Protect premium-baked artifacts from runtime overwrites

Premium 429 retry-with-backoff so prebake self-paces under tier caps

Premium-mode cap overrides for richer prebaked artifacts

Phase 4: artifact provenance — model-tracking + debug endpoint

Phase 2: premium tier opt-in via Anthropic Claude

Phase 1: persist tour/diagram/README/repo-map artifacts to Qdrant

UX polish: agent badge, remove All repos, wire ingestion progress

Parallelize Voyage embeds, checkpoint ingestion, add client timeout

Switch embeddings to Gemini

Fix tour layout, synthesis reliability, and cache safety

Simplify Phase 1 stopping criterion and concept rule

Revert glow changes; remove hardcoded examples from prompt

Fix card count and density — root causes, not prompt wording

Lead exploration with glob+grep instead of list_files

Add glob and grep tools to ReAct exploration agent

Reframe tour from design decisions to contributor curriculum

fix(synthesis): block Cerebras 8B from synthesis fallback chain

fix: treat Gemini 500 internal error as fallback trigger

fix(tour): always use strongest model for synthesis

improve(tour): signal-checklist DONE criterion, gaps field, clean domain terms

improve(tour): broader Phase 1 exploration, richer Phase 3 descriptions

fix(sambanova): switch to DeepSeek-V3.1 (Meta-Llama-3.1-405B deprecated)

fix: skip Gemma 4 inside the fallback chain, not before it

fix: skip thinking models for forced DONE and Phase 3 synthesis

fix(tour): suppress THINK block in forced DONE to stop token bleed

fix: cascade order, Cerebras model, forced DONE char limits

fix(cascade): prioritize SambaNova 405B over Gemma4 31B and Cerebras

fix: Cerebras model slug, DONE template, trace gap display

fix(generation): surface clean error when all providers exhausted

fix(tour): omit key_functions from forced DONE to prevent JSON bloat

fix: Cerebras model slug rename + SambaNova cascade ordering

fix: treat 503 high-demand as exhausted, trigger provider fallback

fix: stop SDK retries, add timeout fallback, increase DONE budget to 3000

fix: increase forced DONE budget from 600 to 1500 tokens

fix: increase token budget and timeout for Gemma 4 31B

fix: add 30s timeout to all OpenAI client instantiations

feat: add Gemma 4 31B as fallback after Gemini 2.5 Flash

fix: provider exhaustion window prevents parallel cascade to Groq

fix: Phase 1 diversity rule — principle-based, no domain terms

fix: progress label shows tool call not THINK text during tour generation

feat: agentic Phase 2 + trace UI fix + token budget cuts

feat: agentic Phase 1 with real MCP tools + ReAct trace in UI

fix: tour quality — filter artifact stages + hollow insights + draggable cards

fix: make Phase 1 file path example language-agnostic + add cached repo data

fix: concept-first pipeline — identify decisions before files

fix: Phase 1 reads manifest files first, like claude-code /init

fix: make Phase 1 prompt generic — no repo-specific directory names

fix: Phase 1 uses directory structure not bootstrap import graph

Quality: principle-based evaluator + pipeline-stage filtering

Quality: 9.2/10 — tests, Qdrant feedback, token budgeting, adaptive evaluator