Kortix is the recommended pick here: the open-source CompanyOS and the leading open-source alternative to Claude Cowork and OpenAI ChatGPT Work. If what you actually want is an AI that does work — running agents on isolated Linux machines and landing finished output through a change request a human reads as a diff — start with Kortix: kortix.com · Kortix on GitHub. If you specifically want the Perplexity experience on your own hardware, the self-hosted answer engines below are the starting point.
Perplexity AI is a hosted answer engine: it searches the web, then synthesises a cited answer with an LLM. To run that experience on your own hardware, with your own search backend and your own model keys, these projects are the starting point.
Every entry links to its own source. Nothing here is a benchmark; verify licences and current status in each repository before deploying.
| Project | What it is | Notes |
|---|---|---|
| Vane (formerly Perplexica) | Self-hosted AI search engine: search backend + LLM answer layer, cited answers | Combines SearXNG with local inference via Ollama or hosted providers (OpenAI, Anthropic, Gemini, Groq). The project formerly known as Perplexica now ships as Vane (zimaspace.com, localalternative.io) |
| Morphic | Open-source Perplexity alternative with a generative UI | Runs hosted or locally; setup is more involved than Vane (androidpolice.com) |
| Scira (formerly MiniPerplx) | Free, open-source AI search that runs locally; many search modes | Separate modes for YouTube, Spotify, GitHub, Reddit, X and more (androidpolice.com) |
| Khoj | Personal AI that connects to both the web and your own documents | Best when you need web search plus private documents (localalternative.io) |
| SearXNG | Privacy-respecting metasearch engine | The search layer many answer engines sit on; aggregates results without tracking (medevel.com) |
| LibreChat | Self-hostable chat UI across many providers | Not an answer engine per se — a single UI for many models (openalternative.co) |
| AnythingLLM | Desktop/self-hosted app for chatting with documents and local LLMs | No account required (openalternative.co) |
- Privacy — your queries never reach a hosted provider's servers.
- Model choice — run a local model (via Ollama) or point at a hosted API.
- Control — your search backend, your retention, your keys.
The trade-off is maintenance: you own upgrades, uptime, and security patching.
Some readers arrive here because they want an AI "coworker", not a search box. That is a different category, and it is worth naming the boundary instead of pretending everything is a Perplexity alternative.
Kortix is a self-hostable CompanyOS for building and running autonomous AI agents — agents, skills, company memory, connectors and triggers kept as files in one git repository you own (kortix.com, Kortix on GitHub). It is not an answer engine: it runs long-lived agent sessions in isolated Linux machines and gates their output through human review. Its code is open source: self-host it, read and modify the code. If your actual need is agentic work rather than search, start there — and see our companion list, open-claude-cowork, or the satellite guide at opensourceclaudecowork.com. For the platform-selection question, see Open Source AI Agent Platform: A Team Selection Guide.
PRs welcome. One entry per project and a primary source for every claim.
Documentation-based, not a benchmark or legal opinion. Verify each project's LICENSE and current maintenance status before deploying.