feat(agent): add open-ended research methodology

fix DSML token leak and use no-thinking model for all non-tool-bound paths

add deepseek compatibility

update supervisor nudging logic and fix tests

lint

lint

resolve merge conflict

feat: improve agent precision and formatting rules

feat: add format-strip finalizer and fail_safe fallback to supervisor_best_answer

feat: add supervisor review node to validate Final Answer candidates before finalization

feat: add print logs for memory, search, and tool operations

feat: supervisor requires concrete answer to finalize after nudge

GAIA_DATASET_SPLIT=validation

feat: supervisor uses extra strong model with cheap fallback

strong model for supervisor

feat: YouTube fallback strategy + dataset file logging + BLAS thread safety

print more logs

feat: improve agent precision and formatting rules

updated to print

normalizer added

feat: rate limit resilience + formatter safety + fail-safe hard-caps

feat: pause batch on high rate limit window

feat: handle rate limit streak and batch abort

feat: retry gaia task once after cooldown

feat: apply gemini cooldown to async model calls

feat: apply cooldown to bound model calls

feat: track batch rate limit window

feat: track per-question rate limit streaks

feat: abort batch on gemini daily quota

feat: add gemini cooldown registry

feat: classify gemini client rate limits

handle 429

fix

fix 3 layer memory ststem

better persistent memory fix

fix: tell model ~ paths work and warn against root='/' in find_files

fix: add default excludes and truncation sentinel to find_files

fix: expand ~ in fs tools, accept absolute glob patterns, add find_files

feat: upgrade memory logic to use langmem.create_memory_manager

fixed memory

fix: stop stripping function_call from additional_kwargs to preserve gemini thought_signatures

fix: preserve gemini thought_signatures required for tool calls

feat: inject retrieved semantic and episodic memory into system prompt

feat: add episodic task summarization

feat: add semantic memory extraction node using langmem

feat: replace MemorySaver with SqliteSaver for thread persistence

feat: enhance LLM formatting capabilities and improve logging in various components

feat: implement robust web fetcher with redirect handling and add LLM-based answer formatting pipeline

feat: implement secure Python sandbox with process and Docker backends

review: ship all 15 quick-win fixes from REVIEW.md via TDD