🔮 KodaAI

A privacy-first, Perplexity-style AI search & chat app — powered by Ollama Cloud. No OpenAI, no Anthropic, no telemetry. Search-augmented, cited answers with a polished dark UI.

This build uses Ollama Cloud (https://ollama.com), not a local Ollama daemon. You bring your own Ollama Cloud API key; inference runs on your Ollama account.

---

Features

• Search-augmented chat (agentic RAG) — SearXNG web search → scrape → grounded, cited answer
• Token-by-token streaming with an animated cursor
• Inline citations [1] [2] linked to source cards
• Auto follow-up questions after every answer
• Thread history persisted in localStorage
• Model switcher — lists your Ollama Cloud models
• Focus modes — All · No Search · Code · Academic
• Privacy dashboard — shows the active model and a 0 third-party-AI-calls counter
• Mobile responsive, keyboard-navigable, accessible

---

Quick start

1. Get an Ollama Cloud API key

Sign in at https://ollama.com, then create a key under Settings → Keys. Cloud models include gpt-oss:120b, gpt-oss:20b, qwen3-coder:480b, deepseek-v3.1:671b, and more.

2. Configure environment

cp .env.example .env.local

Edit .env.local:

OLLAMA_BASE_URL=https://ollama.com
OLLAMA_API_KEY=sk-...your-key...
OLLAMA_DEFAULT_MODEL=gpt-oss:120b
SEARXNG_BASE_URL=http://localhost:8080

3. Install & run

npm install
npm run dev

Open http://localhost:3000.

Using pnpm? pnpm install && pnpm dev works identically.

---

Web search (for "All" / "Academic" modes)

Web search runs through the Ollama Cloud Web Search API, authenticated with the same OLLAMA_API_KEY — no separate search key or server required. Results already include page content, so no scraping step is needed. If search fails for any reason, KodaAI degrades to "No Search" mode with a banner.

Backend priority: Ollama Cloud Web Search → SearXNG → Brave.

SearXNG fallback (only needed if you run a local Ollama daemon, which has no web search):

docker run -d --name searxng -p 8080:8080 \
  -v searxng-data:/etc/searxng searxng/searxng:latest

SearXNG must return JSON — enable formats: [html, json] in its settings.yml.

Brave fallback (optional): set BRAVE_SEARCH_API_KEY in .env.

---

Want to run fully local instead of cloud?

The Ollama client is endpoint-agnostic. Point it at a local daemon:

OLLAMA_BASE_URL=http://localhost:11434
OLLAMA_API_KEY=
OLLAMA_DEFAULT_MODEL=llama3.2

ollama serve
ollama pull llama3.2

---

Architecture

User query
  └─ /api/search  → Ollama Cloud Web Search (top 5 + page content)
       └─ /api/scrape → cheerio fallback, only for results lacking content
            └─ /api/chat → Ollama Cloud (streamed SSE)
                 └─ follow-up questions (second, non-streaming call)

Path	Purpose
app/api/chat	Streaming chat (SSE) + follow-up generation
app/api/search	Ollama Cloud web search (SearXNG / Brave fallback)
app/api/scrape	URL → readable text (fallback only)
app/api/ollama/models	Lists Ollama Cloud models
lib/ollama.ts	Cloud client (bearer auth, streaming)
lib/store.ts	Zustand state (threads, model, focus mode)
hooks/useChat.ts	Orchestrates search → scrape → stream

---

Tech stack

Next.js 14 (App Router) · TypeScript · Tailwind CSS v3 · Radix UI primitives · Framer Motion · Zustand · react-markdown + remark-gfm + rehype-highlight · Lucide icons · Ollama Cloud.

---

Scripts

npm run dev     # development server
npm run build   # production build
npm run start   # serve production build
npm run lint    # eslint

---

Privacy notes

• Inference runs on your Ollama Cloud account — not on OpenAI/Anthropic.
• Web search (when enabled) goes to your SearXNG instance.
• Threads live in your browser's localStorage; nothing is uploaded by KodaAI.
• The privacy badge tracks third-party AI calls (always 0 by design).

---

Built as a privacy-respecting alternative to hosted AI search. 🔒