Activity

fix: Compare products entity_id format + cart message rendering crash (#106)

feat: Make tool guard error messages specific for LLM self-correction (#53) (#92)

feat: Deduplicate tool calls in _should_continue to short-circuit identical retries (#52) (#91)

feat: DST context management — fix stale price filters persisting across queries (#88) (#90)

feat: Add reflection node to prevent LLM retry loops on blocked tool calls (#51) (#89)

security: Remove SQL tools, harden PII redaction, add cart/order session limits (#55, #56, #57) (#87)

fix: Revert pixel INSERT to loop within single transaction (#86)

fix: Batch backend cleanup — pixel response, ORM models, structlog, auth logging, Redis startup (#79-83) (#85)

fix: Show suggestion pills only on product responses (#78) (#84)

perf: Backend performance fixes — async Redis, auth caching, N+1 fix, batch inserts (#41, #43, #44, #45) (#77)

feat: Add context window telemetry and compaction quality metrics (#65) (#71)

refactor: Replace regex action detection with tool-call inspection (#61) (#70)

fix: Harden MEM-4 price parsing and error fallback (#61) (#68)

feat: Enrich tool result summaries with price ranges and user actions (#61) (#67)

feat: Add product search history ring buffer for back-references (#64) (#66)

Add dedicated app pages, activity analytics dashboard, and CI pipeline

feat: LangGraph Guardrails System for StorefrontAgent v4 (#35)

fix: Add HMAC auth headers directly in bridge fetch instead of relying on interceptor (#25)

fix: HMAC auth for prompt chips on production (#24)

feat: LLM-generated product prompt chips for chat engagement (#23)

Redesign apps page to horizontal list layout matching CV page structure

Add GateKpr and ShopExtract to apps page, redesign layout

Add Legal and Responsible Use section to README

Use official test stores in notebooks and README, add Testing section

Remove hardcoded store names from notebooks, use input() for URLs, fix 3.10 compat in monitoring cell

Fix asyncio.timeout (3.11+) to asyncio.wait_for (3.10+) in platform detection

Fix Python 3.10 CI: add future annotations, fix mock transports, pass client directly

Fix Python 3.10 compatibility: StrEnum and datetime.UTC

Fix Python 3.10 compatibility: replace datetime.UTC with timezone.utc

Rewrite demo notebook: clean cells, no fallbacks, sample data from GitHub

Bump to v0.1.2 with updated PyPI documentation

Update README with Colab badge, universal scraping docs, env vars, interactive demo section

Add Colab/Kaggle-compatible demo notebook

Fix tests badge URL: test.yml not tests.yml

Simplify installation to single pip install, bump to v0.1.1

Update README with full LLM provider documentation

Add support for all LLM providers: Ollama, Mistral, Gemini, DeepSeek, Cohere, and 15+ more

Wire LLM tier into extraction fallback chain with configurable API key and model

Add interactive demo notebook with all feature examples

Add comprehensive README, documentation, and example scripts

Add GitHub Actions CI, PyPI publish workflow, and build config

Add comprehensive unit tests for all modules

Fix silent failures and improve error handling

Add marketplace validation, multi-format export, and CLI

Add catalog analysis and competitive intelligence

Add compare, match, and monitor features

Initial package scaffold with core extraction engine

fix: Remove search_shop_catalog from agent tools — text-only output breaks product cards

fix: Prioritize discover_products_semantic over search_shop_catalog in JIT rules

fix: Intercept and fix LLM-provided shop_domain in tool calls

fix: Pass shop_domain to JIT tool docs for search_shop_policies

fix: Strip bare bracket references [1][2][3] from agent text

fix: Include image_url in DB-to-Qdrant sync path

fix: Use correct X-Internal-API-Key header for auto-ingestion on install

Qdrant RF + Preference Injection + Rich Page Context (#19)

Widget UX fixes + Qdrant search improvements (#18)

Qdrant search optimization + UX improvements

fix: Widget UI/UX improvements and security hardening (#16)

fix: Mobile widget viewport and screen reader announcement visibility (#15)

fix: Fallback to raw_data for image_url in SQL product tool

fix: Always compact large tool results in JIT context

chore: Reorganize repository structure

feat: JIT Context Optimization for StorefrontAgent (#14)

fix: Product cards not displaying on second query in same session

perf: Replace MutationObserver with history API for SPA navigation detection

fix: Product card persistence after page refresh

fix: Restore product cards from localStorage on page refresh

debug: Add comprehensive logging to product extraction function

fix: Clarify prompt to always call discover_products_semantic for new product searches

fix: Add widget HMAC authentication headers to iframe API client

feat: Remove Channels tab from merchant dashboard

fix: Add missing useWhatsAppAutomation hook

fix: Update WhatsApp routes to use react-router imports

feat: Add circuit breaker for WhatsApp feature

fix: Remove business_management scope from WhatsApp OAuth

fix: Open WhatsApp OAuth in new window instead of iframe

fix: Remove STATE_PREFIX export after OAuth database migration

fix: Use get_async_session for OAuth state management

fix: Use database for OAuth state instead of Redis

fix: Correct WhatsApp OAuth endpoint URL to match backend

feat: Implement WhatsApp Business API Phase 3

fix: Remove broken migration 026, use corrected UUID types

fix: WhatsApp Phase 2 code review fixes

feat: Add WhatsApp Business API Phase 2 - OAuth, Phone Verification, Analytics

fix: Address code review issues for WhatsApp integration

feat: Add WhatsApp Business API integration (Phase 1)

fix: Add asyncio.wait_for timeouts to checkpoint operations

fix: Extract products only from current turn, not previous turns

fix: Stop product_gallery from showing on non-product queries

debug: Add SSE logging to diagnose product grid clearing issue

fix: Product grid UI bugs - stale grid, duplication, page context

docs: Update AAMS dashboard URL and add Playwright usage instructions

fix: Replace hardcoded staging URLs with dynamic frontend variable

fix: Remove debug console.log statements from widget

fix: Close chat window when switching to minimized mode

fix: Safari state management with timestamp-based transition guard

fix: Complete remaining 5-week plan items

fix: Simplify widget state acknowledgment to prevent desync

fix: Comprehensive state management overhaul for widget open/close

fix: Add widget-bridge state synchronization via acknowledgment

fix: Prevent desktop Safari from being detected as mobile

docs: Add staging test store URL and password to CLAUDE.md

fix: Safari compatibility - prevent header element collapse

fix: Add Safari 13-14 flexbox gap fallbacks for header elements

fix: Add Safari 13-14 browser compatibility

fix: Add transition guard to prevent widget state desync on rapid clicking

feat: Week 5 widget modernization - tokens, dark mode, accessibility, performance

feat(widget): Complete modernization - remove unused tokens.css, standardize TOKENS usage

fix(widget): Add missing checkoutPanel token to design-tokens

fix(widget): Add variant validation to handleBuyNow function

feat(widget): Week 4 improvements - bug fix, accessibility, architecture & bundle optimization

feat(widget): Complete Week 3 design system consolidation & testing

feat(widget): Week 3 accessibility, performance & design system improvements

feat(widget): Complete Week 2 UI modernization with CSS class migration

fix: Restore displayMode to docked when opening from minimized state

docs: Add comprehensive UI refactoring plan and best practices

fix: Preserve image_url in product webhook updates

feat: Add on_sale_only filter to discover_products_semantic

fix: Prevent widget layout crush on reopen in iframe mode

feat: UI modernization + cart fix for agentic widget

fix: Sync product title and description to Qdrant on webhook updates

fix: Eliminate cart.js 404 spam and false isMobile in iframe context

fix: Stop cart.js 404 spam in iframe and fix isMobile false positive

feat: Convert widget to Rufus-style docked side panel with 4 iframe sizing modes

feat: Forward page context to agent and add OpenAI EU base URL support

fix: Make widget clickable by replacing pointer-events:none with dynamic iframe resizing

fix: Keep staging frontend always running to prevent widget 503s

fix: Add X-Shop-Domain header to widget API client and harden prompt config import

fix: Increase staging frontend memory to 512MB for stability

fix: Add widget files to public folder and add debugging logs to brand config

feat: Add iframe-based persistent widget architecture (SIG-72, SIG-73)

feat: Add product comparison mode and quick actions bar

feat: Add Rufus-style docked sidebar, page context detection, and proactive engagement

fix: Remove sslmode from connection string to fix SSL config conflict

fix: Add SSL config for Supabase PostgreSQL connection in shop registration

feat: Add Mistral LLM provider support for lower latency

fix: Add timeout protection and logging to identify connection hang

chore: Add logging for OpenAI model diagnosis

chore: Add logging for LLM call timing diagnosis

fix: Add asyncio.timeout protection around LLM streaming loop

docs: Add Shopify theme extension deploy instructions and fix Qdrant addresses

infra: Harden Qdrant production infrastructure

fix: Update Qdrant staging config to match production for 6PN connectivity

fix: Ensure tool_progress SSE events display in widget UI

chore: Add debug logging for tool_progress SSE events

fix: Update theme extension with latest widget build (tool_progress support)

fix: Use flattened tool_progress fields from SSE chunks

feat: Add tool_progress streaming support to frontend widget

fix: Enable StorefrontAgent (v4) on staging via AGENT_VERSION env var

fix: Use dispatch_custom_event for tool progress streaming

fix: Add null-safe handling for get_stream_writer in all tools

feat: Add custom streaming events with AI sales personality

Fix Bloomers date range to Jan 2025 - Jan 2026

Update CV and homepage - no longer at Bloomers

Fix scheduled deploys not refreshing activity commits

Add live demo link to Sigma AI page

Enable StorefrontAgent (v4) in production (#13)

fix: Fix all ruff linting errors in backend (#12)

fix: Simplify production smoke tests to only check critical endpoints (#11)

Staging (#10)

fix: Enable Qdrant Flycast auto-start on staging

fix: Improve migration system for staging/production isolation

feat: Add staging environment infrastructure

feat: Add stress testing suite for widget chat API

feat: Add LangSmith production tracing to storefront agent

Add AAMS webhook integration for uninstall and subscriptions

refactor: Change api_keys to admin_api_keys table

feat: Add api_keys table migration for AAMS

Update Shopify link styling to match website theme

Add Shopify App Store link to Sigma AI page

fix: use _build_graph check instead of graph property to avoid RuntimeError

fix: add ensure_initialized() call before accessing StorefrontAgent graph

feat: Add LangSmith native evaluators and improve agent tools

feat: Add 100-query LangSmith evaluation dataset and scripts

feat: Add LangSmith evaluation skill with full tracing

Add daily scheduled rebuild for activity page

docs: Add git commit guidelines - no watermarks

fix: Prevent duplicate priorities for featured products

feat: Add get_featured_products tool for merchant's hero products

fix: Yield text before products for proper streaming UX

fix: Generate meaningful fallback responses for product galleries

fix: Auto-refresh cart panel UI after remove/update operations

fix: Use Shopify Ajax cart API for remove/update cart operations

fix: Address Shopify App Store review issues (2.1.2)

fix: Shopify App Store review issues - cart errors, AI hallucination, prompt save

Add research sources section to GEO page

Add GEO audit feature with detailed report UI

Add deep crawl feature to GEO optimizer (up to 10 pages)

Add dedicated pages for DiMedi and ORY Berlin apps

Add DiMedi and ORY Berlin to apps page

chore: Clean up repository and update documentation

chore: Remove unused test artifacts and evaluation files

feat: Secure order lookup with triple verification via Shopify Admin API

feat: Add UX improvements - avatar gallery, color presets, language selection, skills toggles

feat: Add language and skill suggestions to auto-generate agent

feat: Add enabled_skills support across frontend and backend

feat: Add AI skills toggles for assistant capabilities

feat: Add language selection to AI assistant configuration

feat: Add avatar gallery and color presets to widget design studio

Revert to API for showcase data

Use static JSON for showcase data

Add llm.txt for LLM crawlers

Update GEO-optimized LLM content on homepage

Add showcase section to geo page

Remove emojis and special characters from GEO page

Fix GEO Optimizer polling to use correct API endpoints

Add interactive GEO Optimizer tool to apps section

Add GEO Optimizer page with navigation

Fix blog post title capitalization for consistency

Enlarge navbar menu and update email to hello@umerkhan.eu

Add Google Search Console verification meta tag

fix: Theme toggle now works on first click with event delegation

feat: Redesign website with consistent side-label layout and Three.js particles

fix: Pass GITHUB_TOKEN as build argument for Fly.io deployment

Redesign website with modern minimal UI and add activity page

fix: Clear ALL shop caches on uninstall (both shop_helpers and shop_resolver)

fix: Resolve correct shop_id after app reinstall

fix: Resolve app loading issue on fresh install

fix: Prevent Fly.io internal port 8080 from leaking into URLs

fix: Add data-astro-reload to prevent View Transitions port bug

feat: Add /apps page with app listing, update navbar

feat: Add Sigma AI to navbar

fix: Use absolute URLs for sigma-ai page links

Merge article-page-integration: Add Sigma AI pages

feat: Add Sigma AI landing page, privacy policy, and terms of service

feat: Add freemium usage tracking with 1000 sessions/month limit

fix: Hide internal Linear URL from merchants

fix: Ensure agent includes actual ticket URL in responses

docs: Add detailed installation and app embed troubleshooting

chore: Update documentation tool description for merchant context

security: Prevent code/data type exposure to merchants

fix: Add markdown rendering for support chat responses

Add merchant support chat widget with Linear integration

Add Fly.io deployment and homepage experience section

fix: Prevent hallucination and improve conversation memory

Fix: Prevent generic greetings on brand-specific queries

Add prompt guidance for show-all, catalog overview, and long conversations

fix: Clear shown_products state on empty search results

fix: Prevent Dawn theme CSS from hiding widget container

Fix sessions/analytics APIs and add dashboard metrics

feat: Add collections webhooks for real-time sync

fix: Pass BACKEND_API_URL explicitly to npm build process

fix: Expose BACKEND_API_URL to client bundle at build time

fix: Use correct field name 'body' for Article in Shopify GraphQL

feat: Add read_content scope for Shopify Pages API

fix: Remove invalid query argument from Shopify Pages GraphQL

fix: ShopCredentialManager to fetch access token from Supabase Vault

feat: Add Auto-Generate prompt configuration with LangGraph agent

feat: Professional agent responses with horizontal product scroll

fix: Cart synchronization with Shopify theme cart

feat: Redesign checkout panel with variant selectors and fix double-bubble

feat: Redesign Agentic widget to match OpenAI Shopping Demo

fix: SQL parameter mismatch in widget design PATCH endpoint

feat: Add Agentic widget with widget type selector in Design Studio

feat: Integrate theme settings into widget frontend

feat: Add Widget Design Studio for customizable chat widget themes

Fix redundant product cards on follow-up questions

fix: add SSE line buffering for large product_gallery events

fix: SQL tools now return JSON-serializable product data for product cards

feat: add full variant data to product search responses

feat: add graceful cart operations with ordinal references and button sync

fix: restore entity_id extraction for payload

fix: centralize Qdrant point ID generation to prevent duplicates

fix: use Shopify category taxonomy instead of productType for filtering

fix: use full GID for Qdrant point ID to enable proper upserts

fix: remove invalid 'close' SSE event type causing pydantic validation error

feat: add comprehensive Qdrant payload filtering for brand, price, stock, and tags

feat: fix Qdrant 6PN connectivity and enhance learning system

feat: add compliance webhooks and preference retrieval service

feat: add continuous learning system with preference extraction

feat: register customer auth routes and add API tests

feat: add Customer Accounts OAuth and order tracking tools

Enhance merchant prompt and session persistence

fix: per-turn tool counter, remove end_on, strengthen cart prompt

feat: add StorefrontAgent v4, widget-dev, ingestion fixes

fix: correct cart prompt params and add image_url to Qdrant

docs: add Phase 2 completion documentation

feat: add StorefrontAgent v4 - single agent architecture

fix: ensure semantic search for price queries + add logging

fix: use semantic search for price queries

fix: correct tool name in ProductReaderAgent prompts

feat: auto-ingest products and collections on app install

fix: use SAVEPOINT for PostgreSQL transaction isolation

feat: auto-sync products to Qdrant on ingestion

fix: remove typing chunk that widget doesn't handle

fix: resolve chat widget 422 errors with session ID compatibility

fix: enable Shopify managed installation for token exchange

Fix embedded Shopify auth and add backend API integration

fix: add Polaris AppProvider for SSR context

fix: replace RouteErrorBoundary with SimpleErrorBoundary for SSR compatibility

fix: add /health endpoint to fix 503 errors

feat: add comprehensive error boundary system

fix: skip authenticate.admin() for embedded apps with session tokens

feat: migrate to file-based routing with flatRoutes

fix: remove duplicate AppProvider from root.tsx

fix: use route() instead of layout() for proper path handling

fix: prevent login form rendering for embedded apps

fix: properly handle embedded auth in login route

fix: remove auth/login route blocking embedded app

fix: resolve auth/login 500 error by restoring login routes

feat: enable data ingestion route with all dependencies

fix: resolve 500 error by disabling unstable_newEmbeddedAuthStrategy

fix: disable analytics route until backend API is ready

fix: register routes in routes.ts and add analytics page

fix: restore missing app routes accidentally removed in Fly.io migration

feat: migrate to Fly.io deployment with PostgreSQL

feat: add ChatWidgetAgent integration example for merchant profiles

fix: add start-dev.sh and enhance CORS for Cloudflare tunnels

chore: add Shopify app configuration and setup instructions

feat: complete Phase 5 UI and comprehensive documentation

fix: resolve test failures in merchant config test suite

docs: finalize Phase 5 guide and update status to 75% complete

test: add comprehensive test suites for Phase 6

docs: update implementation status - Phase 4 complete (60%)

feat: implement Phase 4 - Prompt Configuration API

fix: migration bug and add async database pool

docs: update implementation status - Phase 3 complete (45% done)

feat: add ChatWidgetAgent integration example for merchant profiles

feat: create SalesAgentComposer with merchant customization support

docs: add implementation status tracker for merchant prompts feature

feat: implement MerchantConfigService with Redis caching and validation

feat: add database migration and ORM models for merchant prompt customization

Merge feature/deep-agents-migration: Docker containerization and production setup

feat: Complete Docker containerization and production setup

feat: implement local SQL LLM with comprehensive testing

feat: fix product prices in Qdrant + LangChain SQL Agent analysis

fix: implement proper message trimming using middleware

feat: implement message trimming to prevent token overflow

refactor: remove MCP layer, use direct Qdrant for semantic search

fix: MCP client connection issues and add SQL fallback

Implement Phase 3: Advanced Features - Subagents and Long-Term Memory

Implement Phase 2: Deep Agents Core Migration

Phase 1 COMPLETE: Deep Agents POC - All tests PASSED

Add comprehensive Deep Agents migration documentation

Pre-Deep Agents migration checkpoint - hierarchical multi-agent architecture

Update graphs.py and langgraph.json with new orchestrator naming

Archive legacy agents and rename master_orchestrator_v3 to orchestrator

Update MetafieldReaderAgent prompt to prioritize product discovery tools

Add product discovery tools for MetafieldReaderAgent fuzzy→exact identifier conversion

Add load_dotenv() to llm_config.py to automatically load environment variables

Replace hardcoded ChatOpenAI with get_llm() in all agents - full Ollama support

Replace hardcoded ChatOpenAI with centralized get_llm() supporting Ollama

Add centralized LLM configuration supporting Ollama, OpenAI, and Anthropic

Update product read tools to use new price column

Fix checkpointer pattern following LangChain Academy best practices

CRITICAL FIX: Restore checkpointers for conversation memory

Add comprehensive conversation memory documentation

Add startup script for LangGraph dev with PostgreSQL persistence

Fix LangGraph dev compatibility while preserving conversation memory

CRITICAL FIX: Restore conversation memory by enabling checkpointers

Fix PostgreSQL connection issues by disabling SSL for local database

Enforce mandatory tool usage in reader agents

Integrate ReadSupervisor into MasterOrchestratorV3

Add safe state access for LangGraph Studio compatibility

Fix message content handling in MasterOrchestratorV3 and ReadSupervisor

Add comprehensive test suite for MasterOrchestratorV3

Fix checkpointer initialization for LangGraph API compatibility

Fix checkpointer initialization order in MasterOrchestratorV3

Fix relative imports in graphs.py for LangGraph Studio compatibility

Configure new multi-agent system for LangGraph Studio

Phase 3: MetafieldReaderAgent with 45 tools and dynamic selection

Phase 2: ReadSupervisor + ProductReaderAgent with dynamic tool selection

Phase 1: Foundation - MasterOrchestrator V3 with LLM routing and Pydantic validation

Add comprehensive multi-tenant architecture to implementation plan

Add comprehensive multi-agent implementation plan

Add LibrarianAgent with 45 specialized metafield tools

Refactor LibrarianAgent tools into modular structure

Add LibrarianAgent Phase 2: Complete Qdrant tools implementation

Add LibrarianAgent Phase 1: Database migration and SQL tools

Fix KeyError user_query - use local variable and remove from state dict

Fix TypedDict to use total=False for LangGraph Studio compatibility

Fix orchestrator to properly invoke ProductAgent V2 as subgraph

Fix AttributeError list object has no attribute lower

Handle multiple message formats in orchestrator (Studio dict/object/string)

Fix product_agent_node to properly handle compressed_instruction fallbacks

Fix Master Orchestrator to accept message-based input from LangGraph Studio

Add Master Orchestrator documentation and quick API test

Fix LangGraph API checkpointer error - remove custom MemorySaver

Implement Master Orchestrator with supervisor pattern for multi-agent routing

Fix token overflow and context length errors in ProductAgent V2

Implement ProductAgent V2 with LangGraph architecture