Local-first agent infrastructure

Shrink the context your agents need.

TinySuite is a family of open-source, token-light tools for local LLMs and MCP agents.

Built for developers who want smaller prompts, better retrieval, and local-first agent infrastructure.

TinySearch logoTinySearch
$ tinysearch research "best local MCP search flow"

search -> crawl -> chunk -> rerank -> grounded prompt

RESULTS
1. Source-backed snippet
2. Source-backed snippet
3. Source-backed snippet

Start with TinySearch.

TinySearch is the first TinySuite product: a self-hosted web research layer that returns compact, source-grounded context for MCP agents.

TinySearch

Live
Shrink the web for local LLMs.

A self-hosted web research layer for MCP agents. Search, crawl, rerank, and return only the source-grounded chunks that matter.

TinyContext

Planned
Memory that fits your context window.

A future local-first memory layer for agents. TinySearch is the live product today.

Why TinySuite exists

Most agent tools are built like context is free.

TinySuite takes the opposite approach: search less, retrieve better, keep prompts smaller, and make local models more useful.

Token-light by default

Return only the context the model needs.

Local-first

Self-hosted tools without hosted dashboards or account requirements.

Agent-native

Designed for MCP workflows, local LLMs, and practical automation.

TinySearch is live

Compact web research for MCP agents.

TinySearch gives your agent a compact research workflow: search -> crawl -> chunk -> rerank -> grounded prompt.

Instead of dumping full webpages into context, TinySearch returns the most relevant source-backed snippets.

QUESTION
What changed in the latest release?

RESULTS
1. Title
   URL
   Relevant extracted text...

INSTRUCTIONS
Answer only from the results. Cite source URLs.

Start here

Install TinySearch, then connect your MCP client.

The docs walk through Docker, client setup, source mode, examples, and troubleshooting.