fix(chat/demo): remove memory search — bubble chat is a public guide bot only, no db access

fix(agent): loosen memory retrieval threshold to 0.70 and show raw_text for CSV data

fix(chat): lower similarity threshold to 0.70 and use raw_text for CSV-ingested data

feat: add /chat/demo endpoint to power KaalChat widget with memory

feat(ui): populate API key input with saved key from localStorage to reduce user confusion

fix: move CREATE INDEX in vector_store after ALTER TABLE migration

fix: move CREATE INDEX statements after ALTER TABLE migration in memory_store schema to prevent UndefinedColumnError on startup

chore: add smriti benchmark runner and eval service

refactor: centralise all magic numbers into config.py driven by env vars

feat: bi-temporal supersession, scope isolation, configurable threshold, benchmark adapter

feat: add article 'How to Give Any AI Agent Long-Term Memory in 3 API Calls'

docs: clarify dual-deployment architecture (Vercel UI + Hugging Face API)

feat: add SEO-optimized article 'The End of AI Amnesia'

chore: update og:url to vercel domain to prevent SEO backlink fragmentation

fix: update Keys.tsx apiCall to use JSON body and Turnstile token for key generation

fix: billing body binding (Pydantic model), surface real error in WelcomeScreen

security: add Turnstile captcha + IP rate limiting + email dedup to key generation

fix: add security headers to next.config (X-Frame-Options, CSP, CORS)

UX: Collapse onboarding from 5 screens to 3, remove forced chat popup, fix all Kaal copy to Smriti

UI: Complete rebrand to Smriti by Kaal the Absolute

fix: add missing KaalChat.tsx to git and fix corrupted .gitignore null-byte wav entry

feat: add KaalChat floating AI widget — consent gate, TTS toggle, template feature cards, typing indicator, 5-msg rate limit display

feat: store Amber voice as base64 py constant to avoid binary in git, fix speaker_wav references in TTS route

feat: add XTTSv2 TTS endpoint + public demo chat with Amber voice, template-first responses, and Chronos memory consent storage

UI: remove Builder and Scale tiers from dashboard Keys page

UI: remove Builder and Scale tiers from dashboard Billing page

fix: restore missing Hexagon import in Sidebar to fix Vercel build after reset

Logo: remove tagline and HourglassIcon import, clean seal + Kaal wordmark only

Rebrand to Kaal — rename from Chronos OS, add Logo component with serif seal design, swap all Hexagon icons for Logo

Update UI and README to reflect new GPT OSS 120B and Llama 3.1 architecture (removed GLM 4.7 references)

Add connected tools list to the Connect UI so users can see their registered tools

Switch agent to GPT OSS 120B — officially recommended Cerebras model for tool-calling agents

Fix Cerebras model ID to llama3.1-70b as 3.3 was deprecated

Switch primary heavy agent model from GLM 4.7 to blazing fast Llama 3.3 70B on Cerebras

Fix router hanging on rate limits by setting max_retries=0 to force instant fallback

Fix pre-launch blocking issues and agent multi-turn reasoning_content error

Replace emoji with Lucide icons in guide; add info tooltips to sidebar nav

Add interactive Getting Started guide to Overview dashboard

Hide paid tiers — show only free Explorer plan until payment is integrated

Update pricing to premium B2B SaaS tiers: /mo Builder and /mo Scale

Fix AgentRunResponse validation error by parsing GLM/DeepSeek reasoning block lists into strings

Update heavy pipeline to use zai-glm-4.7 and update dashboard display

Revert to Llama 3.3 70B to fix API crash with DeepSeek V3.2

Update AI Engine display on dashboard to DeepSeek V3.2

Update DeepSeek model to use explicit cerebras endpoint string

Update heavy pipeline to use DeepSeek V3.2 for optimal reasoning and MoE efficiency

Update heavy pipeline to use GPT-OSS 20B for high efficiency and low token burn

Revert heavy pipeline to Qwen 3 235B on Cerebras

Prioritize Cerebras API over Groq in LLM router for higher limits and speed

Update demo box in Hero section to include 'WHAT'