fix: avoid hugging face build-time model download

fix: stabilize celery reconnect handling

feat: add dual eval review and selectable URL crawling

feat: allow cancelling active ingestion

feat: add safe operator URL ingestion

feat: add hosted gemma review backend support

fix: expose markdown uploads in the UI

feat: add markdown ingestion support

Add image extraction and grounding support

Add hybrid graph substrate and deterministic code graph indexing

fix: harden gemma draft review parsing

feat: add gemma-assisted eval baseline drafts

feat: harden phase 1 retrieval quality workflow

Harden ingestion and retrieval reliability across the pipeline

fix: restore pgvector retrieval and preserve source citations

fix: harden supabase security and stabilize multi-tenant rag flows

fix: harden supabase setup and clean stale retrieval/ingestion code

fix: jwt time out

fix(ingestion): sanitize null bytes and surrogates to resolve Postgres 22P05 error

evaluation fixes eval chunkin_loss types

feat/fix: isolate eval pipeline, stabilize RAG streaming, and resolve auth flow

fix: deployment readiness — auth, naming, Dockerfile, render config

refactor: harden security, optimize deployment, and project-wide cleanup

fix:- structuring stuff

removing stuff

req change

feat: built corpus health API for proactive content gap detection

perf: implement predictive cache prefetching for 0ms follow-up latency

perf: implement sliding window TTL for semantic cache

feat: implemented PageIndex hierarchical tree search and regex query router

feat: added fast-path OCR, centroid EMA, and lexical micro-features

feat: implement active learning, F1 evaluation gate, and stateless cache

feat: implement ingestion backoff, CI build modes, and rebrand to Morpheus

fix(deploy): track missing background worker and ml dependencies

feat(deploy): prepare codebase for Railway deployment and patch JWT security

feat(rag): implement native RAPTOR indexing and fix distillation

feat: rate limiting, token budget guard, and rerank distillation loop

fix(graph): wire onGraphTabVisible() on tab switch via switchView wrapper in main.js

fix(cache): persist version key on first store — increment now goes N→N+1 not 0→1

fix: CrossEncoder sigmoid normalization and per-reranker threshold tuning

Stop tracking backend/core/intent_model.pkl

feat: semantic cache with Redis, multi-provider Groq/Gemini routing, image relevance filtering

fix: - getting groq working and adding image relevance so it works

After applying, Groq should be hit first (it's wired correctly), Gemini second if Groq fails. The generation path will work reliably now.

feat: multi-provider LLM routing, Redis query cache, and pipeline hardening

adding groq and gemini ,cuz open router alone was causing token limit reached , also removeed the ambiguity between .env and config , everything about models chosen declared lives in config now

fixed _build_llm in pipeline bypassing providerfactory from provider.py

fix — add a second parse attempt for double-encoded JSON:

T1.1. testing

chunk incement for context capture from small corpa , intent classifier fetching to supabase now , secondary to json