Perplexica evolved fast. The 2025 v1.11 overhaul brought a new wizard, live config updates, and dynamic model fetching. No more hardcoded limits.
You control everything: AI providers, search depth, embeddings, even custom headers. Run it local for zero cost or on a VPS for always-on access.
I once hit slow answers on a weak laptop. Tweaked to Groq + custom params—answers flew in seconds. Let’s dive in.
Why Advanced Configs Matter for Perplexica Users
Basic setup gives quick wins. Advanced ones fix real issues:
- Faster responses with Groq or optimized Ollama
- Deeper research via Quality mode + more pages
- Custom embeddings for better accuracy
- Private multi-device access without re-entering keys
- GPU/NPU boosts for local runs
In 2026, with AI privacy hot in USA/Canada, these tweaks keep your data yours.
Start with the New Setup Wizard (2026 Must-Do)
After Docker run (single command still king):
- Hit http://localhost:3000
- Wizard pops up: Pick provider first (Ollama local, Groq fast, OpenAI quality)
- Add keys live—no restarts needed
- Choose default model (dynamic fetch shows latest like Llama 3.1, Gemini 2.5, Claude Opus 4.1)
Pro move: Set Groq as primary for speed. Free tier handles tons of queries.
Need reliable hosting? Hostinger VPS shines for 24/7 Perplexica: https://hostinger.com?REFERRALCODE=TAMZID99. Cheap, fast, easy Docker support.
Perplexica Custom Models: Ollama, Groq, and More
Mix providers for best results.
Ollama Advanced Setup:
- Pull strong models: ollama pull llama3.1:70b or qwen2:72b
- In wizard: Set Ollama URL (http://host.docker.internal:11434 or your IP)
- For better quality: Use instruct variants (e.g., llama3.1-instruct)
- Custom params (temp 0.7 default; edit code in ollama.ts for top_k 40, num_ctx 8192 if you fork)
Groq for Blazing Speed:
- Get free API key at groq.com
- Add in wizard—models like Llama3-Groq-tool-use shine for function calling
- Set as default in settings for “Speed” mode
Other Providers:
- LM Studio or Transformers for local extras
- AIML API, Lemonade for NPU/GPU accel
- Custom OpenAI-compatible: Set base URL + key (great for vLLM or private endpoints)
Tip: Start with Groq + Ollama fallback. Switch modes mid-chat.
Perplexica Search Modes: Speed vs Quality Deep Dive
Three modes rule:
- Speed Mode — Quick, fewer sources, Groq shines
- Balanced Mode — Everyday sweet spot
- Quality Mode — Deep research, more pages from SearXNG
Advanced tweak: In settings, bump search depth (some users edit code to >7 pages; feature request open on GitHub). Use “Research” focus for academic/YouTube/Reddit.
Real win: Quality mode + Gemini 2.5 for complex queries. Cut research time in half.
Embeddings and Reranking: Boost Answer Accuracy
Perplexica reranks results with embeddings.
Advanced:
- Default: Built-in
- Custom: Set embedding provider (e.g., nomic-embed-text via Ollama)
- In settings: Pick model for similarity scoring
Users report jina-embeddings-v2-base-en gives sharper citations. Test in wizard.
Docker Advanced Configuration: Ports, Volumes, Env Vars
Single-command basic? Good. Advanced for power:
Use docker-compose.yml for control.
Example tweaks:
- Change port: -p 8080:3000
- Persistent data: -v perplexica-data:/home/perplexica/data
- Env vars: Add -e LOG_LEVEL=debug for troubleshooting
- Custom config: Mount config.toml (but 2026 wizard handles most live)
For remote access: Use reverse proxy (Nginx) + HTTPS. Expose only Perplexica port.
Host on VPS? Add –restart unless-stopped for uptime.
Custom Headers, System Prompts, and File Uploads
2026 extras:
- System instructions: In settings dialog, add custom prompts (e.g., “Always cite Canadian sources”)
- File uploads: PDFs/docs now work great—ask “Summarize this report”
- Custom headers: For special APIs, edit code (GitHub issue #835 has examples)
Weather widget? Toggle units (Imperial/Metric) in settings.
Troubleshooting Advanced Setups
- Stuck brainstorming? Stronger model or Groq fix most
- Slow local? GPU passthrough or Lemonade NPU
- Config not saving? Avoid read-only config.toml mounts
- Multi-device? Wizard keys persist per browser (request for server-side defaults open)
Check logs: docker logs perplexica
Perplexica: The Best Open Source Perplexity AI Clone You Can Run Today (2026 Guide)
My Daily Advanced Perplexica Workflow
I run Groq primary + Ollama fallback. Quality mode for work research. Upload docs often. On VPS via Hostinger—zero downtime.
Saved hours weekly. You can too.
Ready to level up? Tweak one thing today—like Groq integration. Feels like cheating.
What config are you trying next? Comment—I reply quick.