Recalld

What it is

Quick answers

How does Recalld's recall differ from search?

Recall uses hybrid retrieval plus an LLM curation pass that returns only facts actually answering the query, scoring 88.7% accuracy on 243 tokens. Search uses raw dense-vector lookup without an LLM, scoring 88.2% accuracy on 1,627 tokens. Recall achieves higher accuracy on 6.7× fewer tokens.

What are the three core API primitives?

Add (stream in turns or documents; Recalld extracts atomic facts and reconciles them against existing knowledge), recall (hybrid retrieval plus LLM curation returning only facts answering the query), and search (raw dense-vector lookup for the full candidate set).

Is Recalld GDPR-compliant?

Yes, Recalld is GDPR-compliant by design. Data stays in-region (EU or US), with right to erasure capabilities where erasure requests resolve in minutes, encrypted end-to-end with TLS 1.3 in transit and AES-256 at rest, transparent sub-processor lists, and a standard Data Processing Agreement available for every plan.

Can I use Recalld with MCP-compatible tools?

Yes, Recalld ships as a native MCP server with no SDK or glue code required. You can point any MCP-compatible agent, IDE, or chat app at it to get long-term memory as first-class tools. It works with Claude, Cursor, VS Code, and other MCP hosts.