Skip to content

Everything you need to own your AI memory

01

Zero-friction ingestion

Paste, upload, watched folder, or MCP ingest. All paths feed one durable archive with full provenance tracking.

// ingest via MCP
ingest(({
  source: "chatgpt-export.zip",
  metadata: {
    provider: "openai",
    imported_at: Date.now()
  }
})
// ✓ 1,247 memories archived
02

Intelligent extraction

Facts, summaries, and embeddings with source spans and confidence tiers. Bring your own API key — your model, your bill.

// extracted fact
{
  fact: "Prefers TypeScript",
  source_span: "msg_291",
  confidence_tier: "high",
  derivation_method: "llm_extract"
}
03

Hybrid search & retrieval

Keyword, semantic, or hybrid search with Reciprocal Rank Fusion. Context-budgeted retrieval keeps responses within token limits, with source attribution on every result.

// context assembly response
{
  query: "typescript patterns",
  items: [ /* 8 results */ ],
  budget_used: 2841,
  provenance: [ /* sources */ ]
}
04

MCP-native retrieval

Agents and tools retrieve memory without manual intervention. Works with Claude Code, Cursor, and any MCP-compatible client.

05

Review & curate

Browse extracted facts, inspect provenance, promote to canonical memory, and resolve duplicates. Full visibility into what Recalium knows.

Facts 1,247 total

Prefers TypeScript over JavaScript for new projects

source: chatgpt-export/msg_291 · confidence: high

Uses Tailwind CSS for styling

source: claude-export/conv_18 · confidence: high

"I usually reach for Tailwind — it keeps everything colocated..."

↑ source span highlighted on hover

06

Private by default

Localhost-only by default. Sensitivity gate classifies content before indexing. Cascade deletion and a full audit trail.

Ready to own your AI memory?

Get Started Star on GitHub