TinySearch
Web research for the model. It searches, reads and reranks pages locally, then returns compact evidence with its sources attached. No search API key.
- Search and page reading
- Source-grounded results
- A browser for pages that need clicking
TinySuite builds small, open tools for local agents. Our flagship, TinyWebUI, brings TinySearch and TinyContext together in one chat.
TinyWebUI
TinyWebUI is a small chat workspace for any OpenAI-compatible model. It comes with web research and long-term memory already connected, and keeps your chats, documents and tool output in one SQLite file on your machine.
npx tinywebuiNeeds Node.js 22.13+ and uv. Then open http://127.0.0.1:7777 and add a model.
No account, no hosted backend and nothing to build: the page ships prebuilt. Settings live in a config file you can read, and every change in the Settings panel is written back to it.
npx tinywebui starts the server on 127.0.0.1:7777. It needs Node.js 22.13 or later, and uv for the bundled tools.
OpenRouter, OpenAI, Groq, or a local runtime like Ollama, vLLM, LM Studio or llama.cpp. Anything that speaks the OpenAI API.
TinySearch and TinyContext are already connected, so the first chat can search the web and remember what matters.
TinyWebUI comes with TinySearch for web research and TinyContext for memory, both connected from the first run. You can also use either one on its own as an MCP server in Claude Code, Cursor, or any other MCP client.
Web research for the model. It searches, reads and reranks pages locally, then returns compact evidence with its sources attached. No search API key.
Memory across chats. It stores concise facts in local SQLite and recalls only the ones that fit the available context.
TinySuite favors selected evidence and durable facts over raw pages, old transcripts, and undifferentiated history. TinyWebUI applies the same rule to the chat itself: large tool results are compacted once, prompt caches stay warm, and every round shows where its tokens went.
Guides and engineering notes on web search, memory and context for agents.
TinySearch 0.7.0 improves scripted-page scraping, puts main content first, uses one Playwright driver, isolates browser sessions, and locks build dependencies.
Add persistent memory to Claude Code, Cursor, Codex or your own Python agent. Copy-paste setup, working code, and why it beats chat history and vector databases.
Add web search to Claude Code, Cursor, Codex or your own Python agent with any LLM. No API key. Copy-paste setup, working code, and how it beats search APIs.