What it does

OpenKIWI is a self-hosted agentic automation system with a web UI for models, agents, workflow building, and scheduled heartbeats. It is not a machine-translation evaluation library, but a full open-source stack for running AI agents inside your own environment. It fits teams that want security-first containers, multi-model setup, Telegram or WhatsApp channels, and a transparent agent loop without depending on someone else's SaaS.

Plans

Free

Open source, self-hosted

Access

Open-source and self-hosted. The quickstart recommends bringing the stack up with Docker Compose, then configuring models, agents, messaging, and tools.

Screenshots

OpenKIWI workflow builder with a visual chain of steps — This is the clearest starting visual because it shows that OpenKIWI can assemble tool pipelines on a node-based canvas, not just run agent chat.

OpenKIWI agent configuration screen — The agent screen shows that the product is built around managed personas, scheduled tasks, and explicit configuration instead of a single generic bot.

OpenKIWI model selection and connection screen — This captures the multi-model part of the product well: local and remote models can be connected through one interface without breaking the workflow itself.

Usage examples

Autonomous heartbeats on a schedule

An agent does not have to wait for a manual prompt - it can run on a cron schedule and execute instructions from HEARTBEAT.md on its own. In practice, that is useful for digests, health checks, and recurring internal tasks.

Documentation

Agent connected to Telegram or WhatsApp

OpenKIWI lets you connect an agent to Telegram or WhatsApp and restrict access through allowlists. That matters when you want phone-based access without exposing the system to everyone.

Documentation

Application scenarios

Scenario: a team brings up a self-hosted stack with agents, models, and tools inside its own Docker containers instead of relying on an external cloud. OpenKIWI is valued here for its security-first approach and explicit allowlists.

Scenario: an agent monitors websites, prepares a digest, or pulls data from external APIs on a schedule and sends the result to Telegram or WhatsApp. At that point it stops being a chat toy and becomes a real background automation loop.