graphsight 0.3.0__tar.gz → 0.3.1__tar.gz

{graphsight-0.3.0/graphsight.egg-info → graphsight-0.3.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: graphsight
-Version: 0.3.0
+Version: 0.3.1
 Summary: See why your AI agent retrieved what it did — interactive graph viewer for agent retrieval traces. One command, zero dependencies, nothing leaves your machine.
 Author: Arush Karnatak
 License: MIT
@@ -28,20 +28,42 @@ Dynamic: license-file
 [![Python 3.10+](https://img.shields.io/badge/python-3.10+-blue.svg)](https://pypi.org/project/graphsight/)
 [![License: MIT](https://img.shields.io/badge/license-MIT-yellow.svg)](https://github.com/Kcodess2807/graphsight/blob/main/graphsight/LICENSE)
 
-**See exactly why your AI agent retrieved what it did.**
+**See which retrieved documents your agent actually used — and which it ignored.**
 
-Your agent answered a question. Which documents did it actually pull? What
-scores did they get? How are they connected to each other? Most stacks make
-you dig through logs to answer that. Graphsight renders the run as an
-**interactive graph in your browser** — one command, zero dependencies,
+Your agent answered a question. Which documents did it actually pull? Which
+of those did the answer come from? What scores did they get, and how are they
+connected? Most stacks make you dig through logs. Graphsight renders the run
+as an **interactive graph in your browser** — one command, zero dependencies,
 nothing leaves your machine.
 
+![Graphsight showing a retrieved-but-unused document](https://raw.githubusercontent.com/Kcodess2807/graphsight/main/docs/media/retrieved-vs-used.gif)
+
+*A real run. `PR #101` scored **0.910** — the highest of anything retrieved — and the answer
+never used it. `PR #412` scored **0.340** and is the one that answered.*
+
 ```bash
 pip install graphsight
 graphsight path/to/trace_state.json
 ```
 
-## What you get
+## Retrieved vs. used
+
+The signal that isn't in your logs. When a trace carries the final answer,
+every retrieved item is scored by lexical overlap against it and rendered
+either **highlighted** (surfaced in the answer) or **dimmed** with the label
+*"retrieved, unused."* That splits the two classic retrieval failures at a
+glance:
+
+- **Right document retrieved, ignored by the model** → a dimmed node with a
+  high retrieval score. Your retriever worked; your prompt or context order
+  didn't.
+- **Wrong document trusted** → a highlighted node that shouldn't be.
+
+The overlap is a lexical heuristic, labeled as such (threshold 0.2). No LLM
+re-reads your evidence, and no score is invented — a trace with no answer
+attached makes no usage claims at all.
+
+## What else you get
 
 - **Every retrieved item as a typed node** — PR, Service, Person, Ticket,
   Document, Repo, Library, Team, Tool — with its retrieval score.
@@ -86,10 +108,13 @@ graphsight .graphsight/
 
 The [graphsight-langgraph](https://pypi.org/project/graphsight-langgraph/)
 `capture()` helper appends every agent run there automatically, so your
-debugging history accumulates with zero ceremony. Traces render with the
-**retrieved vs. used** split: items that surfaced in the answer highlighted,
-items retrieved but ignored dimmed — the two classic retrieval failures,
-visible at a glance.
+debugging history accumulates with zero ceremony — no setup, no database.
+
+Because the history is a directory of plain files, you can **compare two
+runs side by side**: what the retrieval returned before a prompt change and
+after it, which items appeared or vanished, and which flipped between used
+and ignored. That is usually the fastest way to answer "what did my edit
+actually do to retrieval?"
 
 ## Sharing traces with your team
 
@@ -122,8 +147,11 @@ producers:
   graphsight graphsight_out/trace_state.json
   ```
 
-- **The TraceRAG engine** — the graph-memory backend this project grew out
-  of; its `/api/trace` responses are the same shape.
+- **The Graphsight graph-memory engine** — the backend this project grew out
+  of: GitHub events become a live knowledge graph with typed, timestamped
+  edges (`AUTHORED`, `RESOLVES`, `TOUCHES`), queried by a hybrid
+  vector + graph router. Its `/api/trace` responses are the same shape. See
+  the [main repository](https://github.com/Kcodess2807/graphsight).
 
 Adapters for LlamaIndex and raw OpenTelemetry spans are planned; all
 producers emit the same schema and render in this same viewer.

{graphsight-0.3.0 → graphsight-0.3.1}/README.md

@@ -4,20 +4,42 @@
 [![Python 3.10+](https://img.shields.io/badge/python-3.10+-blue.svg)](https://pypi.org/project/graphsight/)
 [![License: MIT](https://img.shields.io/badge/license-MIT-yellow.svg)](https://github.com/Kcodess2807/graphsight/blob/main/graphsight/LICENSE)
 
-**See exactly why your AI agent retrieved what it did.**
+**See which retrieved documents your agent actually used — and which it ignored.**
 
-Your agent answered a question. Which documents did it actually pull? What
-scores did they get? How are they connected to each other? Most stacks make
-you dig through logs to answer that. Graphsight renders the run as an
-**interactive graph in your browser** — one command, zero dependencies,
+Your agent answered a question. Which documents did it actually pull? Which
+of those did the answer come from? What scores did they get, and how are they
+connected? Most stacks make you dig through logs. Graphsight renders the run
+as an **interactive graph in your browser** — one command, zero dependencies,
 nothing leaves your machine.
 
+![Graphsight showing a retrieved-but-unused document](https://raw.githubusercontent.com/Kcodess2807/graphsight/main/docs/media/retrieved-vs-used.gif)
+
+*A real run. `PR #101` scored **0.910** — the highest of anything retrieved — and the answer
+never used it. `PR #412` scored **0.340** and is the one that answered.*
+
 ```bash
 pip install graphsight
 graphsight path/to/trace_state.json
 ```
 
-## What you get
+## Retrieved vs. used
+
+The signal that isn't in your logs. When a trace carries the final answer,
+every retrieved item is scored by lexical overlap against it and rendered
+either **highlighted** (surfaced in the answer) or **dimmed** with the label
+*"retrieved, unused."* That splits the two classic retrieval failures at a
+glance:
+
+- **Right document retrieved, ignored by the model** → a dimmed node with a
+  high retrieval score. Your retriever worked; your prompt or context order
+  didn't.
+- **Wrong document trusted** → a highlighted node that shouldn't be.
+
+The overlap is a lexical heuristic, labeled as such (threshold 0.2). No LLM
+re-reads your evidence, and no score is invented — a trace with no answer
+attached makes no usage claims at all.
+
+## What else you get
 
 - **Every retrieved item as a typed node** — PR, Service, Person, Ticket,
   Document, Repo, Library, Team, Tool — with its retrieval score.
@@ -62,10 +84,13 @@ graphsight .graphsight/
 
 The [graphsight-langgraph](https://pypi.org/project/graphsight-langgraph/)
 `capture()` helper appends every agent run there automatically, so your
-debugging history accumulates with zero ceremony. Traces render with the
-**retrieved vs. used** split: items that surfaced in the answer highlighted,
-items retrieved but ignored dimmed — the two classic retrieval failures,
-visible at a glance.
+debugging history accumulates with zero ceremony — no setup, no database.
+
+Because the history is a directory of plain files, you can **compare two
+runs side by side**: what the retrieval returned before a prompt change and
+after it, which items appeared or vanished, and which flipped between used
+and ignored. That is usually the fastest way to answer "what did my edit
+actually do to retrieval?"
 
 ## Sharing traces with your team
 
@@ -98,8 +123,11 @@ producers:
   graphsight graphsight_out/trace_state.json
   ```
 
-- **The TraceRAG engine** — the graph-memory backend this project grew out
-  of; its `/api/trace` responses are the same shape.
+- **The Graphsight graph-memory engine** — the backend this project grew out
+  of: GitHub events become a live knowledge graph with typed, timestamped
+  edges (`AUTHORED`, `RESOLVES`, `TOUCHES`), queried by a hybrid
+  vector + graph router. Its `/api/trace` responses are the same shape. See
+  the [main repository](https://github.com/Kcodess2807/graphsight).
 
 Adapters for LlamaIndex and raw OpenTelemetry spans are planned; all
 producers emit the same schema and render in this same viewer.

{graphsight-0.3.0 → graphsight-0.3.1}/graphsight/__init__.py

@@ -8,4 +8,4 @@ Serves the bundled Graphsight Studio UI locally and opens the trace in the
 browser. No backend, no accounts, no network calls beyond localhost.
 """
 
-__version__ = "0.3.0"
+__version__ = "0.3.1"

graphsight-0.3.0/graphsight/dist/assets/arc-Bsma0Xbp.js → graphsight-0.3.1/graphsight/dist/assets/arc-DGOsvhEX.js

@@ -1 +1 @@
-import{$ as ln,a0 as an,a1 as G,a2 as q,a3 as z,a4 as un,a5 as y,a6 as tn,a7 as K,a8 as _,a9 as rn,aa as o,ab as on,ac as sn,ad as fn}from"./mermaid.core-Cr4Ida0l.js";function cn(l){return l.innerRadius}function yn(l){return l.outerRadius}function gn(l){return l.startAngle}function dn(l){return l.endAngle}function mn(l){return l&&l.padAngle}function pn(l,h,I,D,v,A,B,a){var O=I-l,i=D-h,n=B-v,d=a-A,u=d*O-n*i;if(!(u*u<y))return u=(n*(h-A)-d*(l-v))/u,[l+u*O,h+u*i]}function V(l,h,I,D,v,A,B){var a=l-I,O=h-D,i=(B?A:-A)/K(a*a+O*O),n=i*O,d=-i*a,u=l+n,s=h+d,f=I+n,c=D+d,C=(u+f)/2,t=(s+c)/2,m=f-u,g=c-s,R=m*m+g*g,T=v-A,P=u*c-f*s,S=(g<0?-1:1)*K(on(0,T*T*R-P*P)),$=(P*g-m*S)/R,j=(-P*m-g*S)/R,w=(P*g+m*S)/R,p=(-P*m+g*S)/R,x=$-C,e=j-t,r=w-C,F=p-t;return x*x+e*e>r*r+F*F&&($=w,j=p),{cx:$,cy:j,x01:-n,y01:-d,x11:$*(v/T-1),y11:j*(v/T-1)}}function hn(){var l=cn,h=yn,I=z(0),D=null,v=gn,A=dn,B=mn,a=null,O=ln(i);function i(){var n,d,u=+l.apply(this,arguments),s=+h.apply(this,arguments),f=v.apply(this,arguments)-un,c=A.apply(this,arguments)-un,C=rn(c-f),t=c>f;if(a||(a=n=O()),s<u&&(d=s,s=u,u=d),!(s>y))a.moveTo(0,0);else if(C>tn-y)a.moveTo(s*G(f),s*q(f)),a.arc(0,0,s,f,c,!t),u>y&&(a.moveTo(u*G(c),u*q(c)),a.arc(0,0,u,c,f,t));else{var m=f,g=c,R=f,T=c,P=C,S=C,$=B.apply(this,arguments)/2,j=$>y&&(D?+D.apply(this,arguments):K(u*u+s*s)),w=_(rn(s-u)/2,+I.apply(this,arguments)),p=w,x=w,e,r;if(j>y){var F=sn(j/u*q($)),L=sn(j/s*q($));(P-=F*2)>y?(F*=t?1:-1,R+=F,T-=F):(P=0,R=T=(f+c)/2),(S-=L*2)>y?(L*=t?1:-1,m+=L,g-=L):(S=0,m=g=(f+c)/2)}var H=s*G(m),J=s*q(m),M=u*G(T),N=u*q(T);if(w>y){var Q=s*G(g),U=s*q(g),W=u*G(R),X=u*q(R),E;if(C<an)if(E=pn(H,J,W,X,Q,U,M,N)){var Y=H-E[0],Z=J-E[1],b=Q-E[0],k=U-E[1],nn=1/q(fn((Y*b+Z*k)/(K(Y*Y+Z*Z)*K(b*b+k*k)))/2),en=K(E[0]*E[0]+E[1]*E[1]);p=_(w,(u-en)/(nn-1)),x=_(w,(s-en)/(nn+1))}else p=x=0}S>y?x>y?(e=V(W,X,H,J,s,x,t),r=V(Q,U,M,N,s,x,t),a.moveTo(e.cx+e.x01,e.cy+e.y01),x<w?a.arc(e.cx,e.cy,x,o(e.y01,e.x01),o(r.y01,r.x01),!t):(a.arc(e.cx,e.cy,x,o(e.y01,e.x01),o(e.y11,e.x11),!t),a.arc(0,0,s,o(e.cy+e.y11,e.cx+e.x11),o(r.cy+r.y11,r.cx+r.x11),!t),a.arc(r.cx,r.cy,x,o(r.y11,r.x11),o(r.y01,r.x01),!t))):(a.moveTo(H,J),a.arc(0,0,s,m,g,!t)):a.moveTo(H,J),!(u>y)||!(P>y)?a.lineTo(M,N):p>y?(e=V(M,N,Q,U,u,-p,t),r=V(H,J,W,X,u,-p,t),a.lineTo(e.cx+e.x01,e.cy+e.y01),p<w?a.arc(e.cx,e.cy,p,o(e.y01,e.x01),o(r.y01,r.x01),!t):(a.arc(e.cx,e.cy,p,o(e.y01,e.x01),o(e.y11,e.x11),!t),a.arc(0,0,u,o(e.cy+e.y11,e.cx+e.x11),o(r.cy+r.y11,r.cx+r.x11),t),a.arc(r.cx,r.cy,p,o(r.y11,r.x11),o(r.y01,r.x01),!t))):a.arc(0,0,u,T,R,t)}if(a.closePath(),n)return a=null,n+""||null}return i.centroid=function(){var n=(+l.apply(this,arguments)+ +h.apply(this,arguments))/2,d=(+v.apply(this,arguments)+ +A.apply(this,arguments))/2-an/2;return[G(d)*n,q(d)*n]},i.innerRadius=function(n){return arguments.length?(l=typeof n=="function"?n:z(+n),i):l},i.outerRadius=function(n){return arguments.length?(h=typeof n=="function"?n:z(+n),i):h},i.cornerRadius=function(n){return arguments.length?(I=typeof n=="function"?n:z(+n),i):I},i.padRadius=function(n){return arguments.length?(D=n==null?null:typeof n=="function"?n:z(+n),i):D},i.startAngle=function(n){return arguments.length?(v=typeof n=="function"?n:z(+n),i):v},i.endAngle=function(n){return arguments.length?(A=typeof n=="function"?n:z(+n),i):A},i.padAngle=function(n){return arguments.length?(B=typeof n=="function"?n:z(+n),i):B},i.context=function(n){return arguments.length?(a=n??null,i):a},i}export{hn as d};
+import{$ as ln,a0 as an,a1 as G,a2 as q,a3 as z,a4 as un,a5 as y,a6 as tn,a7 as K,a8 as _,a9 as rn,aa as o,ab as on,ac as sn,ad as fn}from"./mermaid.core-B4Rsz1_K.js";function cn(l){return l.innerRadius}function yn(l){return l.outerRadius}function gn(l){return l.startAngle}function dn(l){return l.endAngle}function mn(l){return l&&l.padAngle}function pn(l,h,I,D,v,A,B,a){var O=I-l,i=D-h,n=B-v,d=a-A,u=d*O-n*i;if(!(u*u<y))return u=(n*(h-A)-d*(l-v))/u,[l+u*O,h+u*i]}function V(l,h,I,D,v,A,B){var a=l-I,O=h-D,i=(B?A:-A)/K(a*a+O*O),n=i*O,d=-i*a,u=l+n,s=h+d,f=I+n,c=D+d,C=(u+f)/2,t=(s+c)/2,m=f-u,g=c-s,R=m*m+g*g,T=v-A,P=u*c-f*s,S=(g<0?-1:1)*K(on(0,T*T*R-P*P)),$=(P*g-m*S)/R,j=(-P*m-g*S)/R,w=(P*g+m*S)/R,p=(-P*m+g*S)/R,x=$-C,e=j-t,r=w-C,F=p-t;return x*x+e*e>r*r+F*F&&($=w,j=p),{cx:$,cy:j,x01:-n,y01:-d,x11:$*(v/T-1),y11:j*(v/T-1)}}function hn(){var l=cn,h=yn,I=z(0),D=null,v=gn,A=dn,B=mn,a=null,O=ln(i);function i(){var n,d,u=+l.apply(this,arguments),s=+h.apply(this,arguments),f=v.apply(this,arguments)-un,c=A.apply(this,arguments)-un,C=rn(c-f),t=c>f;if(a||(a=n=O()),s<u&&(d=s,s=u,u=d),!(s>y))a.moveTo(0,0);else if(C>tn-y)a.moveTo(s*G(f),s*q(f)),a.arc(0,0,s,f,c,!t),u>y&&(a.moveTo(u*G(c),u*q(c)),a.arc(0,0,u,c,f,t));else{var m=f,g=c,R=f,T=c,P=C,S=C,$=B.apply(this,arguments)/2,j=$>y&&(D?+D.apply(this,arguments):K(u*u+s*s)),w=_(rn(s-u)/2,+I.apply(this,arguments)),p=w,x=w,e,r;if(j>y){var F=sn(j/u*q($)),L=sn(j/s*q($));(P-=F*2)>y?(F*=t?1:-1,R+=F,T-=F):(P=0,R=T=(f+c)/2),(S-=L*2)>y?(L*=t?1:-1,m+=L,g-=L):(S=0,m=g=(f+c)/2)}var H=s*G(m),J=s*q(m),M=u*G(T),N=u*q(T);if(w>y){var Q=s*G(g),U=s*q(g),W=u*G(R),X=u*q(R),E;if(C<an)if(E=pn(H,J,W,X,Q,U,M,N)){var Y=H-E[0],Z=J-E[1],b=Q-E[0],k=U-E[1],nn=1/q(fn((Y*b+Z*k)/(K(Y*Y+Z*Z)*K(b*b+k*k)))/2),en=K(E[0]*E[0]+E[1]*E[1]);p=_(w,(u-en)/(nn-1)),x=_(w,(s-en)/(nn+1))}else p=x=0}S>y?x>y?(e=V(W,X,H,J,s,x,t),r=V(Q,U,M,N,s,x,t),a.moveTo(e.cx+e.x01,e.cy+e.y01),x<w?a.arc(e.cx,e.cy,x,o(e.y01,e.x01),o(r.y01,r.x01),!t):(a.arc(e.cx,e.cy,x,o(e.y01,e.x01),o(e.y11,e.x11),!t),a.arc(0,0,s,o(e.cy+e.y11,e.cx+e.x11),o(r.cy+r.y11,r.cx+r.x11),!t),a.arc(r.cx,r.cy,x,o(r.y11,r.x11),o(r.y01,r.x01),!t))):(a.moveTo(H,J),a.arc(0,0,s,m,g,!t)):a.moveTo(H,J),!(u>y)||!(P>y)?a.lineTo(M,N):p>y?(e=V(M,N,Q,U,u,-p,t),r=V(H,J,W,X,u,-p,t),a.lineTo(e.cx+e.x01,e.cy+e.y01),p<w?a.arc(e.cx,e.cy,p,o(e.y01,e.x01),o(r.y01,r.x01),!t):(a.arc(e.cx,e.cy,p,o(e.y01,e.x01),o(e.y11,e.x11),!t),a.arc(0,0,u,o(e.cy+e.y11,e.cx+e.x11),o(r.cy+r.y11,r.cx+r.x11),t),a.arc(r.cx,r.cy,p,o(r.y11,r.x11),o(r.y01,r.x01),!t))):a.arc(0,0,u,T,R,t)}if(a.closePath(),n)return a=null,n+""||null}return i.centroid=function(){var n=(+l.apply(this,arguments)+ +h.apply(this,arguments))/2,d=(+v.apply(this,arguments)+ +A.apply(this,arguments))/2-an/2;return[G(d)*n,q(d)*n]},i.innerRadius=function(n){return arguments.length?(l=typeof n=="function"?n:z(+n),i):l},i.outerRadius=function(n){return arguments.length?(h=typeof n=="function"?n:z(+n),i):h},i.cornerRadius=function(n){return arguments.length?(I=typeof n=="function"?n:z(+n),i):I},i.padRadius=function(n){return arguments.length?(D=n==null?null:typeof n=="function"?n:z(+n),i):D},i.startAngle=function(n){return arguments.length?(v=typeof n=="function"?n:z(+n),i):v},i.endAngle=function(n){return arguments.length?(A=typeof n=="function"?n:z(+n),i):A},i.padAngle=function(n){return arguments.length?(B=typeof n=="function"?n:z(+n),i):B},i.context=function(n){return arguments.length?(a=n??null,i):a},i}export{hn as d};