File size: 1,228 Bytes
5afdc95
 
ee0e1b4
 
 
5afdc95
ee0e1b4
5afdc95
 
ee0e1b4
 
 
d31595e
ee0e1b4
 
 
 
 
 
 
d31595e
5afdc95
 
ee0e1b4
 
 
 
 
 
 
 
 
 
d31595e
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
---
title: HistAgent Chat
emoji: 🧬
colorFrom: indigo
colorTo: red
sdk: gradio
sdk_version: 5.44.1
app_file: app.py
pinned: false
license: other
models:
  - Qwen/Qwen3-8B
  - Qwen/Qwen3-Embedding-8B
datasets:
  - wli13/HistAgent-data
tags:
  - histopathology
  - spatial-transcriptomics
  - biology
  - chat
  - retrieval
---

# HistAgent Chat

HistAgent Chat answers questions about selected tissue spots using ranked genes,
cell-state estimates, functional programs and spatial context contained in a
structured evidence card.

The online demonstration uses Qwen3-8B with the HistAgent evidence-card schema,
system policy and evidence constraints. The interface therefore describes the
language-model component as **adapted Qwen3-8B**. It does not expose
chain-of-thought and is intended for research use.

The Atlas Explorer tab searches a balanced public index of 100,000 measured ST
spots using 4,096-dimensional Qwen3-Embedding-8B vectors and cosine similarity.
The demonstration index and checksum manifest are published in the
[HistAgent public evidence-bank release](https://github.com/zipging/HistAgent/releases/tag/public-evidence-bank-v1).
The manuscript analyses use the complete 2.23-million-spot evidence bank.