xKSj | UdZddlmZddlZddlZddlmZddlmZmZddl m Z m Z m Z ddl mZddlmZdd lmZddlZej(j+d d Zej(j+d d Zej(j+ddZej(j+ddZdZej(j+dddk7Zeej(j+ddZeej(j+ddZda da!de"d<ddZ#ddZ$ddZ%eddZ&e d e&!Z'ejPe'dd"Z)Gd#d$eZ*Gd%d&eZ+Gd'd(eZ,e'j+d)d*Z- d dd+Z.e'j_d,e d-fdd.Z0e'j_d/e d-fdd0Z1e'j_d1e d-fdd2Z2hd3Z3dd4Z4dd5Z5 d dd7Z6dd8Z7dd9Z8dd:Z9e'j+d;dZ:e'j+d?dd@dd6de d-f ddAZ;GdBdCeZGdHdIeZ?e'j_dJe d-fddKZ@e'j+dLde d-f ddMZAGdNdOeZBe'j_dPe d-fddQZCe'j_dRe d-fddSZDGdTdUeZEe'j_dVe d-fddWZFe'j+dVde d-f ddXZGGdYdZeZHe'j_d[e d-fdd\ZIGd]d^eZJe'j_d_e d-f dd`ZKGdadbeZLe'j_dce d-fdddZMe'j+deddfdge d-f ddhZNe'j+diddje d-f ddkZOGdldmeZPe'j_dne d-f ddoZQGdpdqeZRe'j_dre d-f ddsZSGdtdueZTGdvdweZUe'j_dxe d-fddyZVe'j+dzdd6e d-f dd{ZWe'j_d|e d-fdd}ZXe'j_d~e d-fddZYe'j_de d-fddZZe'j+dde d-f ddZ[y)u[ Jarvis memory service — the shared brain (deploy on Railway). DUMB ON PURPOSE. There is NO model on this server. The intelligence lives in whatever model is already running on a device (Jarvis's Claude on Mac/Windows now, a local model on the phone later, future screenshot/transcribe daemons) — each one decides WHAT to remember and WHEN to recall, and structures the fact itself. This service only: • stores each memory as a node with a LOCAL embedding (FastEmbed, CPU — no LLM, no API), and • answers searches by MEANING (vector similarity) + returns CONNECTED facts (1 graph hop), so a vague query pulls back the linked context. Backed by FalkorDB (a graph DB with native vectors). Every device/daemon is a thin client that POSTs facts and GETs searches over HTTPS. Endpoints GET /health — liveness + memory count POST /remember — {text, kind?, when?, source?, group?, links?:[{to,type}]} → {id} GET /search?q=&k=&group=&hops= — connected facts by meaning GET /memories?group=&limit=&offset= — newest-first list (HUD viewer) GET /graph?group= — whole graph {nodes,edges} for the HUD graph view POST /link — {from, to, type} POST /unlink — {from_id, to} cut the edge(s) between two nodes POST /backfill_links — {dry_run?,threshold?} auto-link all existing memories GET /entity?name= — everything mentioning an entity ("pull up Al Temimi") POST /summary — {name, text} upsert an entity's current-state summary POST /retire — {id} soft-retire a memory (stale but kept, for dedup) POST /attach_entities — {memory_id, entities} attach entities (backfill hook) POST /insight — {text,title,itype} reflection finding · GET /insights RECALL is SALIENCE-ranked: relevance × recency × confidence (latest+firm facts surface first; nothing is deleted). Facts carry a `confidence` (0..1). AUTO-LINKING (Layer 1): every /remember & /supersede draws edges to the most similar existing memories (embedding + shared words) — no LLM. AUTOLINK=0 off. ENTITIES (Layer 2): the device model extracts people/companies/amounts; this service stores them as shared :Entity nodes + (:Memory)-[:MENTIONS]->(:Entity). POST /supersede — {old_id, text, ...} mark old fact stale + add new POST /forget — {id} permanently delete a memory + its links Env: FALKOR_HOST/FALKOR_PORT, MEMORY_API_KEY (bearer the devices send), MEMORY_GROUP (default namespace), EMBED_MODEL. ) annotationsN)asynccontextmanagerdatetimetimezone)FastAPIHeader HTTPException) BaseModel)FalkorDB) TextEmbeddingMEMORY_API_KEY MEMORY_GROUPahmed MEMORY_GRAPHjarvis EMBED_MODELzBAAI/bge-small-en-v1.5iAUTOLINK10AUTOLINK_THRESHOLDz0.85 AUTOLINK_MAX5zTextEmbedding | None _embeddercxttj|gDcgc] }t|c}Scc}wN)nextrembedfloat)textxs app.py_embedr$Ms0"9??D6#:; <;E!H; << >-@A::>>*;<D FB__Z (FC  $$-;.N P Q C =>D  $$-;.N P Q  5TB C CFLLN * 1!5T BC C CFLLN * 1!5T BC D CFLLN * 21#6d CDsHC98D:E;9 D7+D22D7: E8+E33E8; F9+F44F9cKttj4d{ddddd{y77#1d{7swYyxYwwr)rF mcp_serverlifespan)apps r# _lifespanrKss8M""$$ %$$$$$$sA#AAAA AAAAA A AAz Jarvis Memory)titlerIcDtr|dtk7r tddyy)NzBearer izbad or missing API key status_codedetail)API_KEYr ) authorizations r#_authrSs(=ggY$774LMM8wr*ceZdZUded<dZded<dZded<dZded <dZded <dZd ed <dZ d ed <dZ ded<dZ ded<y)RememberrAr!fact str | NonekindNwhenrsourcegrouplist[dict] | Nonelinksentities float | None confidencetone) __name__ __module__ __qualname____annotations__rXrYrZr[r]r^r`rar)r*r#rUrUsY ID*D*!FJ!E:#E #"&H&#J #D*r*rUc0eZdZUded<ded<dZded<y)LinkrAfrom_idtorelatedrWtypeN)rbrcrdrerkr)r*r#rgrgs L G D* r*rgceZdZUded<ded<dZded<dZded<d Zded <dZded <dZd ed <dZ ded<dZ ded<y) SupersederAold_idr!rVrWrXNrYrrZr[r\r^r_r`ra) rbrcrdrerXrYrZr[r^r`rar)r*r#rmrmsS K ID*D*!FJ!E:"&H&#J #D*r*rmz/healthcz tjdjdd}d|dS#t$rd}YwxYw)Nz MATCH (m:Memory) RETURN count(m)rT)okmemories)r=r> result_setr@)ns r#healthrusI LL; < G G J1 MA &&  s %, ::ctjj}| t|}tj d|||||t |xsd|t||xsdd |S)NzCREATE (m:Memory {id:$id, text:$text, kind:$kind, source:$source, group:$grp, created_at:$now, refers_to:$when, valid:true, confidence:$conf, tone:$tone}) SET m.embedding = vecf32($emb)r) idr!rXrZgrpr&rYembconfra)uuiduuid4hexr$r=r>r+r ) r!rXrYrZr[ryr`ramids r#_creatersi **,  C {Tl LL HD$&hj$*"Sz"DJB 8 9 Jr*z /remember)defaultc t||jjs tdd|jxst }|jj}t |}t||jxsd|j|jxsd|||j |jnd|j}|jxsgD]a}t|jdd j}|s/t j#d ||t|jd d d ct$rt'||||ng}t)||j*|} |t-|| dS)Nz empty textrNrVr?ryr`rarirMMATCH (a:Memory {id:$a}),(b:Memory {id:$b}) CREATE (a)-[:LINK {type:$t}]->(b)rkrjabt)rw auto_linkedr^)rSr!stripr r[ GROUP_DEFAULTr$rrXrYrZr`rar]rAr9r=r>r _auto_link_attach_entitiesr^len) bodyrRrxr!ryr~lkriautoentss r#rememberrs1 - 99?? LAA ** % C 99?? D ,C $ +VTYY++)3C040KT__QUyy "CzzR b! " ( ( * LL4RVVFI-F)GH J /7:c4c *BD C 4Dc$iT BBr*z/linkct|tjd|j|j|j xsddddiS)NrrjrrqT)rSr=r>rhrirkrrRs r#linkrsC - LL ,lltyy/EIFH $<r*z /supersedec <t||jxst}tj d|j t d|jj}t|}t||jxsd|j|jxsd|||j |jnd|j}tj d||j dt rt#||||t%||j&|}||d S) uA fact changed: mark the old one stale (kept for history — this is the 'temporal' bit) and add the new current one, linked to what it replaced.>MATCH (m:Memory {id:$id}) SET m.valid=false, m.invalid_at=$nowrwr&rVrrrzUMATCH (a:Memory {id:$a}),(b:Memory {id:$b}) CREATE (a)-[:LINK {type:'replaces'}]->(b)rr)rwr^)rSr[rr=r>rnr+r!rr$rrXrYrZr`rarrrr^)rrRrxr!ryr~rs r# superseders - ** % C LL%-1[[&LN 99?? D ,C $ +VTYY++)3C040KT__QUyy "C LL=-/3c3' C 4D4 ((r*>-riandoisitmemyofriandaredidforhasherhimhisshesirthewaswhoyoudoesfromhaveintoknowtellthatthemtheythiswerewhatrYwithyouraboutrwherewhichrrcddl}|jd|xsdjDchc] }|tvr|c}Scc}w)uGContent words (names/terms) of a fact — used to spot shared entities.rN [a-z0-9]{3,}r)refindallrB_STOP)r!_rews r#_significant_wordsrsI{{?TZR4F4F4HI I!~ I  s?cptjd||djdd}t|dkDS)NzDMATCH (a:Memory {id:$a})-[l:LINK]-(b:Memory {id:$b}) RETURN count(l)rr)r=r>rsr:)rrrts r# _edge_existsrs? Na $*Q (() +A q6A:r*Fc |tn|}|tn|}t|}i}tj d|dz|||dj } | D]I\} } } t |t| xsdz} t| d| zz}| |vs ||| dkDsC|| f|| <Kt|dd}|rd d jd tt |Dzd z}||d t|Dcic] \}}d|| c}}}tj ||j D]E\} } t |t| xsdz} | dk\s'dd| zz}| |vs ||| dkDs?|| f|| <Gt|jdd}g}|D]h\} \}} ||kr|St ||k\r|St|| r/|stj d|| d|j| | t!|ddj|Scc}}w)uLAYER 1 — connect a fact to the memories most like it, automatically. Candidates come from BOTH vector similarity (meaning) and shared significant words (so two notes that both name "Altimi" always link, even when the embeddings are only lukewarm). Best `max_links` above `threshold` get a :LINK edge of type 'similar' (skipping any pair already connected). No LLM.NzCALL db.idx.vector.queryNodes('Memory','embedding',$k,vecf32($emb)) YIELD node, score WHERE node.group=$grp AND node.valid=true AND node.id <> $nid RETURN node.id, node.text, score )kryrxnidrgQ?rzKMATCH (m:Memory) WHERE m.group=$grp AND m.valid=true AND m.id <> $nid AND ( OR c3&K|] }d| ywztoLower(m.text) CONTAINS $wNr).0rs r# z_auto_link..+7$5q9<$5z) RETURN m.id, m.text)rxrr333333?c|ddS)Nrr))kvs r#z_auto_link..7s Aqr*TkeyreversezTMATCH (a:Memory {id:$a}),(b:Memory {id:$b}) CREATE (a)-[:LINK {type:'similar'}]->(b)r)rwr!score)rrrr=r>rsrr listjoinrange enumeratesorteditemsrappendround)new_idr!ryr[ threshold max_linksdry_runmy_wordscandvrowscidctextrsharedcombinedwordskwqrrparamsrankedlinkedscs r#rr sd'0&7"YI ) 1 yI!$'HD LL ?"nS    $  #UEX 25;B ??@<$-/ d?hc15!5)DI # N2A E ({{7$)#e*$5777' ' v?,5e,<=,>JC$6u{$CCDF{.d?hc1&=!)5 1DI ?DJJL&94 HF F"[b% >  M v;) #  M  $  LL;3' )  S%%A,GH# M/>s*G5cddj|xsdjjS)z=Canonical key for an entity so the same name reuses one node. r)rrBsplit)names r# _norm_entityrPs' 88TZR&&(..0 11r*c d}|xsgD]}t|xsijddj}t|dkr?t|xsijddjj xsd}t |}|st jd||||tdt jd |||d |d z }|S) NrrrrrkthingzeMERGE (e:Entity {key:$key, group:$grp}) ON CREATE SET e.name=$name, e.etype=$etype, e.created_at=$now)rrxretyper&zWMATCH (m:Memory {id:$mid}),(e:Entity {key:$key, group:$grp}) MERGE (m)-[:MENTIONS]->(e))r~rrxr) rAr9rrrBrr=r>r+)mem_idr^r[rtrErrrs r#rrUs A ^^AG==,-335 t9q= ag2]]6734::<BBD 4    LteJ  !   )3u 5 7 Q'( Hr*cddlm}m}|jddd|jddzz}d}|jdxsd}|rU |j|j|j |z j d z }d d t|d d z zz}d |jdddz z}||z|zS#t$rY+wxYw)uRank facts by relevance + keyword hits + recency + confidence — the Generative-Agents recency×importance×relevance idea. Recency and confidence only RE-ORDER results; nothing is ever dropped from memory.rrrgg333333? keyword_hits created_atrg@g?g@g^@g?r`r?) rrr9r&r' fromisoformat total_secondsmaxr@)f_dt_tzbasereccaage_daysrzs r# _saliencerns: 55# na(@!@ @D C |  "B  (3+<+?C !%% d+c1 2D #:    s AB?? C  C z/search rcddl}t||xst}|rdnd}t|} tj d|zdzt |d| |dj} g} t} | D]h| jd| jddd d d ttd d d td ndddj|jd|jD cgc] } | tvr| c} dd}|rd|j!ddzdzdj#dt%t'|Dzdzdz}d|it)|D cic] \}} d||  c} }}tj ||jD]d| vr t+fd|D}| jddd d d tdd|zzd d td nd|dd | jd| j-t.d!"|dkDr|| rz| Dcgc]}|d# }}tj d$d%|ij}|D]Ad| vs | jd| jddd d d!d&Cd'| iScc} wcc} }wcc}w)(u_HYBRID recall: vector similarity (meaning) UNIONed with keyword matches (names/terms). A long detailed memory has a blurry embedding and can lose a short vague one on pure vector search, so we also surface any fact whose text literally contains the query's significant words — that's how "Ziyad" or "Journey Joy" reliably comes back.rNrzAND node.valid=true zlCALL db.idx.vector.queryNodes('Memory','embedding',$k,vecf32($emb)) YIELD node, score WHERE node.group=$grp z]RETURN node.id, node.text, node.kind, node.refers_to, node.created_at, score, node.confidencerrryrxrrrF)rwrVrX refers_tor rr`rrr$MATCH (m:Memory) WHERE m.group=$grp znode.zm.zAND (rc3&K|] }d| ywrr)rs r#rzsearch..s%O.s*E%Q110B0B0D+Dq%s))r g?) rwrVrXrr rr`r rTrrwzuMATCH (m:Memory)-[l:LINK]-(n:Memory) WHERE m.id IN $ids AND n.valid=true RETURN DISTINCT n.id, n.text, n.kind, l.typeids)rwrVrXrrfacts)rrSrr$r=r>rrssetaddrrr rrBrreplacerrrrsumsortr)qrr[hops include_stalerRrrxvalidryrowsr#seenrrkwrrhitsrr"neighr!s @r#searchr2s9 -  =CB%;E )C << 35: ; 2 2!QiC0  2 3=*  E 5D  1 AaD!A$!#$Q4qt$U1Q4[!434Q43CE!A$K % ' ( OQWWY? ?1? " $E 3emmGT6R R%Ors) r[limitoffsetr+rXrRrxr,kfrr-r9r!s r#rrrrs, -  =CB%8E $ "BVQCsO,.F v <<.6; : :     LL.69JJ    j $$% 'EJN)PJNQtQqT1Q4adtqtQqTZR AJN)P QQ)Ps)*Cc&eZdZUded<dZded<y)ForgetrArwNrWr[rbrcrdrer[r)r*r#r?r? GE:r*r?z/forgetct||jjs tddtj dd|jjiddiS)uPermanently delete a memory (and its links) — Ahmed removed a WRONG fact from the HUD viewer. Unlike /supersede (which keeps the old node as history), this is gone for good; use it only for genuinely bad/incorrect memories.rzno idrNz)MATCH (m:Memory {id:$id}) DETACH DELETE mrwrqT)rSrwrr r=r>rs r#forgetrCsQ  - 77==?G<< LL< (* $<r*z/graphc t||xst}|rdnd}tjd|zdzd|ij}tjd|rdndzdzd|ij}tjd d|ij}tjd |rdndzd zd|ij}|D cgc]} | d | d | d| dd} } | |D cgc]} d| d z| d d| dxsdzddc} z } |D cgc]} | d | d | dxsdd} } | |D cgc]} | d d| d zddc} z } | | dScc} wcc} wcc} wcc} w)uThe WHOLE memory graph for a group — :Memory nodes + :LINK edges, PLUS Layer-2 :Entity nodes and (:Memory)-[:MENTIONS]->(:Entity) edges — for the HUD's graph view. Ahmed walks/zooms this to see what's connected to what.rr5rz)RETURN m.id, m.text, m.kind, m.created_atrxzJMATCH (a:Memory)-[l:LINK]->(b:Memory) WHERE a.group=$grp AND b.group=$grp z"AND a.valid=true AND b.valid=true zRETURN a.id, b.id, l.typezAMATCH (e:Entity) WHERE e.group=$grp RETURN e.key, e.name, e.etypezMMATCH (m:Memory)-[:MENTIONS]->(e:Entity) WHERE m.group=$grp AND e.group=$grp zRETURN m.id, e.keyrrrr)rwr!rXr zent:zentity:rrj)rrirkmentions)nodesedgesrSrr=r>rs) r[r+rRrxr,nrowserowsentrowsmentrowsr!rFrGs r#graphrMs  -  =CB%8E LL.63 46;S\ CCM:  LL /2$H J $ $',S\  34>:  llK   j || /"5 7   %cl ,-7J   aD!A$!AaD I    !ad]AaD AaDOG4BH  E d!A$!0A B   !!t6AaD=*E!!EU ++   !sE#!E E.Ec0eZdZUded<ded<dZded<y)UnlinkrArhriNrWr[r@r)r*r#rOrOs L GE:r*rOz/unlinkczt|tjd|j|jdddiS)uCut the connection(s) between two memories (either direction) — Ahmed disconnected two nodes in the graph view. The nodes stay; only the edge goes.z=MATCH (a:Memory {id:$a})-[l:LINK]-(b:Memory {id:$b}) DELETE lrrqT)rSr=r>rhrirs r#unlinkrQ s7 - LLGll)+ $<r*z/entityc ^t||xst}t|}|sdgiStj d||dj }i}|D]C\}}} } } } } }|j ||| | xsd| xsdgd}|dj| | |dEdt|jiS)uEverything hanging off one entity — the 'pull up Al Temimi' view. Matches by canonical key with light fuzziness (substring either way).r^zMATCH (e:Entity)<-[:MENTIONS]-(m:Memory) WHERE e.group=$grp AND m.valid=true AND (e.key=$key OR e.key CONTAINS $key OR $key CONTAINS e.key) RETURN e.key, e.name, e.etype, e.summary, e.read, m.id, m.text, m.created_at ORDER BY e.name, m.created_at)rxrr)rrksummaryreadr#r#)rwrVr ) rSrrr=r>rs setdefaultrrvalues)rr[rRrxrr-rekeyenamerrSereadr~mtextmcreatedrEs r#entityr\+s  -  =C t C B << 5 C  " #-*  DDH@eUGUC OOD5%.5mU[b,.#0 1 ' e8LM EI T[[]+ ,,r*c0eZdZUded<ded<dZded<y)SummaryrArr!NrWr[r@r)r*r#r^r^Es I IE:r*r^z/summaryc lt||jxst}t|j}|r|j j s tddtjd|||jj |j j tdddiS)uUpsert a current-state SUMMARY onto an entity — the reflection engine's rolling 'profile' so 'pull up Al Temimi' returns the gist, not just raw facts.rname and text requiredrNzMERGE (e:Entity {key:$key, group:$grp}) ON CREATE SET e.name=$name, e.etype='thing', e.created_at=$now SET e.summary=$text, e.summary_at=$nowrrxrr!r&rqT rSr[rrrr!rr r=r>r+rrRrxrs r# summary_setrdKs - ** % C tyy !C diioo'4LMM LL 1C):"8: 7 8 $<r*z/readc lt||jxst}t|j}|r|j j s tddtjd|||jj |j j tdddiS)uTHEORY OF MIND — upsert a psychological 'read' onto a person/company: what they seem to want, how they decide, the state of the relationship, and how Ahmed should approach them. Models THEM, for Ahmed. (Reuses the Summary shape: {name, text}.)rr`rNzMERGE (e:Entity {key:$key, group:$grp}) ON CREATE SET e.name=$name, e.etype='thing', e.created_at=$now SET e.read=$text, e.read_at=$nowrarqTrbrcs r#read_setrf]s  - ** % C tyy !C diioo'4LMM LL +C):"8: 7 8 $<r*c&eZdZUded<dZded<y)ValuesrAr!NrWr[r@r)r*r#rhrhqs IE:r*rhz/valuesc t||jxst}|jj s t ddt jd||jj tdddiS)uVALUE MODEL — upsert Ahmed's north-star: his priorities, what he chases, what he avoids. A singleton profile the reflection engine and the brain consult so advice fits HIM, not generic best-practice.rz text requiredrNzOMERGE (p:Profile {id:'values', group:$grp}) SET p.text=$text, p.updated_at=$now)rxr!r&rqT) rSr[rr!rr r=r>r+)rrRrxs r# values_setrjvsj  - ** % C 99?? ODD LL .TYY__.xzBD $<r*ct||xst}tjdd|ij}|sdddS|ddxsd|ddxsddS)NzGMATCH (p:Profile {id:'values', group:$grp}) RETURN p.text, p.updated_atrxr)r! updated_atrrrH)r[rRrxr-s r# values_getrmsk -  =C <<Q    j  "--GAJ$"DGAJ4D" EEr*c&eZdZUded<dZded<y)RetirerArwNrWr[r@r)r*r#rororAr*roz/retirect|tjd|jj t dddiS)uSoft-retire a memory — mark it stale (hidden from recall) but KEEP it as history, unlike /forget which deletes. Used by the consolidation pass to fold away near-duplicate facts without losing anything.rrrqT)rSr=r>rwrr+rs r#retirerqs9  - LLQ  ;= $<r*c0eZdZUded<ded<dZded<y)AttachEntitiesrA memory_id list[dict]r^NrWr[r@r)r*r#rsrssNE:r*rsz/attach_entitiesct||jxst}t|j|j |}d|iS)uBackfill hook — attach extracted entities to an EXISTING memory (used by the engine-side entity backfill over the memories saved before Layer 2).attached)rSr[rrrtr^)rrRrxrts r#attach_entities_eprxs:  - ** % C0.92 RETURN count(node))rErxrF duplicate)createdreasonzCREATE (i:Insight {id:$id, text:$text, title:$title, itype:$itype, group:$grp, created_at:$now, surfaced:false}) SET i.embedding = vecf32($emb)rr{)rwr!rLr|rxr&ryT)rwr)rSr!rr r[rr$r=r>rsr:r@r{r|r}rLr|r+)rrRr!rxrydupiids r# insight_addrs - 99?? D ODD ** % C ,C ll !(+C"8:;E*  3s1vay>A%$ < < **,  C LL )D4::+;::. 3 ( )$ ''    s=C88 DDz /insightsTc  t||xst}|rdnd}tjd|zdz|t dt |ddj }d|Dcgc]}|d |d|d |d |d d c}iScc}w)NzAND i.surfaced=false rz%MATCH (i:Insight) WHERE i.group=$grp zYRETURN i.id, i.text, i.title, i.itype, i.created_at ORDER BY i.created_at DESC LIMIT $limrd)rxr8insightsrrrr)rwr!rLr|r )rSrr=r>rr:rs)r[unsurfaced_onlyr;rRrxcondr-r!s r# insights_listrs -  =C&5 "2D <</$6 0 0C3uc?34  67Aj  FJLFJ!"!adQqT#$Q4qt=FJL MMLsBz/insights/searchrc ddl}t||xst}t|}t }g} t j dt|ddz||djD]W} | d|vs |j| d|j| d| d| d| dtt| ddd Y |jd |jD cgc] } | t vr| c} dd } | rd d j#dt%t'| Dzdz} d|it)| D cic] \} } d| |  c} } } t j | |jD]A} | d|vs |j| d|j| d| d| d| ddd C d|dt|diS#t$rYwxYwcc} wcc} } w#t$rY7wxYw)uUnsurfaced insights RELEVANT to a query — used to raise a reflection contextually (only when Ahmed asks about something related), never on a timer.rNzCALL db.idx.vector.queryNodes('Insight','embedding',$k,vecf32($emb)) YIELD node, score WHERE node.group=$grp AND node.surfaced=false AND score > 0.5 RETURN node.id, node.text, node.title, node.itype, scorerrrrr)rwr!rLr|rrrz?MATCH (i:Insight) WHERE i.group=$grp AND i.surfaced=false AND (rc3&K|] }d| yw)ztoLower(i.text) CONTAINS $wNr))rrts r#rz"insights_search.. rrz') RETURN i.id, i.text, i.title, i.ityperxrrr)rrSrr$r$r=r>rrsr%rrr r@rrBrrrrr)r)r[rrRrrxryr.outr!rrrrtrs r#insights_searchrs-  -  =C )CDC    #1ay1}SE GHRz RA t41 !A$!qt%&qTE%!+q4IKLR OQWWY? ?1? " $E P{{7$)#e*$577799J %8H I8H11QC!8H IJ \\#v.99Q4t#HHQqTNJJadAaD1Q4)*1 =>: JSAY (('     !J   s<;G.AG G:G*G!;9G! GG! G-,G-c&eZdZUded<dZded<y) InsightSeenrArwNrWr[r@r)r*r#rrrAr*rz /insight/seencdt|tjdd|jiddiS)Nz.MATCH (i:Insight {id:$id}) SET i.surfaced=truerwrqTrSr=r>rwrs r# insight_seenrs0 - LLA" $<r*cFeZdZUdZded<dZded<dZded<dZd ed <y) BackfillNr_rz int | NonerFboolrrWr[)rbrcrdrrerrr[r)r*r#rr(s*"I|" Iz GTE:r*rz/backfill_linksc t||jxst}|j |jnt}|j xst }tjdd|ij}t}g}|D]\}} t| xsd} t|| xsd| ||||jD]`} t|| df} | |vr|j| t!|dks8|j#| xsddd| d xsddd| d d bt!|t!|||j|d S) uLAYER 1 backfill — run auto-linking over EVERY existing memory once, so today's loose cloud gets wired up. `dry_run` reports what it WOULD link (with samples) without touching the graph, so the threshold can be tuned first.NzHMATCH (m:Memory) WHERE m.group=$grp AND m.valid=true RETURN m.id, m.textrxr)rrw<r!r)rrr)rF linked_pairsrrsamples)rSr[rrrrrr=r>rsr$r$rr frozensetr%rr) rrRrxthrmxrFpairsrrntextryrpairs r#backfill_linksr/s@  - ** % C NN6$..rrZr[rr+)rrRtids r#task_addr^s - 99?? LAA **,  C LL ?DIIOO-dhhn" 'x 0Km   #;r*z/tasksc t||xst}tjd|rdndzdzd|ij}d|Dcgc].}|d|d|d xsdt |d |d |d xsdd 0c}iScc}w)Nz MATCH (t:Task) WHERE t.grp=$grp rzAND t.done=false z`RETURN t.id, t.text, t.due, t.done, t.created_at, t.done_at ORDER BY t.done, t.due, t.created_atrxtasksrrrrrr)rwr!rdoner done_at)rSrr=r>rsr)r[ include_donerRrxr-r!s r# task_listrms -  =C <<*2#6 8 / /    !j  9=?9=AqT1Q4! #AaDz1"#A$,$09=? @@?s3A>z /task/donecvt|tjd|jt dddiS)Nz7MATCH (t:Task {id:$id}) SET t.done=true, t.done_at=$nowrrqT)rSr=r>rwr+rs r# task_doner}s1 - LLJ 35 $<r*z /task/reopencdt|tjdd|jiddiS)Nz6MATCH (t:Task {id:$id}) SET t.done=false, t.done_at=''rwrqTrrs r# task_reopenrs. - LLI" $<r*z /task/deletecdt|tjdd|jiddiS)Nz MATCH (t:Task {id:$id}) DELETE trwrqTrrs r# task_deleters+ - LL3dDGG_E $<r*z/reminders/duect||xst}t}tj d||dj }|D]}tj dd|did|Dcgc]}|d|d|dd c}iScc}w) uuTasks whose reminder time has passed and aren't done/fired — the scheduler polls this to know what to announce.zMATCH (t:Task) WHERE t.grp=$grp AND t.done=false AND t.due<>'' AND t.due<=$now AND coalesce(t.fired,false)=false RETURN t.id, t.text, t.due)rxr&z(MATCH (t:Task {id:$id}) SET t.fired=truerwrrrr)rwr!r)rSrr+r=r>rs)r[rRrxr&r-r!s r# reminders_duers  -  =C *C << %.1#&> @AK   ?$!N TJT1Q41ad;TJ KKJs*B)r!rAreturn list[float])rrA)rNone)rJr)rRrWrr)NrN)r!rArXrArYrWrZrAr[rAryzlist[float] | Noner`r rarWrrA)rrUrRrW)rrgrRrW)rrmrRrW)r!rArr$)rrArrArr)NNF)rrAr!rAryrr[rArr rr:rrrru)rrArrA)rrAr^z list | Noner[rArr:)rdictrr ) r)rArr:r[rWr*r:r+rrRrW) r[rWr;r:r<r:r+rrXrWrRrW)rr?rRrW)r[rWr+rrRrW)rrOrRrW)rrAr[rWrRrW)rr^rRrW)rrhrRrW)r[rWrRrW)rrorRrW)rrsrRrW)rrzrRrW)r[rWrrr;r:rRrW)r)rAr[rWrr:rRrW)rrrRrW)rrrRrW)rrrRrW)r[rWrrrRrW)rrrRrW)\__doc__ __future__rr7r{ contextlibrrrfastapirr r pydanticr falkordbr fastembedr rHr8r9rQrr<rr?rr rr:rr=rrer$r+rFrKrJmountrSrUrgrmrurpostrrrrrrrrrrr2rrr?rCrMrOrQr\r^rdrfrhrjrmrorqrsrxrzrrrrrrrrrrrrrrrr)r*r#rs+X# *'22# **..)2 . ~w7 ZZ^^NH 5 jjnn],DE  ::>>*c *c 12::>>*>GH2::>>.#67 "& &=2C<Oi8 N y !9!  ''379=  /  ,6 BE  +9?9MCC.'171E,;A$;O)), M:>$<<36<<)3<L2  2&Da %&,T&:AA#AAH!%Ba#(T)/)=Q Q0:Q&QQ6Y )5;D5I  "%&,T&:",#",",JY )5;D5I*.'-d';-$--2i *;A$;O"'8>t8L&Y )9?9M  #'+1$+? F( F FY )5;D5IY  39$3G&0i*;A$;O((:&*D!.4T.B M M!+ M M 04q06t0D()#-()()V) /-3D-A *y /5d/CK",KKD"9"Y '5;D5I  "&U*0*> @' @ @,8>t8L.:@:N.:@:N  &*.4T.BL!+LLr*