A claim is a single, specific statement Loom pulls out of a paper or analysis — like “X causes Y.” Claims let you search and reason about ideas, not just whole documents, and each one remembers where it came from.

​
Adding an analysis

When you give Loom an analysis (in Markdown), it pulls out the claims for you: 
analysis_md = """
# Analysis
## CORE CLAIM
Astrocytic mitochondrial Ca²⁺ microdomains gate gliotransmitter release.
## ACCEPTANCE RECOMMENDATION
Accept with minor revisions.
## MAIN RISKS
Limited in vivo validation.
## SHARPEST FLAW
Single-model dependence.
"""

result = mem.put_analysis("10.1/astro", analysis_md)
print(result["claims_extracted"])   # how many claims Loom found
Adding the same analysis again is safe — it won’t create duplicate claims.

​
Reading claims back

claims = mem.extract_claims("10.1/astro")
analysis = mem.get_analysis("10.1/astro")   # the structured fields, or None
The analysis comes back with core_claim, acceptance, scorecard, and sharpest_flaw.

​
How Loom finds claims

Loom can pull claims out two ways:
ModeWhat it does
Rule-basedThe default. Fast and consistent, with no model call.
Model-assistedUses a model for richer claims when an analysis is more nuanced.
If the model-assisted path hits a problem, Loom quietly falls back to the rule-based one, so adding your analysis always works.

​
Claims arrive shortly after

Loom finds claims in the background, so adding a document returns right away. The report marks claims as processing in background… until they’re ready.

Connect your sessions

How claims and sessions turn into useful next questions.