fix: remove PydanticOutputParser format_instructions from switchboard + planner (was making free LLMs answer the schema)

fix: synthesizer — remove PydanticOutputParser format_instructions injection (was causing LLMs to answer the schema instead of the query)

fix: dream/curiosity LLM calls, self-reflection topic extraction, daemon stale thought cleanup + error logging

fix: context_engine — dedup pending thoughts, hard cap 20, 24h age limit, fix topic extraction stopwords

fix: smart_router.py — Gemini parsing crash, in-memory rate state, provider cooldown, correct HTTP-Referer

fix: _model.py — correct HTTP-Referer URL, guard empty responses, clean docstrings

fix: update FREE_MODEL_LADDER in _model.py — replace dead OpenRouter model IDs with working ones

fix: restore all missing endpoints — daemon, memory, prompts, self, cases, intelligence, cache, traces, curation, domain; fix doubled router prefixes

fix: silence /?logs=container 404, HF_STORE_REPO dataset repo created

fix: daemon.run() is sync blocking loop — run in ThreadPoolExecutor not asyncio.create_task

fix: use >= version constraints in requirements to unblock Docker build on HF Space

fix: NEWS_API_KEY standardization, remove Dockerfile dead lines, fix HF variable collision

fix: HF Spaces compatibility and self-improving AI - port 7860, CORS, yfinance waterfall, persistent memory via HF Dataset, skill distillation, news symbol fixing, and service integration

Fix curiosity engine: integrate smart router for free LLM providers (Gemini/Groq/Cloudflare/Ollama)

Fix imports in backend/app/main.py to use relative imports

Fix: add missing dataset_persistence stub module

Fix Dockerfile - set WORKDIR to /app/backend and update README

Fix Dockerfile - set PYTHONPATH to include backend

Fix Dockerfile for HF Spaces - handle playwright failure gracefully

Add yfinance to requirements

Fix: markets page - real stock data via Yahoo Finance

feat: add curation layer for self-improvement (Phase 2)

feat: add domain classifier training module for Phase 3 self-improvement system

feat: add domain classifier for self-improvement (Phase 3)

feat: add curation layer for self-improvement (Phase 2)

feat: add observation layer for self-improvement (Phase 1)

stabilize: remove duplicate daemon, non-fatal config, lazy graph, feature flags, gated routers

fix: use Actually Relevant API as primary news source, Alpha Vantage as fallback

fix: add Alpha Vantage fallback for news when NewsAPI fails on hosted servers

feat: self-training engine — response critique, prompt optimization, response refinement

feat: autonomous self-improvement — HF dataset search, extraction, fine-tuning builder

feat: self-reflection engine live — gaps detected, dataset building, opinions forming

feat: add self-reflection engine — opinions, corrections, knowledge gaps, training dataset

fix: make discovery queries go through LLM, faster daemon cycles

feat: Janus v0.4.0 - living intelligence system

feat: Add Workspace tab with workflows, sessions, curiosity discoveries

feat: Fix Pulse tab URLs, add Curiosity Engine, integrate into daemon

feat: Janus living intelligence system — complete build

feat: MiroOrg v2 — native simulation, intelligent caching, adaptive intelligence, embedded charts, premium UI overhaul

feat: MiroOrg v2 full rebuild — conditional graph, new model layer, finance/mirofish nodes, research tools, sentinel layer, frontend overhaul

feat: MiroOrg v1.1 — AI Financial Intelligence System

chore: update .gitignore and sanitize .env.example

Update code

feat: Upgrade MiroOrg Basic to v0.3.0

Switch backend brain from Gemini to DeepSeek

Add backend v0.2 endpoints for cases, prompts, health, config, and agents

Remove stray duplicate agent file

Working backend with LangGraph agent pipeline

Initial clean project structure