Context Streams

As you use Recurse, you're already building something valuable: a structured knowledge base. You subscribe to sources, upload documents, make connections. Recurse extracts frames and builds relationships.

Context Streams let you share parts of it. Select a subgraph—specific sources, documents, the connections between them—and package it for others. Your team subscribes to your research methodology. Your community accesses your domain expertise. They're not just reading your writing—they're accessing how you organize and connect information.

This works both ways. You subscribe to fifteen newsletters from thoughtful people in your field, but they pile up. Ninety-three unread issues. Context Streams solve this: subscribe to someone else's knowledge base directly (like Neal Stephenson's information editors in "Fall; or, Dodge in Hell"). Their newsletters, their synthesis, their connections—your queries draw from their curation. As they learn, your knowledge updates.

The key is domain authority—who you trust. An AI researcher maintains their understanding of RAG patterns. You recognize their expertise, subscribe to their stream. Your queries now draw from years of their curation. You're not evaluating sources yourself—you're trusting their evaluation.

How It Works

Create a stream using the context assembler. Select which parts of your knowledge base to include: source subscriptions, documents, synthesis notes, frame relationships. Choose visibility: private, team, or public.

When someone subscribes, that subgraph integrates with their knowledge base. When you add sources or make connections, their streams update automatically.

Example: Without streams, asking about RAG best practices means Googling blog posts, reading GPT's outdated advice, spending days building understanding. With a stream like #rag-patterns, the system retrieves from the curator's subgraph—their synthesis, papers, blogs, connections. 30 seconds.

Query syntax:

#rag-patterns What are the tradeoffs between dense and sparse retrieval?

@researcher What's new in retrieval augmented generation?

Use #collection-slug to target specific streams you're subscribed to, or @username to query a curator's public streams.

Creating and Sharing

You're already building your knowledge base through normal use—uploading docs, subscribing to sources, querying your knowledge. You're not creating content to share. You're just making what you already built accessible. Creating a stream means exposing a subgraph of it. Share your research with your team. Make a subgraph public for your domain. Provide subscription access to clients. Control which subgraph gets exposed—share your research publicly while keeping client work private.

Privacy and ownership: Subscriptions don't copy raw documents—they point to sources that subscribers ingest into their own graphs, with clear attribution and audit trails. You control which subgraph gets exposed. Subscribers can scope which conversations use which streams.

This creates natural knowledge exchange: you build expertise, others build expertise. Share subgraphs, subscribe to trusted experts in domains you touch occasionally. You're doing what you're doing anyway—now others can benefit from it.

Related Concepts