Bump the langsmith Python SDK and note the CLI update path in the runbook

Switch the default chat model to DeepSeek V4 Flash via OpenRouter

Rescue account-state provider errors and page ops on provider incidents

Deploy straight to prod on push to main; retire the dev Space flow

Replace the client-level fallback with a turn-level rescue

Refresh MODEL_PRICING against current provider price sheets

De-package graphrag and move to the openai v3 SDK

Replace llama-index with a local Document dataclass

Upgrade backend dependencies; adapt to langchain 1.3.15 and ruff 0.16

Archive private KB bundles for fast cold starts

Only number inline citations that resolved to a source card

Tag LangSmith traces by deployment environment

Remove Claude Haiku 4.5 from the selectable model list

Show a loading skeleton in the sidebar until the tool registry arrives

Always open the sidebar on page load; make collapse session-only

Fix narrow-viewport e2e test for the drawer-based mobile sidebar

Add collapsible desktop sidebar with persisted preference

Fix font-size cascade bug and raise the type scale floor

Rebrand frontend to Towards AI identity and fix mobile layout

Fix stale claims across repo docs

Improve KB command ergonomics, tool-output streaming, and wiki linting

Improve retrieval streaming and tracing

Merge pull request #15 from towardsai/codex/prod-v2-structured-800k

Add model-aware prod v2 memory preset

Add frontend regression test suite

Fix follow-up chat payload compaction

fix: preserve DeepSeek thinking events

Merge pull request #14 from towardsai/fix/deepseek-default-leftovers

docs(evals): F38 numeric refinements from the repricing verification writeup

docs: staleness sweep for core and pipeline docs

docs(evals): findings F35-F38, battery v2 confound corrections, staleness fixes

feat(evals): paired lockstep runner, lossless subagent grading, deprecated-battery guard

feat(experiments): DeepSeek stage-1 compaction arms + prefix-preserving summarization

chore: ignore the local full_stack_course working dir (private course content)

fix(chat): keep one thread on one provider, drop the unselectable Gemini

test(chat): cover effective_tool_names for the default DeepSeek model

fix(tests): join live text deltas without separators, test a selectable Anthropic model

docs: describe the DeepSeek default + Gemini 2.5 fallback accurately

fix(retrieval): store the BM25 index as gzipped JSON instead of pickle (#13)

Format DeepSeek default tests

Use direct DeepSeek as default model

Use OpenRouter DeepSeek with Gemini fallback

chore(versions): capture 2026-07-06 docs refresh, purge retired openai_cookbooks

fix(versions): llms.txt sources get version=null, not a Last-Modified date

fix(pipeline): single-source runs rebuild all_sources_data.jsonl safely

fix(evals): check_triggers evaluates every run before aggregating

fix(evals): never judge faithfulness on silently truncated evidence

fix(pipeline): make upload_data_to_hf report failure honestly

fix(registry): retire sources from every registry dict, not just the first

fix(data-hygiene): gitignore the retire workflow's .bak backups