Self-Host Perplexica: Open Source Perplexity 2026

Q: When Perplexica Isn't the Right Choice?

Be honest about the tradeoffs: Don't use Perplexica if: You want a polished, no-maintenance AI assistant — Perplexity AI is better You're on a low-power VPS without enough RAM — SearXNG alone is lighter You need mobile-first access — Perplexica's web interface isn't optimized for phones You need real-time data with sub-second responses — cloud APIs will always be faster for latency Use Perplexica if: You're researching sensitive topics and don't want your queries logged You already run Ollama an

Perplexity AI is genuinely good — but every query you run trains their models, builds their data profile, and eventually ends up funding a closed-source moat. Perplexica is the answer: a 33,000-star MIT-licensed AI search engine that runs entirely on your hardware, uses your choice of LLM, and routes web searches through SearXNG instead of a vendor-controlled black box.

This guide gets you from zero to a working Perplexica instance in under 15 minutes.

TL;DR

Perplexica is the best open-source Perplexity alternative in 2026. At 33k GitHub stars, MIT licensed, and actively maintained through v1.12.1, it's mature enough for daily use. Run it with Ollama for 100% local operation, or wire it to OpenAI, Anthropic, or Groq if you want faster inference. Every search stays on your hardware — no query history, no ad targeting, no training data contribution.

Key Takeaways

33,000 GitHub stars — the most popular self-hosted AI search engine by a wide margin
MIT licensed — fork it, modify it, run it commercially without restriction
Model-agnostic — Ollama (local), OpenAI, Anthropic Claude, or Groq all work out of the box
Powered by SearXNG — aggregates Google, Bing, DuckDuckGo, and 70+ engines without tracking
3 focus modes — Speed (fast), Balanced (daily use), Quality (deep research) tunable per query
v1.12.1 released December 2025 — actively maintained, 976 commits on master
Docker Compose install — five commands from clone to running instance

Why Perplexica Over Perplexity AI

Perplexity AI charges $20/month for Pro. The free tier limits your daily searches and uses their proprietary model stack. More importantly, your search history is stored, analyzed, and potentially used for model training. Perplexity's privacy policy is standard enterprise: "we may use your content to improve our services."

Perplexica flips this completely. The architecture is:

Your query → SearXNG (aggregates search engines) → LLM of your choice → Cited answer

Nothing leaves your network unless you configure it to use a cloud API. With Ollama, the entire stack — search aggregation, AI inference, answer generation — runs on your machine.

The tradeoff is real: Perplexica requires setup, hardware, and occasional maintenance. But for privacy-conscious users, researchers handling sensitive topics, or developers who want full control over their AI stack, this is the better choice.

What You're Actually Installing

Perplexica isn't a single binary — it's a small stack of services:

Service	Role	Port
Perplexica frontend	Chat interface (Next.js)	3000
Perplexica backend	Search orchestration + LLM calls	3001
SearXNG	Privacy-preserving meta-search engine	4000

The backend handles the logic: it takes your question, constructs search queries, sends them to SearXNG, fetches the top results, and passes the context to your LLM with a prompt asking it to synthesize a cited answer.

Docker Compose spins all three up together. You don't need to configure SearXNG separately — Perplexica's docker-compose.yaml handles it.

System Requirements

Before installing, confirm you have:

Docker + Docker Compose — v24+ recommended
RAM: Minimum 4GB free. 8GB+ if running Ollama locally alongside Perplexica
Disk: ~2-5GB for Docker images, plus model storage if using Ollama
LLM backend (pick one):
- Ollama — free, fully local, needs GPU for acceptable speed (CPU works but is slow)
- OpenAI API key — fastest results, ~$0.01 per search at gpt-4o-mini
- Anthropic API key — Claude Haiku is a good balance of speed and quality
- Groq API key — free tier available, fastest inference for Llama-based models

Installation: Docker Compose (Recommended)

Step 1: Clone the repository

git clone https://github.com/ItzCrazyKns/Perplexica.git
cd Perplexica

Step 2: Configure

Copy the sample config and open it for editing:

cp sample.config.toml config.toml

Open config.toml. The key sections:

[GENERAL]
PORT = 3001
SIMILARITY_MEASURE = "cosine"  # or "dot_product"

[API_KEYS]
OPENAI = ""           # paste your key here if using OpenAI
GROQ = ""             # paste if using Groq
ANTHROPIC = ""        # paste if using Anthropic

[API_ENDPOINTS]
OLLAMA = ""           # e.g. "http://host.docker.internal:11434" for local Ollama
SEARXNG = "http://searxng:4000"  # leave as-is for Docker setup

If using Ollama locally on Linux: You must expose Ollama to the Docker network. Set the environment variable before starting Ollama:

OLLAMA_HOST=0.0.0.0:11434 ollama serve

Then set the endpoint in config.toml:

OLLAMA = "http://host.docker.internal:11434"

Step 3: Start

docker-compose up -d

Wait 30-60 seconds for all containers to initialize. The first run pulls images (~1.5GB total).

Step 4: Open and configure the model

Navigate to http://localhost:3000. On first load, you'll see the Settings gear icon — click it to select your LLM:

Chat model: The model that generates answers. Groq's llama-3.3-70b-versatile is a great free option for cloud. For local, llama3.2:3b works on 8GB RAM, llama3.1:8b is better on 16GB+.
Embedding model: Used for document reranking. Leave at default unless you're customizing.

Step 5: Run your first search

Type a question. Try something where citations matter:

"What changed in Node.js 24 vs Node.js 22?"

You'll see Perplexica query SearXNG, pull sources, and generate a cited answer with numbered footnotes linking to exact URLs. This is the core loop.

Search Modes Explained

Perplexica's three focus modes change how it approaches each query:

Mode	Speed	Depth	Best For
Speed	~3-5s	Shallow	Quick factual lookups, recent events
Balanced	~8-15s	Medium	Everyday research, most queries
Quality	~20-40s	Deep	Technical comparisons, research papers

Switch modes per query in the interface. Quality mode runs multiple search rounds and uses more context tokens — it costs more on paid APIs but produces noticeably better synthesis on complex topics.

Beyond the base modes, you can narrow the search source:

All Sources — default, aggregates everything
Academic — prioritizes Google Scholar, arXiv, Semantic Scholar
Reddit/Discussions — community sentiment, real user experiences
News — time-sensitive results prioritized
YouTube — video content surfaced with summaries

Using File Upload for Local Analysis

Perplexica supports uploading PDFs and text files to query against them locally. Drop a research paper into the chat and ask questions directly — the backend embeds the document and runs similarity search against your query before calling the LLM.

This is particularly useful for:

Analyzing internal documents without sending them to any cloud service
Summarizing long PDFs with source citations pointing to specific pages
Cross-referencing a document against web results in a single query

Perplexica vs the Alternatives

Tool	Stars	LLM	Privacy	Setup Complexity
Perplexica	33,000	Any (local or cloud API)	✅ Full local option	Medium
SearXNG alone	16,000	None — raw search	✅ Full local	Low
Morphic	8,000	OpenAI/Vercel AI only	❌ Cloud required	Medium
MindSearch	14,000	Multiple	Partial	High
Perplexity AI	n/a	Proprietary	❌ Vendor-controlled	None (SaaS)

SearXNG is what powers Perplexica's search layer. If you only want privacy-respecting web search without AI synthesis, SearXNG alone is simpler and lighter. But Perplexica is the right choice when you want the full Perplexity-style workflow — question → web research → cited answer.

Morphic is the most polished open-source competitor, but it's designed to run on Vercel and requires OpenAI. It's not truly self-hostable in the local sense — it's more "deploy your own cloud instance." Perplexica runs on your Raspberry Pi if you want it to.

Keeping Perplexica Updated

Perplexica releases frequently — v1.12.1 landed December 2025. To update:

cd Perplexica
docker-compose down
git pull origin master
docker-compose pull
docker-compose up -d

Check the GitHub releases page for changelog. Most updates add model support or fix SearXNG integration bugs — watch for breaking config.toml changes in the release notes.

When Perplexica Isn't the Right Choice

Be honest about the tradeoffs:

Don't use Perplexica if:

You want a polished, no-maintenance AI assistant — Perplexity AI is better
You're on a low-power VPS without enough RAM — SearXNG alone is lighter
You need mobile-first access — Perplexica's web interface isn't optimized for phones
You need real-time data with sub-second responses — cloud APIs will always be faster for latency

Use Perplexica if:

You're researching sensitive topics and don't want your queries logged
You already run Ollama and want to add web context to your local LLM
You want to customize prompts, add focus modes, or build on top of the search pipeline
You're building an internal research tool for a team or organization

Methodology

GitHub data: ItzCrazyKns/Perplexica, March 2026
Installation verified against v1.12.1 docker-compose.yaml and official README
Comparison data from openalternative.co and direct GitHub repositories
Search mode descriptions from Perplexica DeepWiki

Looking for other self-hosted AI tools? See our guides on self-hosting Dify, Best Open Source Cursor Alternatives, and Self-Hosted AI Agent Frameworks. Compare all options on the OSSAlt homepage.

Comments