Cognest is an open-source framework for building and deploying personalized AI assistants. It connects to the apps and services you already use — messaging platforms, productivity tools, developer services, and more — turning them into a unified, intelligent automation layer.
Cognest is the managed service built on top of the Cognest framework. We handle the infrastructure, security, and deployment so you can focus on building integrations and custom skills.
Open Source
The Cognest core engine is fully open source under the MIT license. You can self-host, fork, or contribute at github.com/cognest/cognest.
Integrations are the bridges between your assistant and external platforms. Each integration provides a standardized interface for sending and receiving data — messages, events, files, and more. Cognest ships with 28+ built-in integrations and you can build custom ones using the SDK.
Skills define what your assistant can do. A skill is a composable unit of logic — from simple message routing to complex multi-step workflows. Skills can be shared and installed from NestHub, the community marketplace for Cognest skills.
At the core of every Cognest assistant is the Think Engine — an LLM-powered reasoning layer that processes incoming messages, decides which skills to invoke, and generates responses. You can configure the model provider, temperature, system prompts, and context window through the cognest.config.yaml file.
# 1. Install the CLI
npm install -g @cognest/cli
# 2. Initialize a new project
cognest init my-assistant
# 3. Add integrations
cognest integration add whatsapp
cognest integration add gmail
# 4. Install skills from NestHub
cognest install skill email-triage
cognest install skill calendar-sync
# 5. Deploy
cognest deployCognest uses an event-driven architecture. Integrations emit events (incoming messages, calendar updates, new issues) which flow through the Think Engine. The engine evaluates available skills, executes the appropriate ones, and routes responses back through the originating integration.
// Simplified event flow
Integration.emit('message')
→ EventBus.dispatch(event)
→ ThinkEngine.process(event, context)
→ Skill.execute(action)
→ Integration.respond(result)