1.1.2

Added

• DuckDB view import restored: inspect and snapshot import now lists views alongside tables. Export runs with DuckDB enable_external_access=false (and allowed_directories for the snapshot output path) so views that only reference in-catalog data import correctly while definitions that read external files, attach other databases, or access the network are blocked. Disable with DCC_ENABLE_DUCKDB_VIEW_IMPORT=false for table-only mode.

Fixed

• Re-selecting a dataset in the sidebar now invalidates cached profile, sample, and related queries so Columns, Samples, and Charts reflect current data instead of stale TanStack Query results.
• Frontend dependencies upgraded to patched releases, resolving all npm audit findings.
• Vite dev server pre-bundles heavy lazy-loaded deps at startup so first navigation to Columns/Charts/Query no longer triggers 504 (Outdated Optimize Dep) errors.
• Backend dependencies upgraded (starlette, python-multipart, pydantic-settings, and transitive msgpack/pip) so Security Audit pip-audit passes in CI.
• Added pytz dependency so timestamp-with-timezone columns sample correctly.

Full changelog: CHANGELOG.md

1.1.0

Upgrade from 1.0.0

1. Update custom API clients: DuckDB inspect/import APIs use source_id (replaces upload_id).
2. /api/saved-charts was removed; saved chart rows are dropped from local workspaces on next startup.
3. Saved SQL snippets gain an optional description column; existing workspaces migrate in place on startup.
4. Large datasets may rebuild profiles with sample-scoped metrics; the UI polls prepare jobs until ready.

Added

• DuckDB open-from-disk for large local .duckdb files (POST /api/datasets/duckdb/open-local, pick-local native file picker), capabilities endpoint, searchable relation picker, and lazy per-relation row counts.
• Saved SQL snippets now persist an optional description.

Fixed

• Starlette upgraded to a patched release; token middleware uses the routed request path so malformed Host headers can no longer bypass X-DCC-Local-Token.
• gitleaks CI install now verifies the release tarball SHA256 before installation.
• DuckDB relation import now snapshots through a direct read-only connection to the source file so views that reference catalog-qualified names (e.g. oddsfox.schema.table in dbt-built databases) export correctly.
• The local API token is generated once per backend process and reused for background job error redaction.
• Profile refresh and DuckDB import jobs observe cancellation between expensive stages.

Changed

• DuckDB inspect/import APIs use source_id (replaces upload_id). Default inspect is metadata-only (no per-table COUNT(*)).
• Raised default browser upload limits to 2 GiB per file and per batch.
• Upload and path registration now queue a single dataset_prepare background job per dataset.
• Large files use Parquet metadata / bounded counts and sample-scoped null metrics.
• Profiling honors configurable timeouts; prepare jobs report incremental progress.
• Frontend profile loading polls jobs with exponential backoff and deduplicates profile queries across tabs.

Breaking

• Removed unused /api/saved-charts endpoints. Existing local workspaces drop the dcc_saved_charts table on next startup.

1.0.0

[1.0.0] - 2026-05-20

Upgrade from 0.1.0

1. Run make clean-local (or delete .dcc_workspace.duckdb) if you have an existing
   workspace from 0.1.0—the layout and profile cache format changed.
2. Re-open datasets so profiles rebuild (first GET /profile may return
   PROFILE_NOT_READY; the UI polls jobs until ready).
3. Custom Ask integrations must use POST /api/agent/ask/stream only (sync ask removed).

Breaking

• Workspace DB: removed versioned migrations and DCC_WORKSPACE_BACKUP_BEFORE_MIGRATE; existing .dcc_workspace.duckdb files with incompatible dcc_* layouts require make clean-local (or manual delete).
• Profile API: removed legacy fields potential_id_columns, potential_key_columns, primary_date_column; only structure_version: "v4" cache entries are served (stale cache is auto-deleted).
• Agent Ask: removed synchronous POST /api/agent/ask; clients must use POST /api/agent/ask/stream (frontend: askAgentStream only).
• Frontend: removed api.getProfile, exported api.waitForJob, and api.askAgent.

Added

• docs/user-guide.md for product usage (data ingestion, profiles/jobs, SQL, Ask, shortcuts).
• docs/RELEASE.md for maintainer release and tagging steps.
• Root .env.example with commented high-signal DCC_* variables.
• Frontend API type conformance tests (types.test.ts + src/api/__fixtures__/).
• make check-ci for clean-room CI-parity validation (npm ci then make check).
• useDatasetProfile hook and api.fetchDatasetProfile for async first-open profiling (PROFILE_NOT_READY + job polling).
• Registry RLock coverage for concurrent count/unregister paths.
• Hook unit tests for useDatasetProfile, useColumnsTable, useSqlResultsGrid, and useSqlHistory.

Changed

• Frontend: removed the Overview tab; the app opens on Columns by default (/ redirects to /columns).
• Frontend: removed the Quality tab; the header still shows the quality score, column flags and filters live on Columns, /quality redirects to /columns, and the profile-diff dialog was removed from the UI (API unchanged).
• Frontend Columns: default table sort is column name ascending; distribution stats label standard deviation as STDEV (not σ).
• Frontend Ask: focused chat workspace with hero on empty threads, context bar, collapsible history, compact composer, and tighter turn layout.
• Frontend Ask agent: DCC_AGENT_SUMMARIZE_WITH_LLM defaults to true; summarization prompts request direct answers with deterministic fallbacks when the second LLM call fails.
• Frontend Ask composer: compact sticky bar, Options popover (model, row limit, dataset scope by name), always-on suggested prompts, real turn timing summary, conversation list polish.
• Frontend SQL workspace: active-dataset chip, consolidated toolbar, run selection, resizable editor/results split, collapsible schema rail, snippet templates, run timer chip, and editor shortcuts (⌘+Shift+F, ⌘+S).
• Primary tab navigation uses route transitions; Columns and Samples tables use sticky headers.
• Profiler histogram bins and column summary formatting improvements.
• GET /api/datasets/{id}/profile no longer blocks on build_profile; returns PROFILE_NOT_READY with job_id when the cache is empty.
• Upload/register queue profile refresh jobs; refresh endpoint dedupes active profile jobs.
• Datasets API split into datasets_upload, datasets_profile, datasets_inspect, and datasets_jobs modules (URLs unchanged).
• Ask persistence aligned with Workspace.ask (AskStore); agent workflow split into persistence/run modules.
• Query tab: SqlEditor, SchemaDatasetBlock, and useSqlHistory extracted from QueryPage.
• Sample rows API returns structured DuckDB error codes.
• Backend tests reorganized under tests/api/, tests/profiler/, and tests/workspace/.
• Frontend coverage baseline raised to 92% (lines/statements); types.ts excluded from v8 coverage (still validated via types.test.ts).
• Workspace startup drops a legacy schema_version table when the current dcc_* schema validates (avoids requiring make clean-local after the migration removal).
• Documentation reorganized: slim root README.md, tightened docs/5-minute-tour.md, authoritative tier READMEs, expanded SECURITY.md, contributor-focused CONTRIBUTING.md / AGENTS.md.
• Dependency: idna 3.15 (addresses CVE-2026-45409 in 3.14).

Removed

• README Screenshots section and wireframe SVGs under docs/images/.
• Unused frontend/public/icons.svg marketing sprite.
• Unused Label and Kbd UI primitives (and their tests).
• Unused exports cardPadding and severityCssVar from tokens.ts, and chartGrid from chartTheme.ts.
• Stale .streamlit/secrets.toml entry from .gitignore.
• Frontend Quality tab UI (QualityPage, profile diff dialog, issue list page).