Open source alternatives to Perplexity

What Perplexity does

Perplexity takes a question, searches the web in real time and writes a conversational answer with numbered citations that link to its sources. Pro Search produces more thorough answers, and Research mode runs many searches in sequence to produce a longer report. The answers are written by language models that include third-party models from providers such as OpenAI and Anthropic, and paid subscribers can choose the model. Plans range from the free Standard plan through Pro and Max to Education and Enterprise plans, and there is an API.

Why people look for an alternative

• Training on by default for individual plans. For Free, Pro and Max users, the AI Data Retention setting is on by default, which means your data is collected for AI training. You can switch it off under Account settings > Preferences, but the opt-out only covers data collected after that point; data already collected for training cannot be removed.
• Enterprise is handled differently. Perplexity states that Enterprise data is never used to train its models, keeps Enterprise file uploads for seven days and lets admins configure retention. It also says its agreements with third-party model providers forbid them from training on Perplexity data.
• Retention and location. The privacy notice keeps personal data for as long as needed for its stated purposes and says data may be processed in the United States or other countries whose privacy laws may be less strict than yours. After account deletion, personal information is removed within 30 days.
• Free usage limits. The free Standard plan includes basic searches, but only three Pro Searches a day and one Research query a month.
• No self-hosted version. Perplexity is available only as a hosted service, apps and API.

A self-hosted answer engine keeps your queries, history and settings on your own server and lets you choose the search source, often a SearXNG instance. Two points still apply: the search queries reach whatever search engines the backend queries, and the language model sees your question and the retrieved pages. With a hosted API, that content goes to the provider; with open-weight models run locally it stays on your hardware, which needs a suitable GPU or plenty of memory, and local models are generally less capable than the top hosted ones.

What to compare in an alternative

1. Search backend. Which search engines or APIs it uses, and whether you can run that part yourself.
2. Citations. Whether answers link each claim to sources you can open.
3. Depth. A quick-answer mode versus a multi-step research mode that reads more pages.
4. Models. Support for local runtimes such as Ollama and for hosted providers.
5. Own documents. Whether it can search your files as well as the web.
6. Setup. The services you need to run (search engine, database, model server) and whether it supports several users.

Which one to choose

Vane, formerly Perplexica, is the most direct open source counterpart: a self-hosted answer engine that searches through SearXNG, cites its sources, has a quick mode and a longer research mode, and works with local models through Ollama or with hosted providers. It is single-user, with no login. Morphic does the same job with richer answer layouts, a quick and an adaptive research mode, built-in user accounts with OIDC, and an unusually detailed account of what a self-hosted instance sends outside your server; its Docker setup also bundles SearXNG. Khoj combines web search with search over your own notes and documents and has a research mode for longer, multi-step answers, which suits people who want one assistant for both.