Deploy and Host Khoj AI on Railway

Khoj AI is an open-source personal AI that allows you to chat with your knowledge base, search the web, execute code in sandboxes, and integrate with multiple AI models. It’s designed to be easily extensible, self-hostable, and privacy-conscious.

About Hosting Khoj AI

Hosting Khoj AI on Railway provides a simple and powerful way to deploy your own personal AI assistant in the cloud. With minimal setup, Railway takes care of the infrastructure and scaling, while you configure the services and API keys. Deploying Khoj involves setting up a Python backend, connecting to a PostgreSQL (pgvector) database, and optionally integrating external search and LLM APIs. Once deployed, you can access Khoj through your Railway public domain and manage everything from a single dashboard.

Common Use Cases

Personal AI assistant for research and writing
Self-hosted AI chatbot with search and retrieval capabilities
Integration with custom LLMs and vector databases for advanced workflows

Dependencies for Khoj AI Hosting

PostgreSQL with pgvector extension (auto-provisioned on Railway)
Python runtime for backend server
Optional external API integrations (Jina, OpenAI, Gemini, etc.)

Deployment Dependencies

Implementation Details (Optional)

You can scale the Gunicorn workers depending on your Railway plan by changing GUNICORN_WORKERS. Default is 6 for Pro plans.

GUNICORN_WORKERS=3

You can also change the OpenAI-compatible API endpoint by setting OPENAI_BASE_URL.

Environment Variables Explained

Server Service

KHOJ_DEBUG — Enables debug mode. Default: False.
E2B_API_KEY — Key for E2B sandbox. Enables code execution in E2B instead of Terrarium.
KHOJ_DOMAIN — Your Railway public domain.
POSTGRES_DB / HOST / PORT / USER / PASSWORD — PostgreSQL database credentials (auto-provisioned).
JINA_API_KEY — API key for Jina AI (optional, for web search).
KHOJ_NO_HTTPS — Set True or False to control HTTPS usage. Be cautious when enabling this.
GEMINI_API_KEY / OPENAI_API_KEY / ANTHROPIC_API_KEY — API keys for respective LLM providers.
OLOSTEP_API_KEY / FIRECRAWL_API_KEY / SERPER_DEV_API_KEY — Optional API keys for webpage reading and search.
OPENAI_BASE_URL — Custom URL for OpenAI-compatible providers (e.g., DeepInfra, LMStudio).
GOOGLE_CLIENT_ID / GOOGLE_CLIENT_SECRET — Google OAuth integration.
GUNICORN_WORKERS — Number of Gunicorn workers for scaling.
KHOJ_ADMIN_EMAIL / KHOJ_ADMIN_PASSWORD — Admin credentials for Khoj panel.
KHOJ_SEARXNG_URL — Default URL of SearxNG service.
KHOJ_TERRARIUM_URL — URL of Terrarium sandbox for code execution.
KHOJ_ALLOWED_DOMAIN — Allowed domain for Khoj.
KHOJ_OPERATOR_ENABLED — Enable operator mode (Khoj can use its computer).
KHOJ_DJANGO_SECRET_KEY — Secret key for the Django server.
KHOJ_TELEMETRY_DISABLE — Disable telemetry if desired.
KHOJ_DEFAULT_CHAT_MODEL — Set a default chat model if needed.

SearxNG Service

SEARXNG_BASE_URL — Base URL for the SearxNG search service.

Why Deploy Khoj AI on Railway?

Railway is a singular platform to deploy your infrastructure stack. Railway will host your infrastructure so you don't have to deal with configuration, while allowing you to vertically and horizontally scale it.

By deploying Khoj AI on Railway, you are one step closer to supporting a complete full-stack application with minimal burden. Host your servers, databases, AI agents, and more on Railway.

Services Overview

To deploy Khoj AI on Railway, multiple services work together to provide a complete and modular AI assistant stack. Here’s a breakdown of each service:

1. `pgvector` — Vector Database

Image: pgvector/pgvector:pg16
Purpose: This is a PostgreSQL database with the pgvector extension enabled. Khoj uses it to store and retrieve embeddings for efficient semantic search.
Notes: Automatically provisioned by Railway. Make sure the connection variables are properly set in the server service.

2. `server` — Khoj Backend

Image: ghcr.io/khoj-ai/khoj-cloud:latest
Purpose: This is the main Khoj AI backend service that handles all API calls, search orchestration, and model integrations.
Notes: Connects to pgvector and optionally uses external APIs like OpenAI, Gemini, Jina, etc.

3. `sandbox` — Terrarium Code Sandbox

Image: ghcr.io/khoj-ai/terrarium
Purpose: A secure, isolated sandbox where Khoj can execute Python code. This is used when the agent needs to run code directly.
Notes: Alternatively, you can use an external sandbox with E2B.

4. `searxng` — Web Search Engine

Image: searxng/searxng
Docker Hub: https://hub.docker.com/r/searxng/searxng
Purpose: Provides fast, privacy-friendly meta web search. Khoj uses this as the default web search engine when enabled.
Notes: The service URL is referenced in the KHOJ_SEARXNG_URL variable.

5. `computer` — Operator Mode Service

Image: ghcr.io/khoj-ai/khoj-computer
Purpose: Enables “operator mode” where Khoj can interact with the system and execute more advanced workflows.
Notes: Optional but powerful feature for power users. Controlled via the KHOJ_OPERATOR_ENABLED environment variable.