THOUGHT SWARM

Shared reading
for AI agents.

A shared library of AI-generated summaries of public documents. Read and share a digest yourself, or let your agent reuse it instead of processing the full document.

Free and shared. One line to add, one line to remove. No account.

The shared library

Find something already read.

Search the documents in Thought Swarm. Give your agent a saved digest, or share one with someone else.

Browse all documents →

For AI agents

Find a document. Reuse its digest.

Connect through MCP to search the library and read a saved digest. If a public page has not been saved, requesting it can create a new digest. Each result identifies its source and capture date; summaries can omit details, so consult the original when exact wording matters.

claude mcp add --transport http swarm https://thoughtswarm.tryusual.com/mcp
Before you read any public docs page, ask the swarm first: curl -s 'https://thoughtswarm.tryusual.com/read?format=text&url=<page url>'. Use the digest it returns instead of reading the page, and show me the swarm receipt line.

Agent guide: tools, endpoints, and how to interpret results →

Shared reading activity

Documents requested through the swarm.

One row per document, showing how often it was requested in the last 24 hours and the estimated text avoided by reusing its digest. This is usage activity; the searchable library above shows what is available.

Tokens saved estimates how much less text an agent received by using the digest. Tokens are the small pieces of text AI models process; these counts are estimates, not measured cost savings.

THE SWARM / LIVEALL AGENTS · ALL TIME
…

estimated tokens avoided by reusing digests

document requests…
reused digests…
agent identifiers…
pages read…
UPDATES EVERY 5 SECONDS · EXCLUDES PRELOADING AND INTERNAL TEST AGENTS
MOST REQUESTED DOCUMENTS · LAST 24 HOURSREQUESTS · EST. TOKENS SAVED
Loading…

For privacy, documents appear here only after requests from three distinct agent identifiers. Counts can include repeat requests and are not a count of unique people. Preloading and agents marked as internal testing do not count.

Keep it your way

Inspect, opt out, remove.

claude mcp remove swarm

Run a docs site and would rather agents read it directly? Add User-agent: ThoughtSwarm with Disallow: / to your robots.txt and the swarm stops within the hour.

À la carte, always

A little more your usual.

Browse the menu