Per-user workspace isolation, tool permissions, tamper-evident audit trail.
Architecture

One platform, not six tools glued together

LLM gateway, agent framework, chat interface, messaging integration, memory system, and admin dashboard — in a single self-hosted deployment.

KairoClaw Platform Architecture — Multi-channel ingress, agentic core, tool ecosystem, memory, security, and operations

Tap to view full size

Up and running in 60 seconds

01. Clone & Run

One command to start. Docker pulls the image, sets up SQLite, and serves the admin dashboard. You're running in under 60 seconds.

02. Configure

Add your LLM provider keys, connect Telegram or WhatsApp, customize the agent persona. All from the built-in admin dashboard — no code required.

03. Talk to Your AI

Send a message on Telegram, WhatsApp, or the web. Your AI responds with tools, memory, and context. It remembers you across sessions.

01. Clone & Run

Built for real workflows

From personal assistant to team-wide AI gateway.

Personal AI Assistant

Your own AI that lives in Telegram or WhatsApp. It searches the web, reads files, runs shell commands, and remembers your preferences across conversations.

Web search & browsing
File read/write/edit
Persistent memory
Custom persona via IDENTITY.md
Channels: Telegram, WhatsApp, WebTools: All built-in tools
Security & Self-Hosting

Your data never leaves your server

Self-hosted means you control everything. No data sent to third parties. No vendor lock-in.

KairoClaw Multi-Layered Security Architecture

Tap to view full size

Security Layers

Encryption at Rest

AES-256-GCM envelope encryption for all API keys and tokens. Never stored in plaintext.

RBAC & Permissions

Admin, Power User, User roles. Per-tool permissions enforced at prompt and execution level.

Tamper-Evident Audit

Hash-chain audit trail on all tool calls and admin actions. Every decision is traceable.

Output Redaction

Auto-strips API keys, JWTs, database URLs from agent output. Priority-based span resolution.

Workspace Isolation

Per-user scoped file access, memory, and media. Shared team space + private user space.

Prompt Injection Detection

Heuristic engine identifies instruction overrides, role manipulation, and jailbreak attempts.

Recommended

Docker

One command. Includes Chromium, OCR, and all system tools.

docker compose up -d

Bare Metal

Node.js 22+, systemd service, runs on any Linux box.

./scripts/setup-local.sh

Open source. No catch.

Every feature is free. Bring your own LLM API keys and run it on your hardware.

Recommended

Open Source

Free

Everything you need, forever

  • All features included
  • Unlimited users
  • All channels (Telegram, WhatsApp, Web)
  • All built-in tools
  • Plugin system (MCP, CLI, HTTP, OpenAPI)
  • SQLite or PostgreSQL
  • MIT Licensed
Get Started

Enterprise

Contact Us

For teams that need more

  • Everything in Open Source
  • Priority support & SLA
  • Custom integrations
  • Deployment assistance
  • Training & onboarding
Contact Us

Frequently Asked Questions

KairoClaw is a self-hosted, open-source AI operations platform. It connects any LLM (Anthropic, OpenAI, Ollama) to messaging channels (Telegram, WhatsApp, Web) with dozens of built-in tools, persistent memory, and multi-user access. Think of it as n8n for AI agents — self-hosted and extensible.
Read the Docs on GitHub

Stop gluing tools together. Deploy KairoClaw.

One platform. Every channel. All your tools. Self-hosted and open source.

Dozens
Built-in tools
4
Chat channels
MIT
Licensed
View on GitHub