LIMIT 1 on expr table, minor tool fixes

Fixes:

• LLM master prompt - LIMIT 1 statement for any queries on expr table. This will reduce occurrence of "input tokens exceeded context memory" errors.

• gene_annotation table - fix column names in tools.py

• retrieve gene list tool - fix file name of protein coding gene list in tools.py

• gene id lookup tool - fix gene annotation file name in tools.py

Updates:

• Troubleshooting page on wiki

Improved documentation, Readme

New:

• A wiki page has been created for the repository see wiki here
• setup.md: instructions to build

Improvements:

• readme: Updated demo videos and thumbnail
• database: change index columns for hgnc_nomenclature to: gene_symbol, alias_symbol, gene_stable_id
• main.py: erase memory: remind user to reload page for function to take effect
• Invisible SEO heading text on index.html - 150 characters total

Removed:

• readme: moved setup steps to setup.md
• Doc files: rm all doc files, these go to another repo

Fixes:

• Doc files: stand_alone_survival: ignore mmstatus1/2/3 columns in
• Doc files: gatk_baf: remove duplicate public_id column
• Refdata: rename columns like 'start', 'end', and 'Chromosome/scaffold name' to 'start_bp', 'end_bp' and 'chromosome'
• Refdata: remove spaces with underscore in column names of gene_annotation
• tools.py: fix gene annotation file name

Updated landing page & bug fixes

New features:

• Add description to index.html
• RetrieveGeneListTool - return list of IGH, housekeeping, or proteincoding genes
• SurvivalDataTool - retrieve right-censored OS/PFS

Bug fixes

• Dockerfile: COPY of main.py into container
• explicitly state censor flag 1=event 0=censor in docs
• close missing parenthesis in docs
• tools.py: file name of Cox PH covariates data file

Misc

• index.html footer margin
• main.py: rm print headers
• moved tool data files to /refdata
• rm tool data files from /src/result

Handle graph.stream() errors, tool bug fixes

Bug fixes to v1.5.0

• fix wrong filename for pre-generated cox ph covariates
• handle graph.stream exceptions by returning the message to user
  • e.g. openai.exception when input token exceeded
• token usage defaults to 0 instead of -1
• fixes to readme such as broken video links, PostgresDB version

Updates on ready-to-use application:
Uptime is much better using AWS apprunner + lightsail DB

AWS AppRunner + Lightsail Postgres18 DB

v1.5.0

Additions ✨

• Check usage button - reports running token usage and running cost
• Mobile version - expandable dropdown menu for command buttons
• Environment variable module (variables.py) to check for environment variables and load them into python

Changes 🌪️

• Changes of env variables:
• SECRET_KEY ➡️ JWT_SECRET_KEY,
• remove MAIL_PORT, MAIL_FROM,
• Added DBHOSTNAME, DBUSERNAME, DBPASSWORD and removed COMMPASS_DSN, COMMPASS_DB_URI
• refer to variables.py

Removed 🗑️

• Scripts and data files related to CoMMpass database - these are DevOps related and not needed for the MyeGPT client
• create_database.py
• create_vectorstore.py
• 📂docs
• 📂omicdata
• 📂clindata
• 📂refdata

Known issues 🚨

• App unresponsive on first login of new account

Updates on ready-to-use version 📱

• Change DB from lightsail instance to lightsail database
• Database no longer has ultra-long gene expression tables - uses just expr and expr_metadata
• Hosting on AWS App runner with 2vCPU and 4GB

Link to ready-to-use app 🔗:

https://c5tuv9quha.ap-southeast-1.awsapprunner.com/

Update to langchain==1.2.7

Code changes:

• Refactored code to work with latest versions of:
• langchain == 1.2.7, langgraph == 1.0.7, langchain-openai == 1.1.7, langchain-postgres == 0.0.16
• Assign ReAct agent logic to new module executor.py
• executor.py is based off LangGraph.prebuilt.create_chat_agent_executor

UI Additions:

• "Fix History" button - this deletes tool calls without corresponding tool results
• If "Fix History" is triggered but no messages exist or no error-causing messages exist, system message is
• Send Message button disabled when model init, "Fix History" or "Erase Memory" is occurring

Release live url:
https://github-release-v1-4-0---app-qm3lx6wcqq-uc.a.run.app/

Security patches

Fixes:

• validate headers each time to prevent cross-origin access
• Prevent log in without email verification
• Prevent access to /api/init or /api/ask without API bypass token
• User name and email always displayed as "admin"

Additions:

• tool for retrieving gene median absolute deviation
• tool for retrieving gene CoxPH summary statistics

Updates:

• minor UI changes e.g. command buttons

Link to live version:

• https://github-release-v-1-3-3---app-qm3lx6wcqq-uc.a.run.app/

The following uptime checks are implemented

Improved security

Continuation of v1.3.0 - chat experience is optimized for GPT-5-mini, GPT-5.2, and GPT-4-turbo models

Better exception handling - reduce incidents of internal server errors

Dedicated API_BYPASS_TOKEN env var to allow a static bearer token to authenticate into all endpoints - this allows Gcloud service agent testing

New endpoint:
POST /ready: to test app readiness. checks token validation, CoMMpass database, user database, and Langchain vectorstore. returns 200 or 500.

Existing endpoints:
GET /
GET /verify
POST /app
POST /token (not exposed)
POST /api/init
POST /api/ask
POST /api/register
DELETE /api/erase_memory
DELETE /api/delete_account

Optimized for GPT-5-mini

Current as of 12 Jan 2026
Changes:

• Optimized specifically for GPT-5 mini models
• AI messages appear as agent, tool calls, and tool results
• Code, tool calls, and tool results are wrapped in scrollable div's
• No more show/hide trace - replaced with thinking messages
• Document vectorstore is fragmented for better retrieval

Fixes:

• account creation issues: server crashes when existing email is entered
• graphs not being plotted or displayed

Support for Gemini AI

Fixes

1. email verification link not working
2. Display text embeddings model name upon double clicking of user icon (through the embed-model-provider-display element)

Additions

1. Support Google Gemini AI LLM models
2. Support Google Gemini Text embedding service
3. Support for authentication-based evaluation through eval.py