The basic loop is a well-trodden side project: send the question to a search API, pull the text of the top pages, and have a model answer with numbered citations. An agent builds that, with follow-up questions and a searchable history, in one sitting. What it can't build is what sits behind Perplexity's answers: its own index and ranking, licensed sources such as Statista and PitchBook, the Computer agent, the Comet browser and the apps. The bill doesn't disappear either: you pay per search and per token, and on a large model heavy use can approach the price of Pro.
Before you buildPerplexity's free plan still gives practically unlimited basic searches and three Pro searches a day; check whether that covers you before building.
Build me a personal answer engine to replace Perplexity Pro. Requirements: - Next.js 15 (App Router) + TypeScript on http://localhost:4800: one question box; the answer streams in with numbered citations [1], [2] that link to a source list under it. - Search: Brave Search API (BRAVE_API_KEY in .env) for the top 8 results, or a local SearXNG instance when SEARXNG_URL is set. - Fetch each result with a 6-second timeout, extract readable text with @mozilla/readability + jsdom, keep the 1,500 words most relevant to the question, and skip pages that fail. - Answer with the Vercel AI SDK (`ai` package) using ANTHROPIC_API_KEY or OPENAI_API_KEY, or Ollama when OLLAMA_MODEL is set, with a model picker per thread. The system prompt requires a citation on every factual sentence and a plain "I couldn't find this" when the sources don't support an answer. - Follow-ups stay in the thread (earlier Q&A plus fresh search results); show 3 suggested follow-up questions under each answer. - Research mode: the model writes 3-5 sub-queries, searches each, and writes a sectioned report with citations; at most 25 pages fetched per report. - History: threads in better-sqlite3 at ./data/answers.db, a /library page with SQLite FTS5 search, and export of any thread as Markdown. - No accounts, no telemetry, local-first; only search and model calls leave the machine. - Out of scope: image or video generation, a browser, mobile apps, file connectors and paywalled data sources. - README: getting a Brave Search key, running SearXNG in Docker as the free option, and the rough cost per question for each model.
Opens your agent with the prompt filled in, you press enter. Or copy it and paste it into an empty folder.
The first prompt builds the core. Run these one at a time in the same folder to grow it into your own Perplexity Pro.
Add document search to the answer engine in this folder. Requirements: - A /files page to upload PDF, DOCX, Markdown and TXT; extract text with pdf-parse and mammoth and keep the originals in ./data/files/. - Split into chunks of about 800 tokens, embed them with nomic-embed-text through Ollama, and store the vectors in SQLite with sqlite-vec. - A switch per question: Web, My files or Both. File citations show the file name and page and open the file. - Out of scope: Google Drive, Dropbox and other connectors. - Update the README with the Ollama model pull command.
Make the answer engine in this folder usable as my browser's search engine. Requirements: - GET /search?q=... opens a new thread with that question. - Serve an OpenSearch description at /opensearch.xml and link it from the page head so Chrome and Firefox can add it. - A query starting with "!g " goes straight to Google instead. - Update the README with the steps for Chrome and Firefox.
Add saved questions to the answer engine in this folder. Requirements: - Save any question as a watch that runs daily or weekly, scheduled with node-cron. - Each run answers it fresh and stores the result next to the earlier ones. - Show what changed since the last run: new sources and a one-paragraph summary of the difference written by the model. - A /watches page lists them with the date of the last change. - Out of scope: email and push notifications. - Update the README.
Add an Academic mode to the answer engine in this folder. Requirements: - Query the arXiv API and the Semantic Scholar Graph API (optional key in .env) instead of the web search API. - Use abstracts, plus full text for open-access PDFs under 40 pages. - Cite papers as authors, year and title, with a DOI or arXiv link. - Back off and retry when an API returns a rate-limit error. - Update the README.
For $20 a month people get quick, cited answers and several frontier models without juggling API keys or usage bills. Analysts and students lean on Research mode and the premium data sources, and many simply like having the same history on the phone, in the browser and on the desktop.
Estimate · Brave's monthly credit covers about 1,000 searches. Each answer feeds the model around 10,000 tokens of page text: well under a cent on a small model, a few cents on a large one. Thirty questions a day on a large model costs about as much as Pro.
Checked 2026-10-13 against Perplexity Pro's docs: 1, 2, 3, 4
| Plan | Monthly | Annual, per month | What you get |
|---|---|---|---|
| Free | $0 | $0 | Basic searches with no practical cap, 3 Pro searches a day, 1 Research run a month, limited file uploads. |
| Pro | $20 | $17 | Higher weekly limits on Pro Search, Research, uploads and file and app creation; a choice of frontier models; a weekly Computer allowance and a one-time 4,000 bonus credits. |
| Max | $200 | $167 | The highest limits and newest models, 10,000 Computer credits a month, the Max assistant in Comet, priority support. |
| Education Pro | $10 | — | Pro features plus Learn Mode for students and educators verified through SheerID. |
| Enterprise Pro | $40 | $33.33 | 400 Pro searches a week, 50 Research runs a month, shared Projects, file connectors, seat management; data not used for training. |
The free plan gives practically unlimited basic searches, 3 Pro searches a day and 1 Research run a month.
Kinda. The core is a weekend build with the prompt on this page, but real gaps remain: Perplexity's own search index and answer ranking, Licensed premium sources such as Statista, PitchBook and CB Insights.
About $20 a month on the Pro plan, $240 a year (checked 2026-10-13).
Perplexity's own search index and answer ranking; Licensed premium sources such as Statista, PitchBook and CB Insights; The Computer agent, the Comet browser and the phone apps; Image and video generation; Several frontier models under one flat subscription. If any of those matter to you, keep paying.
Yes: Vane (open source), Morphic (open source), GPT Researcher (open source).