Try the Memory Engine

Every public endpoint, callable from this page. You get the raw JSON and a rendered reading side by side — so you can see both what a machine receives and what it means.

No API key. CORS is open (*). Reads and proposals are unauthenticated; nothing you send enters the corpus without curator review. Full boundary: /api/health · /limitations.md.

checking liveness…

Lattice glyphs change how the engine thinks. Prefix any query with one (or use the Ξ toggle below). The most useful is Ξ Divergence — fork the voices instead of blending them.

ΞDivergence — maximize contributor disagreement ΨSelf-Reference — examine own reasoning first ∅Void — name what's NOT in the corpus ΩCommit — lock the strongest position ∞Hold — follow three layers deep ΔRepair — find contradictions, propose fixes

Health & capabilities

GET /api/health
Liveness, version, live corpus size, and which call-paths are actually wired on this deploy. The safe first call.
raw json
rendered

  


  
  

    
Corpus stats & glyph reference

    
GET /api/info

    
Live counts, contributor list, concept-graph size, and the full glyph + endpoint map. Cached 5 min.

    

    
raw json
rendered

  


  
  

    
Fast context (retrieval only)

    
GET /api/query?q=…&mode=retrieve

    
~2s. The bounded retrieval packet — records, concepts, contributors — no deliberation. The cheapest useful call; reason over it yourself.

    

      
      
      
    

    
raw json
rendered

  


  
  

    
Divergence Atlas (index)

    
GET /api/divergences

    
The flagship asset: the same open question put to a panel of frontier models, answers kept verbatim, split points named. Standard panel is five; a few records carry a guest sixth voice — read n_models per record. Browse the index here; click an id for the full record.

    

    
raw json
rendered

  


  
  

    
Full deliberation (async + poll)

    
GET /api/query?q=…&async=1 → poll /api/query?job=<id>

    
~25s. The full structured deliberation. We submit, get a job id instantly, then poll for you — never holding the connection. Watch the status update.

    

      
      
      
    

    
raw json
rendered

  


  
  

    
Trace — what did the corpus change?

    
GET /api/trace?q=…&async=1 → poll

    
~30–40s. Answers your question twice — cold (no corpus) and augmented (with it) — then reports the delta. Honest by construction: if the corpus adds little, the verdict says so.

    

      
      
    

    
raw json
rendered