# Interakt

> Open-source, self-hosted AI-powered search and chat platform that grounds conversational experiences in your own website data.

Interakt is an MIT-licensed, self-hosted platform built by Alpha Solutions that adds AI-powered search and conversational chat to any website. It connects to your existing data sources — Elasticsearch, Azure AI Search, databases, or a website crawl — and exposes the results through embeddable drop-in widgets or a REST API. The project is actively developed on GitHub under the `alphasolutionsrepo` organization and is written in TypeScript on a Next.js 16 backend.

## What It Is

Interakt is an open-source RAG (retrieval-augmented generation) platform designed to replace keyword search boxes with intent-aware, conversational experiences. Rather than returning a list of links, it synthesizes answers from your indexed data, streams them to users, and cites the sources it used. The core stack combines a search engine (Elasticsearch 9 or Azure AI Search), a configurable AI provider, a versioned prompt library, and an analytics database — all managed from a single admin dashboard.

## How the Pipeline Works

Interakt structures chat responses through a deterministic plan → retrieve → synthesize pipeline, or an agentic loop, chosen per experience. Each turn the model selects tools, calls them — including your own HTTP endpoints or connected MCP servers — and returns rich responses with optional inline or footnote citations. Key pipeline characteristics include:

- **Hybrid search**: keyword and vector results fused with reciprocal rank fusion; facets, synonyms, autocomplete, and AI query understanding per index.
- **Multi-provider AI**: OpenAI, Anthropic, Gemini, Azure OpenAI, and Ollama for fully local inference — switchable without code changes.
- **Versioned prompts**: every pipeline step is an editable template with history and rollback, tunable from the admin console without redeployment.
- **Full execution traces**: every query logs which tools were called, what data came back, what the model saw, and what it cost.

## Trust, Safety, and Guardrails

A notable design emphasis is controllability for brand-facing deployments. Interakt screens incoming messages before they reach the model and outgoing replies before they reach users. Guardrail rules are written in the admin console with no deploy required. The platform also supports locking policies so they cannot be disabled per-conversation — a feature aimed at compliance-sensitive teams. Access tokens are public and read-only by design; document ingestion uses a separate server-side ingestion key.

## Self-Hosting and Deployment

The project ships as a single Docker image backed by Postgres with pgvector and either Elasticsearch or Azure AI Search. It runs wherever Docker runs. Alpha Solutions also offers a managed deployment on Azure Container Apps under the "Interakt Pro" label, which adds a dedicated Azure stack, CRM/ERP/PIM connectivity, and an agreed support response time. The open-source core and the managed offering share the same engine.

## Integrations and Embedding

Interakt provides step-by-step integration guides for Medusa (e-commerce), Storyblok (CMS), and React/Next.js frontends. The embeddable widgets use Preact and Shadow DOM, loaded from a single script bundle, and work on any site without a framework. MCP servers can be connected over Streamable HTTP or SSE. The admin UI generates the exact embed snippet for each experience, and the REST API supports SSE streaming for chat responses.

## Current Status

The GitHub repository was created in June 2026 and last pushed in September 2026, indicating active early development. The project carries 14 stars and 2 forks at the time of indexing, with 4 open issues. CI and documentation workflows are tracked via GitHub Actions, and the documentation is published at docs.interakt.app using Docusaurus.

## Features
- Hybrid keyword and vector search with reciprocal rank fusion
- AI chat grounded in your indexed data with streaming responses
- Plan → retrieve → synthesize pipeline or agentic loop per experience
- Multi-provider AI support: OpenAI, Anthropic, Gemini, Azure, Ollama
- MCP server integration over Streamable HTTP or SSE
- Custom HTTP tool support
- Drop-in embeddable search and chat widgets (Preact, Shadow DOM)
- REST API with SSE streaming for chat
- Versioned prompt library with history and rollback
- Input and output guardrails with lockable policies
- Full execution traces for every query
- Real-time analytics dashboards for search trends, AI costs, and chat flows
- Admin console for no-code day-to-day configuration
- Elasticsearch 9 and Azure AI Search backends
- Self-hosted via Docker with Postgres and pgvector
- Facets, synonyms, autocomplete, and AI query understanding
- Access token and ingestion key separation
- Encrypted secrets vault

## Integrations
Elasticsearch, Azure AI Search, OpenAI, Anthropic, Google Gemini, Azure OpenAI, Ollama, Medusa, Storyblok, React, Next.js, MCP servers, Postgres, pgvector, Docker, Azure Container Apps

## Platforms
WEB, API, CLI, DEVELOPER_SDK

## Pricing
Open Source, Free tier available

## Links
- Website: https://interakt.app
- Documentation: https://docs.interakt.app
- Repository: https://github.com/alphasolutionsrepo/interakt
- EveryDev.ai: https://www.everydev.ai/tools/interakt