Research-paper corpus
Keep one Markdown note per paper (abstract, key claims, your critique) and Memory Layer becomes a cited, queryable literature memory.
Ingest the corpus
memory ingest ~/research/notes --project lit-review \
--type documentation --tag corpus-2026 --dry-run
memory ingest ~/research/notes --project lit-review \
--type documentation --tag corpus-2026The --tag marks the whole corpus so it can be filtered, re-ingested, or bulk-archived as one unit when the review cycle ends.
Query across the literature
memory query --project lit-review \
--question "Which papers found spacing effects in procedural learning?"Every answer cites the notes it draws from — and refuses when your corpus doesn't actually support an answer, which matters more in research than anywhere else. Retrieval is keyless (lexical); configure an embedding provider to add semantic search for paraphrased questions.
Structured claims (optional, better)
For higher-precision recall, post distilled claims per paper through the HTTP API instead of raw notes — one structured_candidates entry per claim with the paper as provenance:
curl -s http://127.0.0.1:4040/v1/capture/task \
-H "Authorization: Bearer $MEMORY_API_TOKEN" -H "Content-Type: application/json" -d '{
"project": "lit-review", "task_title": "Cepeda 2006 notes",
"user_prompt": "Distill Cepeda et al. 2006", "writer_id": "reading-notes",
"agent_summary": "Distilled distributed-practice meta-analysis.",
"structured_candidates": [{
"canonical_text": "Cepeda et al. 2006 meta-analysis: spacing effects are robust across 254 studies; optimal gap scales with retention interval.",
"summary": "Cepeda 2006: spacing robust; optimal gap scales with retention interval.",
"memory_type": "domain_fact", "confidence": 0.9, "importance": 4,
"tags": ["corpus-2026", "spacing"],
"sources": [{"source_kind": "note", "excerpt": "Cepeda, Pashler, Vul, Wixted & Rohrer (2006)"}]
}]}'
curl -s http://127.0.0.1:4040/v1/curate \
-H "Authorization: Bearer $MEMORY_API_TOKEN" -H "Content-Type: application/json" \
-d '{"project": "lit-review"}'Share it
memory bundle export --project lit-review --out lit-review.mlbundle.zipBundles are privacy-safe by default (no raw captures, writer IDs, or local paths) and import idempotently — a colleague can memory bundle import your corpus and query it immediately. In the Graph tab, co-cited papers cluster visibly as relation edges accumulate between frequently co-retrieved notes.