# Changelog

All notable changes to Chat Switchboard.

## [0.17.2] — 2026-02-28

### Added
- **CodeMirror 6 integration.** Rich editor infrastructure compiled at Docker
  build time via esbuild (IIFE bundle, ~295KB min / ~90KB gzip). Two factory
  functions exposed on `window.CM`:
  - `CM.chatInput()` — Markdown-mode editor for the chat input with
    auto-growing height, Enter=send / Shift+Enter=newline, spell check, and
    WYSIWYG fenced code block decorations (visual container with monospace
    font and accent border, matching claude.ai UX).
  - `CM.codeEditor()` — Full-featured code editor for admin extension panel
    with line numbers, bracket matching, search/replace, fold gutter, and 10
    bundled language modes (Markdown, JavaScript, JSON, SQL, HTML, CSS, YAML,
    Go, Python, Rust).
- **Graceful degradation.** All CM6 integration points check `window.CM`
  availability. If the bundle fails to load, the app falls back to native
  `<textarea>` with zero breakage.
- **Dark/Light/System theme toggle.** New appearance setting with three
  modes. Light theme overrides all CSS variables via `[data-theme="light"]`
  selector. System mode tracks `prefers-color-scheme` media query in real
  time. Theme changes emit `theme.changed` on the EventBus; CM6 editors
  toggle `oneDark` syntax theme via compartment reconfiguration.
- **Vim/Emacs keybinding preference.** Editor keybinding mode (Standard /
  Vim / Emacs) configurable in appearance settings. Applies to code editors
  and extension editors only — chat input always uses standard keybindings.
  Live-switchable on already-open editors via `keymap.changed` event.
  Bundled statically (~40KB for both modes).
- **Inline code shortcut.** Ctrl/Cmd+E wraps selection in backticks or
  inserts an empty inline code pair with cursor between them.
- **Code block shortcut.** Typing ` ``` ` at the start of a line expands
  to a fenced code block with cursor positioned inside. The decoration plugin
  renders code blocks with a styled visual container in the chat input.
- **CI path-based change detection.** New `detect-changes` job classifies
  changed files into frontend/backend/infra/docs buckets. Test jobs skip
  when irrelevant (FE-only changes skip Go tests, docs-only changes skip
  all tests and deploy). Tags always run the full pipeline.

### Changed
- Docker build pipeline: both `Dockerfile.frontend` and unified `Dockerfile`
  now include a `cm6-build` stage (Node 20 Alpine → esbuild → IIFE bundle).
- `ChatInput` abstraction in `chat.js` replaces direct textarea access
  across 7 callsites (`chat.js`, `tokens.js`, `attachments.js`).
- Extension editor in `admin-handlers.js` uses `CM.codeEditor()` with JSON
  and JavaScript modes, replacing bare `<textarea>` with manual Tab handler.
- Service worker excludes `/vendor/codemirror/` from cache (version-busted).
- Debug state snapshot includes CM6 version and language list.
- `ARCHITECTURE.md` updated to v0.17 reflecting CM6 integration, SQLite
  dual-driver, modular frontend file structure, and theme system.
- CI pipeline header updated to v0.17.2 with path gating documentation.
- `build-editor.sh` falls back to `npm install` if `package-lock.json`
  is missing (belt-and-suspenders for local dev).

### Fixed
- **App initialization crash.** `Events.publish` (nonexistent) → `Events.emit`
  (correct API). The unhandled exception during `initAppearance()` killed
  `initListeners()`, leaving the entire app half-initialized — settings modal
  unclosable, keyboard shortcuts unwired, paste handlers missing.
- **Cursor invisible in dark and light mode.** CM6 cursor used `--accent`
  color (low contrast on both themes). Switched to `--text` for consistent
  visibility. Added explicit `borderLeftWidth: 2px`.
- **Placeholder text offset.** Double padding between `.cm-editor` wrapper
  (12px) and `.cm-content` (8px) pushed placeholder 20px below expected
  position. Zeroed `.cm-content` padding — wrapper is the single source of
  truth.

## [0.17.1] — 2026-02-27

### Added
- **SQLite backend.** Full dual-driver database layer — set `DB_DRIVER=sqlite`
  to run with an embedded SQLite database. Pure Go (no CGO), zero external
  dependencies. 19 store files covering all domain stores: channels, messages,
  users, teams, personas, knowledge bases, notes, usage, audit, extensions,
  and more. Feature parity with Postgres including knowledge base vector
  search via app-level cosine similarity computed in Go.
- **Dialect-aware test infrastructure.** `database.SetupTestDB()` detects
  `DB_DRIVER` and provisions either a Postgres test database or a SQLite
  temp file. Exported `database.PH(n)` returns `$N` or `?` per dialect.
  `database.TruncateAll()` uses `TRUNCATE CASCADE` on Postgres and
  `DELETE FROM` with `PRAGMA foreign_keys` toggling on SQLite.
  `dialectSQL()` helper in handler tests converts `$N` placeholders and
  strips `::jsonb` casts at runtime.
- **SQLite CI pipeline.** New `test-sqlite` job runs the full handler
  integration test suite and store tests against an embedded SQLite
  database. Parallel with the existing Postgres test job. Build gate
  verifies `CGO_ENABLED=0` compilation.
- **Generic provider test config.** Live provider integration tests now
  read `PROVIDER`, `PROVIDER_KEY`, and `PROVIDER_URL` environment
  variables instead of hardcoded `VENICE_API_KEY`. Legacy fallback
  preserved. Model selection prefers non-reasoning models to avoid
  thinking budget requirements with low `max_tokens`.

### Changed
- `kb_chunks` table in SQLite schema includes `embedding TEXT` column
  for JSON-encoded float64 vectors (previously omitted as feature-gated).
- `SimilaritySearch` on SQLite loads candidate chunks, decodes JSON
  embeddings, and computes cosine distance in Go — replacing the
  previous "not available" error.
- `InsertChunks` on SQLite now stores embedding vectors as JSON text.
- Live test names genericized: `TestLive_Venice*` → `TestLive_*`.
- CI `build-and-deploy` depends on `[test, test-frontend, test-sqlite]`.

### Fixed
- **SQLite `RETURNING` + `time.Time` scan failure.** The modernc/sqlite
  driver cannot scan `datetime('now')` TEXT columns into `time.Time` via
  `RETURNING`. All 13 SQLite store `Create` methods rewritten to set
  timestamps in Go (`time.Now().UTC()`) and use `ExecContext` instead of
  `QueryRowContext(...).Scan()`. Format: `2006-01-02 15:04:05` (`timeFmt`
  constant in `helpers.go`).
- **SQLite missing `id` in INSERTs.** Unlike Postgres (`DEFAULT
  gen_random_uuid()`), SQLite `TEXT PRIMARY KEY` columns have no
  auto-generation. Added `store.NewID()` / `uuid.New()` to: `usage_log`,
  `model_pricing` (both upsert paths), `team_members` (AddMember),
  `group_members` (AddMember), `refresh_tokens` (CreateRefreshToken).
- **Test seed helpers SQLite-aware.** `SeedTestUser`, `SeedTestChannel`,
  `SeedTestTeam`, `SeedTestTeamMember`, `SeedTestGroup`, `SeedGroupMember`
  now branch on `IsSQLite()` to provide application-generated UUIDs
  instead of relying on `RETURNING id`.
- **Postgres-isms in SQLite stores.** `extension.go` Update used `now()`
  → `datetime('now')`; `ListForUser` COALESCE used `true` → `1`.
- SQLite store bool/int type mismatches: `persona.go` auto-fetch flag,
  `usage.go` exclude-BYOK filter, `user_settings.go` visibility map
  values — all corrected from `0`/`1` to `false`/`true`.

## [0.17.0] — 2026-02-27

### Added
- **Persona-KB binding.** Personas can now have knowledge bases directly
  bound to them. When a user selects a persona with bound KBs, the
  `kb_search` tool is automatically scoped to those KBs and the persona's
  system prompt includes a KB listing hint. Admin and team admin preset
  forms include a KB picker with per-KB auto-search toggles.
  Migration adds `persona_knowledge_bases` join table.
- **Enterprise KB mode.** New `discoverable` flag on knowledge bases
  controls whether users can see and attach KBs directly. When
  `kb_direct_access` platform policy is set to `true`, users cannot
  attach KBs to channels directly — they access KBs exclusively through
  persona bindings curated by admins. New `ListDiscoverableKBs` and
  `SetDiscoverable` endpoints.
- **Role fallback alerts (issue #69).** When a role's primary provider
  fails and the fallback activates, the resolver emits a `role.fallback`
  event on the EventBus with audit log entry. Admin users see a persistent
  dismissable banner. 5-minute per-role cooldown prevents flooding.
- **Chat rename.** Double-click any chat title in the sidebar to edit
  inline. Enter saves, Escape cancels.
- **Utility model auto-naming.** After the first assistant response,
  a background request to the utility role generates a concise title.
  Falls back to truncation when no utility model is configured.
  New endpoint: `POST /channels/:id/generate-title`.
- **Chat token count.** Conversation token estimate shown in the model
  bar, color-coded against context budget.
- **State restore on refresh.** Active chat ID persisted to
  `sessionStorage`, auto-restored on page reload.

### Fixed
- **ResolvePreset group access bypass.** `ResolvePreset()` used raw SQL
  that skipped `resource_grants` checks from v0.16.0. Now uses
  `PersonaStore.UserCanAccess()`.
- **KB create scope authorization.** Now enforces admin role for global
  KBs and team admin role for team KBs.
- **Completion handler persona ID scoping.** `personaID` variable scoped
  inside an `if` block but referenced downstream. Fixed by threading
  through function signatures.
- **Nil slice JSON marshaling.** `ListDiscoverableKBs` returns `[]`
  instead of `null` when no KBs match.

### Changed
- `UpdateDocumentStorageKey` moved from raw `ExecContext` to store method.
- EventBus route table: `role.fallback` → `DirToClient`.
- Resolver gains `.WithBus(bus)` builder (nil-safe, backwards compatible).
- Service worker: `/extensions/` excluded from fetch handler.
- Mermaid renderer v2.0 (pan/zoom, export) promoted to builtin.
- Removed DIAG diagnostics from `TestGroupBasedPersonaAccess`.
- Paste-to-file threshold synced from backend `PublicSettings`
  (`storage.paste_to_file_chars`, admin-configurable, default 2000).
  Resolves hardcoded constant in `attachments.js`.
- Embedding dropdown already had tolerant type filter, manual model ID
  fallback, and auto-switch on empty — confirmed complete, checkbox updated.

## [0.16.0] — 2026-02-27

### Added
- **User groups.** Global and team-scoped groups decoupled from team
  membership. `groups` and `group_members` tables. Admin and team admin
  CRUD for group management. Groups serve as ACL targets for resources.
- **Resource grants.** Three-way grant model (`team_only`, `global`,
  `groups`) for Personas and Knowledge Bases. `resource_grants` table
  with `grant_scope` and `granted_groups UUID[]` columns. Grant picker
  UI on Persona and KB forms with group multi-select.
- **Schema consolidation.** 9 incremental migrations collapsed into
  single `001_v016_schema.sql`. Fresh installs use the consolidated
  file; upgrade path preserved via migration version tracking.

### Changed
- Persona and KB list queries now filter through `resource_grants` for
  non-admin users. Team-only remains the default scope.
- Admin panel shows group membership counts and grant summaries.

## [0.15.1] — 2026-02-26

### Added
- **`attachment_recall` tool.** Two operations: `list` returns filenames
  and metadata for the current channel's attachments; `read` extracts
  and returns content by attachment ID. Channel-scoped access control.
- **`conversation_search` tool.** Full-text search across the current
  channel's message history using PostgreSQL `plainto_tsquery`. Returns
  matching messages with timestamps and role context.
- **Token estimator attachment awareness.** `Tokens.estimateAttachments()`
  accounts for staged file sizes in context budget calculations. Warning
  thresholds include attachment estimates.

## [0.15.0] — 2026-02-26

### Added
- **Background compaction scanner.** Periodic scan identifies channels
  exceeding configurable context thresholds. Automatic summarization via
  utility role compresses old messages into summary nodes. Per-channel
  opt-in/out via `auto_compact` channel setting.
- **Compaction service.** `compaction.Service` orchestrates summary
  generation: estimates token usage, selects messages for compression,
  calls utility role, inserts summary as tree boundary node with metadata.
- **Context budget guard rail.** 80% ceiling prevents compaction from
  triggering mid-generation. Cooldown timer prevents repeated compaction
  of the same channel.
- **Summarize & Continue button.** User-triggered compaction from the
  context warning bar. Reuses compaction service with immediate execution.

### Changed
- Scanner configurable via `global_settings`: threshold percentage,
  cooldown duration, enabled/disabled toggle. Channel-level overrides.

## [0.14.0] — 2026-02-26

### Added
- **Knowledge bases.** RAG pipeline: upload documents → chunk (recursive
  text splitter) → embed via pgvector → `kb_search` tool for semantic
  retrieval. Team and personal KB scopes. Channel KB toggle enables
  per-conversation knowledge access.
- **Document ingestion.** `knowledge.Ingest()` pipeline: file upload →
  text extraction (reuses v0.12.0 pipeline) → chunking with configurable
  overlap → embedding via the embedding role → storage as `kb_chunks`
  with vector index.
- **KB admin panel.** Knowledge Bases section under AI category. Create,
  delete, upload documents, view chunk counts and storage usage. Team
  admin scoped to team KBs.
- **Notes semantic search.** Note search upgraded from exact text match
  to pgvector cosine similarity when embeddings are available.
- **`kb_search` tool.** Registered when embedding role is configured.
  Accepts query text, returns top-K chunks with source document
  attribution. Channel KB bindings control which KBs are searched.

### Changed
- pgvector extension enabled in schema (`CREATE EXTENSION IF NOT EXISTS
  vector`). `kb_chunks` table includes `embedding vector(1536)` column
  with IVFFlat index.

## [0.13.1] — 2026-02-26

### Added
- **`web_search` tool.** Search provider abstraction with two backends:
  DuckDuckGo (HTML scraping, zero config) and SearXNG (self-hosted,
  JSON API). Returns title, URL, and snippet for each result. Configured
  via `SEARCH_PROVIDER` and `SEARXNG_URL` env vars.
- **`url_fetch` tool.** Fetches and extracts text content from URLs.
  Respects robots.txt. Content-type detection with HTML-to-text
  conversion. Configurable timeout and size limits.
- **Tool categories.** Tools now have a `category` field (builtin,
  search, knowledge, browser). Tools toggle UI in chat bar groups
  by category with per-tool enable/disable.
- **Tools toggle UI.** Popup menu on chat input toolbar showing all
  available tools. Per-tool checkbox state sent as `disabled_tools[]`
  in completion requests. Browser extension tools included.

## [0.13.0] — 2026-02-25

### Changed
- **Admin panel refactor.** Replaced 12-tab modal with fullscreen admin
  panel. Four categories (People, AI, System, Monitoring) with section
  sidebar navigation. URL-based routing (`#admin/people/users`).
  Responsive layout, classification banner-aware positioning.
- **CSS design token cleanup.** Consolidated duplicate color/spacing
  variables. Admin panel uses shared token system with main UI.

## [0.12.0] — 2026-02-25

### Added
- **File handling and vision support.** Upload images, PDFs, and documents
  into chat via 📎 button, drag-and-drop, or paste. Multimodal message
  assembly injects base64 images for vision-capable models and extracted
  text for documents. Staged attachment strip shows upload progress and
  extraction status. Auth-aware blob rendering with Bearer tokens. Image
  lightbox viewer. Vision capability hints on image attachments.
- **Storage backend abstraction.** `ObjectStore` interface with two
  implementations: PVC (local filesystem) for single-node/dev and S3
  (minio-go) for multi-node production. S3 backend works with any
  S3-compatible API: MinIO, Ceph RGW, AWS S3. Auto-detection when
  `STORAGE_BACKEND` is not set (tries PVC, disables if not writable).
  Both backends share identical semantics — all handlers, attachment
  CRUD, multimodal assembly, and orphan cleanup work regardless of
  backend.
- **S3 configuration.** `S3_ENDPOINT`, `S3_BUCKET`, `S3_ACCESS_KEY`,
  `S3_SECRET_KEY`, `S3_REGION`, `S3_PREFIX`, `S3_FORCE_PATH_STYLE` env
  vars. Endpoint auto-detects SSL from scheme. Path-style URLs default
  on (required for MinIO/Ceph). Optional key prefix for shared buckets.
  CI pipeline syncs S3 secrets to K8s when `STORAGE_BACKEND=s3`.
- **Text extraction pipeline.** PDF, DOCX, XLSX, PPTX, ODT, RTF text
  extraction via filesystem-based queue. Inline and sidecar modes.
  Crash recovery for items stuck in processing state.
- **Admin storage panel.** Backend health, file count, total size,
  orphan detection and cleanup. Shows PVC path or S3 endpoint/bucket
  depending on active backend. Extraction queue status.
- **Vault CLI commands.** `switchboard vault rekey` re-encrypts all
  provider API keys when rotating the encryption key.
  `switchboard vault status` shows encryption health. Admin UI
  encryption status indicator in Settings tab.
- **Per-chat model persistence.** Server-side `channels.settings`
  JSONB field stores `last_selector_id`. Roams across devices with
  localStorage write-through cache as fallback.

## [0.11.0] — 2026-02-25

### Added
- **Browser extension system.** Full lifecycle: manifest registration, script
  injection, scoped context (`ctx.renderers`, `ctx.tools`, `ctx.events`,
  `ctx.storage`, `ctx.ui`), permission-aware API proxying. Extensions self-
  register via `Extensions.register()` and receive isolated contexts during
  `initAll()`. Admin CRUD endpoints, asset serving (public, no auth needed
  for `<script>` tag loading), auto-seeder for builtin extensions.
- **Custom renderer pipeline.** Block renderers intercept code fences by
  language tag, post renderers process the DOM after insertion. Case-
  insensitive language matching handles LLM capitalization (`Mermaid`,
  `Diff`, `CSV`). Nested ` ```markdown ``` ` fence unwrapping prevents the
  common LLM pattern of wrapping responses in markdown fences from breaking
  inner code blocks.
- **Browser tool bridge.** Extensions register tool handlers via
  `ctx.tools.handle()`. Server collects browser tool schemas alongside
  server tools, includes them in LLM completions. Tool calls route through
  WebSocket EventBus (`tool.call.*` → browser → `tool.result.*` → server).
  30-second timeout with error recovery.
- **Server tools: calculator and datetime.** Auto-register via `init()`,
  zero wiring changes. Calculator: recursive-descent evaluator with
  arithmetic, 17 math functions, constants. Datetime: current date/time/
  timezone/unix/ISO-week in any IANA timezone.
- **6 built-in browser extensions** (self-contained, own CSS via
  `_injectStyles()`/`destroy()` lifecycle):
  - **Mermaid**: block + post renderer, SVG diagrams, dark mode detection,
    local vendor with CDN fallback, max-height constraint with scroll
  - **KaTeX**: block renderer (` ```latex/math/tex ```) + post renderer
    (inline `$...$` and `$$...$$` in text nodes)
  - **CSV Table**: block renderer, RFC 4180 parser with quoted fields,
    sortable columns (click headers), numeric-aware sorting
  - **Diff Viewer**: block renderer, red/green syntax highlighting for
    unified diffs, stats badge (+N/-N), hunk/file headers
  - **JS Sandbox**: browser tool (`js_eval`), sandboxed iframe
    (`allow-scripts` only), console capture, 10s timeout
  - **Regex Tester**: browser tool (`regex_test`), multi-input matching,
    full match details with named groups and indices
- **Admin extension editor.** Edit button on each extension in Admin →
  Extensions. Inline editor shows name, description, manifest JSON, and
  script source with tab-key support. System extensions show overwrite
  warning.
- **Enhanced diagnostics.** Test 5: browser extensions (loaded, active,
  renderers, tool handlers). Test 6: Service Worker cache (registration,
  scope, state, cache names, entry counts). Purge Cache button in Debug
  Log modal footer.
- **Extension-owned styles.** All 4 rendering extensions (mermaid, katex,
  csv, diff) inject their own `<style>` tags during `init()` and remove
  them during `destroy()`. Global stylesheet only keeps `.ext-rendered`
  wrapper. User-created extensions can bring their own CSS without
  touching the global stylesheet.
- **Notes extension rendering.** `runExtensionPostRender()` called in both
  note read mode and preview mode. Mermaid diagrams and KaTeX math now
  render correctly in notes.
- **Database migration 006_extensions.sql.** `extensions` and
  `extension_user_settings` tables.

### Fixed
- **WebSocket token field mismatch.** API saved `{ accessToken: '...' }`
  but EventBus read `tokens.access`. Tool bridge was dead on arrival.
- **Extension asset 401.** Asset route was behind auth middleware, but
  `<script>` tags don't send Authorization headers. Moved to public group.
- **`runExtensionPostRender is not defined`.** Stale Service Worker cache
  served old `ui-format.js` without the function. Added `typeof` guard.
- **Extension init order.** Extensions loaded after `loadChats()` — block
  renderers not registered when messages first rendered. Moved extension
  loading before chat loading.
- **K8s Ingress WebSocket routing.** Traefik `pathType: Exact` doesn't
  reliably win over `Prefix` rules in the same Ingress resource. Changed
  `/ws` and `/health` to `Prefix` for longest-prefix-wins routing.
- **Mermaid SVG oversized.** Added `max-height: 600px` on diagram
  container with overflow scroll, removed mermaid.js hardcoded `height`
  attribute from SVGs so CSS constraints apply.

### Changed
- Extension CSS removed from global `styles.css`. Each extension owns its
  styles via `_injectStyles()` in `init()` with `destroy()` cleanup.
  Idempotent injection guards prevent duplicates.
- Markdown fence language extraction lowercased at the point of extraction
  in `ui-format.js`, making all block renderer pattern matches
  case-insensitive without per-extension workarounds.
- `_unwrapMarkdownFence()` pre-processor strips outer ` ```markdown ```  `
  wrappers when they contain nested fences. Handles think-block
  placeholders and trailing explanation text. Only triggers when nested
  fences are present — plain markdown code blocks still render normally.

## [0.10.5] — 2026-02-24

### Added
- **`ui-primitives.js` — shared rendering primitives and registries.**
  Single source of truth for provider types, role definitions, and reusable
  UI components. Extension-ready via registry pattern (`Providers.add()`,
  `Roles.add()`). Primitives follow the `renderPresetForm()` pattern:
  `(container, options) → control object` with `getValues/setValues/clear`.
  - `Providers` registry — types, labels, default endpoints (was 5 places → 1)
  - `Roles` registry — names, type filters, hints (was hardcoded in 2 places → 1)
  - `renderCapBadges(caps, opts)` — consolidates 3 badge builders (compact + detailed)
  - `renderProviderForm(container, opts)` — replaces 3 form implementations
  - `renderProviderList(container, opts)` — replaces 3 list renderers
  - `renderRoleConfig(container, opts)` — replaces 2 role UIs + 4 handlers
  - `renderUsageDashboard(container, opts)` — replaces 3 usage renderers

### Fixed
- **Admin provider form now auto-fills endpoint on type change.** Was missing
  from admin (worked in user BYOK and team forms). Now all three scopes use
  the same primitive with identical behavior.
- **Team provider form consolidated.** Was two separate HTML forms (create +
  edit) with separate listeners. Now a single dual-mode form matching the
  pattern used by admin and user BYOK scopes.

### Changed
- Provider type definitions removed from `index.html` (2 static `<select>`s)
  and `settings-handlers.js` (1 dynamic build + 2 endpoint maps). All now
  sourced from `Providers` registry in `ui-primitives.js`.
- Provider list rendering uses event delegation instead of inline `onclick`
  handlers. Each list returns `{ refresh, getCache }` control handles.
- Role configuration uses `data-role-*` attributes for event delegation
  instead of `id`-based selectors and global `onchange` handlers.
- Usage dashboards accept options for compact/full mode, custom API calls,
  and extension-provided extra columns.
- All 16 `confirm()` calls replaced with `showConfirm()` — styled modal
  dialog matching the app design instead of browser-native dialog. Supports
  `danger` styling, Escape/Enter keys, click-outside dismiss.
- Sidebar collapse icon now always visible (dimmed) next to the logo,
  brightens on hover. Previously the icon replaced the logo on hover only.
  When sidebar is collapsed, only the collapse icon shows (logo hidden).
- New `.popup-menu` + `.popup-menu-item` shared CSS base for all dropdown
  and flyout menus. `createPopupMenu(anchor, opts)` primitive available for
  future menu creation with consistent behavior.
- Removed ~360 lines of duplicated code across `admin-handlers.js`,
  `settings-handlers.js`, `ui-admin.js`, and `ui-settings.js`.
- Removed ~40 lines of static HTML form markup from `index.html`.

## [0.10.4] — 2026-02-24

### Added
- **`model_type` field across the full pipeline.** Models now carry a type
  classification (`chat`, `embedding`, `image`) sourced from provider APIs
  at sync time — no hardcoded lists. Venice's `/v1/models` returns `type`
  per model; OpenAI-compatible APIs pass through the field when present.
  - New DB column: `model_catalog.model_type VARCHAR(20) DEFAULT 'chat'`
  - Migration: `005_model_type.sql`
  - Propagation: `providers.Model.Type` → `CatalogSyncEntry.ModelType` →
    `CatalogEntry.ModelType` → `UserModel.ModelType` → frontend `model_type`

### Fixed
- **Admin role save didn't refresh UI.** `adminSaveRole()` showed "✓ Saved"
  but never called `UI.loadAdminRoles()`, so dropdowns appeared stale after
  save. Now reloads the roles panel after a successful save.
- **Role model dropdowns showed all models regardless of type.** Embedding
  role showed chat models, utility role showed embedding models. Both admin
  and user role UIs now filter the model dropdown by `model_type`:
  - "embedding" role → only `model_type === 'embedding'` models
  - "utility" role → only `model_type === 'chat'` models

### Changed
- Venice provider now reads the `type` field from each model in the API
  response and normalizes it (`text` → `chat`, `embedding` → `embedding`,
  `image` → `image`).
- OpenAI provider wire type extended with optional `type` field for
  OpenAI-compatible APIs that include it.

## [0.10.3] — 2026-02-24

### Changed
- **Frontend refactor: 2 monolith files → 13 domain-scoped files.** Split
  `ui.js` (2,582 lines) and `app.js` (2,940 lines) into 13 focused files
  averaging ~544 lines each. No features added, no functions renamed, no
  architectural changes. Vanilla JS, no modules, no build step.

  **New file structure:**

  | File | Lines | Domain |
  |------|------:|--------|
  | ui-format.js | 353 | Markdown rendering, `esc()`, code blocks, side panel |
  | ui-core.js | 974 | UI object: sidebar, chat list, messages, streaming, model selector |
  | ui-settings.js | 640 | Settings tabs, teams, providers, user preferences |
  | ui-admin.js | 645 | Admin tabs, users, roles, usage, teams |
  | tokens.js | 123 | Context tracking, token estimation |
  | notes.js | 364 | Notes panel, editor, multi-select |
  | chat.js | 584 | Chat ops, send, regen, edit, branch, summarize |
  | settings-handlers.js | 692 | Settings save, provider CRUD, command palette |
  | admin-handlers.js | 652 | Admin actions, presets, team management |
  | app.js | 567 | State, init, boot, auth, listener dispatch |

  Unchanged: `api.js` (575), `debug.js` (580), `events.js` (327).

- **`initListeners()` decomposed** into domain-specific init functions:
  `_initChatListeners()`, `_initSettingsListeners()`, `_initAdminListeners()`,
  `_initNotesListeners()`, `_initGlobalKeyboard()`. The orchestrator in `app.js`
  dispatches to each.
- **Side panel resize** changed from self-invoking IIFE to `_initSidePanelResize()`
  called during listener init, avoiding DOM timing issues.
- **Service worker** updated with new file list for pre-caching.
- **Policy-gating tests** updated to read from the correct source files after
  the split. All 159 tests pass.

## [0.10.2] — 2026-02-24

### Added
- **Summarize & Continue** — User-triggered conversation compaction using the
  utility model role. Button appears in context warning bar at ≥75% context
  usage. `POST /channels/:id/summarize` calls the utility role to generate a
  summary, inserts it as a tree node with `metadata.type = "summary"`, and
  updates the cursor. Subsequent completions use the summary as a context
  boundary — messages before it are replaced by the summary as a system message.
  Multiple summaries stack. Fork-aware (summaries are tree nodes with their own
  branch position). "Show full history" toggle reveals collapsed earlier messages.
- **BYOK Role Overrides** — Users with personal providers can override the org's
  utility and embedding model roles. Resolution chain: personal → team → global.
  New "Model Roles" tab in Settings (visible when BYOK is enabled). Stored in
  `user.settings` JSONB under `model_roles` key, same shape as team overrides.
- **Utility Rate Limiting** — `utility_rate_limit` global setting (default: 20
  calls/hour/user, 0 = unlimited). Org-funded utility calls check against
  `usage_log` before executing. BYOK calls exempt (user pays their own way).
  Returns 429 with clear message when exceeded.
- **Message metadata in path** — `PathMessage` now includes `metadata` JSONB
  field, enabling summary detection and future message-type extensibility.
- **Per-chat model/preset restore** — Switching between chats now restores the
  last-used model or preset in the selector. Stored in localStorage keyed by
  channel ID. Falls back to the channel's base model, then the global default,
  if the original selection is no longer available. Cleaned up on chat deletion.

### Changed
- **Generation role removed** — `RoleGeneration` ("generation") removed from
  `ValidRoles` and admin Roles tab. Image/media generation will be
  extension-managed (v0.11.0) with its own provider config, not a role slot.
  The current completion/embedding abstraction doesn't fit image gen's
  fundamentally different API surface.
- **Resolver.Complete/Embed signatures** — Now accept `userID` parameter for
  personal role override resolution. Empty string skips personal lookup.
- **Proactive token refresh** — Access token is now refreshed at 80% of its
  lifetime (~12min for 15min tokens). On page reload, a conservative 60s refresh
  is scheduled since token age is unknown. Eliminates the race condition where
  profile succeeds on a near-expired token but subsequent calls fail.
- **Auth guard in `startApp()`** — If token is invalidated during startup
  (e.g., 401 on loadChats after profile succeeded), app returns to login
  instead of rendering a broken UI with cascading 401 errors.
- **Per-chat model restore fallback** — When the stored model/preset is removed,
  falls back to admin default → first visible model (was keeping stale selection).
  Also cleans up the stale localStorage entry.
- **BYOK Role Overrides** — Fixed response parsing (`configs.configs` not
  `configs.data`) and model field mapping (`configId`/`baseModelId` not
  `provider_config_id`/`model_id`).
- **`GetRole` endpoint integration test** — `GET /admin/roles/:role` now
  exercised by CI via `TestIntegration_Roles_GetSingleRole`, catching the
  `GetConfig` signature mismatch that caused the build failure.
- **Auth resilience frontend tests** — 10 tests covering startup auth guard,
  token refresh lifecycle, and the profile-success-then-401 edge case.
- **JS syntax lint in CI** — `node --check` runs on all `src/js/*.js` files
  before frontend tests, catching parse errors that tests alone can't detect.

## [0.10.1] — 2026-02-24

### Added
- **Admin System Prompt** — Global system prompt configured in Admin › Settings.
  Injected as the first system message in every conversation, before user/preset
  system prompts. Users cannot override or disable it. Stored in
  `global_settings['system_prompt']`.
- **Personas tab** — User presets moved from the Models tab to their own
  "Personas" tab in Settings. Tab is hidden when admin disables user presets
  (`allow_user_personas` policy).
- **Team Management modal** — Team admin functionality extracted from Settings
  into its own tabbed modal (Members, Providers, Presets, Usage, Activity).
  Accessed via "Team Management" flyout menu item, or "Manage →" on team cards
  in Settings. Multi-team admins see a picker first; single-team admins go
  straight to the tabbed view. Back arrow in header returns to picker.
- **Preview pane: clear button** — Trash icon in the preview header resets the
  iframe and shows the empty hint. Also auto-clears when deleting a chat that
  had active preview content.
- **Preview/Notes pane: fullscreen** — Expand button in the header toggles
  fullscreen mode (panel fills entire viewport width).
- **Preview/Notes pane: resizable** — Drag the left edge of the side panel
  to resize (280px–70vw). Width resets on close.
- **Code block: download** — "Download" button on every code block. Infers
  file extension from language tag (e.g. `python` → `.py`, `go` → `.go`).

### Changed
- **Personal Usage scoped to BYOK** — `GET /usage` and the user Usage tab now
  only show consumption against personal (BYOK) providers. Global provider
  usage is the org's cost, not the user's. New `QueryByUserPersonal` store
  method filters on `scope = 'personal' AND owner_id = user_id`. Admin usage
  views remain unaffected. Usage tab hidden when admin disables user providers
  (`allow_user_byok` policy).
- **OpenAI streaming usage race** — Deferred the Done event until the usage
  chunk arrives. Previously, finish_reason fired Done before the usage chunk
  (which has `choices: []`), so streaming token counts were always 0 for all
  OpenAI-compatible providers.

### Fixed
- **Usage logging zero-token guard** — Removed early exit in `logUsage` that
  suppressed rows when tokens were 0. Combined with the streaming race above,
  this meant no streaming usage was ever recorded.
- **Live chat completion test** — `TestLive_VeniceChatCompletion` sent wrong
  field names (`messages`/`config_id` instead of `content`/`provider_config_id`).

## [0.10.0] — 2026-02-24

### Added
- **Model Roles** — Named model slots (`utility`, `embedding`, `generation`)
  with primary + fallback bindings. Stored in `global_settings['model_roles']`.
  Team-level overrides via `teams.settings` JSONB. New `server/roles/` package
  with `Resolver.Complete()` and `Resolver.Embed()` for automatic failover.
  Admin API: `GET/PUT /admin/roles/:role`, `POST /admin/roles/:role/test`.
  Team API: `GET/PUT/DELETE /teams/:id/roles/:role`.
- **Provider Embed() interface** — All providers implement `Embed()`. OpenAI,
  Venice, OpenRouter support `/v1/embeddings`; Anthropic returns `ErrNotSupported`.
  New types: `EmbeddingRequest`, `EmbeddingResponse`.
- **Usage Tracking** — Every completion (streaming and non-streaming) logs
  token counts and cost to `usage_log` table. Cost calculated at insert time
  from `model_pricing` (no retroactive recalculation). Provider scope
  denormalized for efficient admin filtering. Admin views exclude BYOK.
  New stores: `UsageStore`, `PricingStore`. Admin API:
  `GET /admin/usage`, `GET /admin/usage/users/:id`, `GET /admin/usage/teams/:id`.
  User API: `GET /usage` (personal summary).
- **Model Pricing** — `model_pricing` table with catalog sync and manual admin
  overrides. Sync from provider APIs (Venice, OpenRouter) auto-populates
  pricing with `source='catalog'`; manual entries never overwritten. Admin API:
  `GET/PUT/DELETE /admin/pricing`.
- **Streaming token capture** — OpenAI: `stream_options.include_usage` +
  parse usage/cache from final chunk. Anthropic: parse `message_start` for
  input/cache tokens, `message_delta` for output tokens. Cache token fields
  (`CacheCreationTokens`, `CacheReadTokens`) on `CompletionResponse`,
  `StreamEvent`, and `streamResult`.
- **Admin Roles tab** — Configure primary/fallback provider+model per role,
  test-fire from UI.
- **Admin Usage dashboard** — Period selector (7d/30d/90d), group-by
  (model/user/day/provider), totals cards, breakdown table, pricing table.
- **Team Usage dashboard** — Team admins can view usage against their
  team-owned providers via `GET /teams/:teamId/usage`. Filters to
  `provider_configs WHERE scope='team' AND owner_id=teamId`. Integrated
  into team management panel with period/group-by selectors.
- **Admin Reset Password** — Button in user list with vault destruction
  warning dialog (two-step confirmation).
- **User Usage tab** — Settings modal "Usage" tab shows personal token
  consumption and estimated costs across all conversations, including BYOK.
  Period selector (7d/30d/90d) and group-by (model/day).
- **Live Venice integration tests** — Gated behind `VENICE_API_KEY` env var.
  Tests: non-streaming completion, streaming completion, usage logging for
  both modes, pricing from catalog sync, and direct embeddings via BGE-M3.
  Uses `qwen3-4b` (Venice Small) — cheapest at $0.05/$0.15 per 1M tokens.
- **Migration 004** — `usage_log` table, `model_pricing` table, `model_roles`
  seed in `global_settings`.

### Fixed
- **UEK re-wrap on password change** — `ChangePassword` now decrypts UEK
  with old password and re-encrypts with new password + fresh salt. Previously,
  password changes silently broke all personal BYOK keys.
- **UEK destruction on admin password reset** — `ResetPassword` now nullifies
  vault columns, evicts UEK from cache, deletes personal provider configs,
  and logs an audit event. Previously, admin resets left orphaned encrypted
  UEK that silently broke personal keys.
- **Streaming completions logged zero tokens** — `StreamEvent` lacked token
  fields and providers discarded usage data from final chunks. Both OpenAI
  and Anthropic streaming parsers now capture and propagate token counts.
- **OpenAI streaming usage race condition** — The OpenAI protocol sends
  chunks in order: content → finish_reason → usage → [DONE]. The parser
  sent `Done=true` on the finish_reason chunk (step 2) before the usage
  chunk arrived (step 3, with `choices:[]`). The receiver returned
  immediately on Done, so usage was captured but never delivered.
  Fix: defer the Done event as `pendingFinish` until the usage chunk
  arrives, then flush with tokens attached. Tool-call finishes flush
  immediately (tool loop needs the event). Affects all OpenAI-compatible
  providers (OpenAI, Venice, OpenRouter).
- **Token accumulation across tool iterations** — `streamResult` in
  `stream_loop.go` now accumulates input/output/cache tokens across
  multi-tool-call iterations instead of only capturing the final iteration.
- **Admin pricing leaked BYOK entries** — `PricingStore.List()` returned
  all model_pricing rows including those from personal BYOK providers.
  Admin panel showed pricing entries for user-private providers they
  shouldn't see, with O(users × models) scale explosion. Now joins on
  `provider_configs` and filters `scope != 'personal'`. Admin `UpsertPricing`
  also validates provider scope, rejecting manual pricing on personal
  providers.
- **Team role handlers used wrong param name** — `ListTeamRoles`,
  `UpdateTeamRole`, `DeleteTeamRole` read `c.Param("id")` but routes use
  `:teamId`. Every team role operation silently got an empty team ID.
- **Usage logging suppressed for zero-token streams** — `logUsage` had an
  early exit when `inputTokens == 0 && outputTokens == 0`. Combined with the
  OpenAI streaming parser bug (below), this silently dropped all streaming
  usage rows. Removed the guard — requests are always recorded.
- **Live chat completion test used wrong field names** — `TestLive_VeniceChatCompletion`
  sent `messages` and `config_id` but the handler expects `content` and
  `provider_config_id`. Test silently skipped on the binding error.

## [0.9.4] — 2026-02-24

### Added
- **API key encryption (vault)** — Two-tier AES-256-GCM encryption for stored
  API keys. Global/team keys encrypted with env-var-derived key (HKDF-SHA256);
  personal BYOK keys encrypted with per-user encryption key (UEK) derived from
  password via Argon2id. Admin cannot recover personal keys without user's
  passphrase. New `server/crypto/` package: `vault.go`, `cache.go`,
  `resolver.go`, `backfill.go` with 11 round-trip tests.
- **UEK lifecycle** — UEK generated on registration, unwrapped on login
  (Argon2id → AES-GCM), cached in `sync.Map` for session duration, evicted
  with memory zeroing on logout. Pre-migration users auto-initialize vault
  on first login.
- **Migration 003_vault.sql** — Adds `encrypted_uek`, `uek_salt`, `uek_nonce`,
  `vault_set` to users table. Adds `api_key_enc` (BYTEA), `key_nonce`,
  `key_scope` to provider_configs. Backfills `key_scope` from existing scope.
- **Startup backfill** — `BackfillEncryptedKeys()` encrypts plaintext API keys
  on first startup with `ENCRYPTION_KEY` set. `EnforceEncryptionKey()` refuses
  startup if encrypted keys exist but env var is missing.
- **CI/CD: encryption secret** — `ENCRYPTION_KEY` Gitea secret synced to k8s
  `switchboard-encryption` secret. Backend manifest references it as optional
  `secretKeyRef`.

### Fixed
- **HTML code blocks render live in chat (XSS)** — When a model's `</think>`
  tag directly abutted a code fence (no newline), the fence wasn't recognized
  by marked.js, causing raw HTML to render as live DOM elements (canvas games,
  styles, etc.). Three-layer fix: (1) DOMPurify switched from permissive
  `ADD_TAGS` (default allows canvas, style, form, etc.) to strict
  `ALLOWED_TAGS` allowlist of only markdown-produced elements; (2) think-block
  placeholders padded with `\n\n` to ensure adjacent fences start on fresh
  lines; (3) unclosed code fences auto-closed before `marked.parse` for
  streaming protection.

## [0.9.3] — 2026-02-23

### Changed
- **Code blocks: button-driven collapse** — Replaced `<details>` wrapper with
  inline collapse/expand toggle button in the code toolbar. Auto-collapses at
  >15 lines with a fade mask; toggle available on all blocks >5 lines. User
  can always expand/collapse regardless of threshold. Smooth CSS transition
  instead of native `<details>` jump.
- **Thinking blocks: always visible** — Thinking blocks are always rendered in
  the DOM regardless of the `showThinking` setting. The setting now controls
  whether blocks start expanded (`<details open>`) or collapsed. User can
  always click to toggle. Setting label updated to "Auto-expand thinking blocks".

### Added
- **Admin default model** — New `default_model` policy in Admin → Settings.
  Dropdown populated from enabled models. When a user has no saved selection
  (fresh login, cleared browser) or their saved model is no longer available,
  the admin default is used before falling back to first visible. Resolution
  chain: `localStorage → admin default → first visible`.
- **Reasoning/thinking support for OpenAI-compatible providers** — Models that
  send `reasoning_content` in stream deltas (Grok, DeepSeek, etc.) now have
  thinking blocks streamed live and persisted as `<think>` tags. Renders as
  collapsible thinking blocks in both streaming and history views.
- **Favicon animation during generation** — Browser tab favicon pulses with
  cascading dot opacity animation while a completion is in progress, restoring
  to the static favicon when done.
- **Tool calls in message history** — Tool call metadata (name, arguments,
  result, error status) is now persisted alongside assistant messages in the
  database. History view renders tool calls as collapsed `<details>` blocks
  showing "🔧 tool_name → done" with expandable input/output JSON.
- **Notes tool: "View note" link** — When a notes tool (`note_create`,
  `note_update`, etc.) completes, a "📝 View note" button appears in both
  the live streaming tool indicator and the history tool call block. Clicking
  opens the Notes panel and navigates directly to the note.
- **Notes panel: Copy button** — "Copy" button added to note read mode,
  copies title + content as markdown to clipboard.
- **Brand hover: jitter-free crossfade** — Sidebar brand logo→collapse icon
  swap uses opacity crossfade instead of `display: none` toggle, eliminating
  layout reflow jitter on hover.

### Fixed
- **Model selector shows unavailable model** — Selector restoration searched
  all models including user-hidden ones. Saved selection (localStorage) for a
  hidden model like Claude Opus would re-select it on every page load even when
  only Grok was visible. Resolution now filters to visible models only.
- **Refresh toast shows wrong count** — "Loaded 33 models" counted all models
  including hidden; now shows only visible count ("Loaded 1 model").
- **Tool calls vanish after completion** — Live streaming tool indicators
  disappeared when `reloadActivePath()` rebuilt messages from DB. Fixed: the
  `getActivePath` query now includes `tool_calls` column, and `PathMessage`
  carries the data through to the frontend renderer.
- **Tool result "undefined results" text** — Operator precedence bug in tool
  result summary parser caused `undefined results` to display for note_create.
  Fixed summary logic to properly branch on title vs count.
- **Regenerate streams at wrong position** — Regen'd response appeared below
  the full conversation (appended at bottom) then snapped to correct position
  after completion. Fixed: display is now truncated to the parent message
  before streaming starts, so the new response streams in-place as a clean
  branch. Backend context was already correct (excludes old response).
- **Regenerate loses tools, reasoning, and tool execution** — Regen handler
  was a stripped-down copy of the completion handler missing tool definitions,
  tool execution loop, reasoning_content forwarding, and tool_calls persistence.
  Model lost access to note tools on regen and fell back to generic capabilities.
  Refactored: extracted `streamWithToolLoop()` into `stream_loop.go` as the
  single canonical streaming implementation. Both `streamCompletion` (normal
  chat) and `Regenerate` now call the shared function — only persistence
  differs. Future streaming features (new event types, tool capabilities)
  automatically apply to all code paths.
- **OpenRouter free models crash** — Free models with nil pricing caused
  `pq: invalid input syntax for type json` on catalog sync. Nil pricing now
  routes through `ToJSON()` producing `"{}"` instead of nil `[]byte`.
- **API key appears unsaved** — `ListGlobalConfigs` returned raw
  `ProviderConfig` structs where `APIKeyEnc` is `json:"-"` (never serialized),
  so `has_key` was always `undefined`. Now returns computed `has_key` field.
- **Case-insensitive usernames** — Login, registration, and all user lookups
  use `LOWER()` in SQL. All user creation paths normalize to lowercase.
  New migration `002_ci_username.sql` adds `LOWER()` unique indexes and
  normalizes existing rows.

## [0.9.2] — 2026-02-23

### Added
- **Collapsible code blocks**: Code blocks over 15 lines auto-collapse into
  `<details>` with language and line count summary. Language label shown in
  top-left corner of all code blocks.
- **HTML preview**: "Preview" button on HTML code blocks opens a sandboxed
  iframe (`allow-scripts`, no `allow-same-origin`). Auto-detected for
  untagged blocks via heuristic.
- **Token count estimate**: Live token counter below input area showing
  approximate tokens and context usage percentage when model has `max_context`.
- **Context length warning**: Dismissable banner above input at 75% (yellow)
  and 90% (red) context usage with guidance to start a new chat.
- **Proxy interception detection**: `_parseJSON()` checks Content-Type before
  `.json()` on all API paths including streaming. Typed `proxyBlocked` errors
  with proxy page title extraction and actionable splash messages.
- **Environment injection**: `window.__ENV__` wired through entrypoint, k8s,
  and index.html for dev/test/production gating.
- **Enhanced diagnostics**: `NET:PROXY` log type, Content-Type capture in
  fetch interceptor, environment info in export header and state snapshot.
- **Team admin audit scoping**: New `GET /api/v1/teams/:teamId/audit` and
  `/audit/actions` endpoints scoped to team members. Activity Log section
  in team manage panel with filter dropdown and pagination.
- **New favicon**: Switchboard panel design with provider-colored jacks.
  Animated SVG (rotating plugs), 32px PNG, 256px PNG, and ICO.
- **Seed users** (dev/test only): `SEED_USERS=user:pass:role,...` env var
  pre-creates active users on startup. Ignored in production. K8s secret
  `switchboard-seed-users` wired in backend manifest.

## [0.9.1] — 2026-02-23

### Removed
- **Static known model table**: Deleted `knownModels` map from backend and
  `KNOWN_MODELS` from frontend. The same model ID can have different capabilities
  depending on the provider (e.g. DeepSeek has tool_calling on OpenRouter but
  not on Venice). A hardcoded table can't represent this.
- **Frontend `lookupKnownCaps()`**: Removed client-side capability guessing.
  Backend is the sole source of truth via catalog → heuristic chain.

### Changed
- **Resolution chain simplified**: catalog (provider API sync) → heuristic inference.
  No intermediate known table. Providers that report capabilities via API are
  authoritative; heuristics are best-effort for unsynced models.
- **All providers updated**: OpenAI, OpenRouter, Anthropic, Venice now call
  `InferCapabilities()` directly instead of the dead known table lookup.
- **Frontend `resolveCapabilities()`**: Now passes through backend caps as-is.
  No client-side merging with a static table.
- [x] EXTENSIONS.md recovered into repo, updated with Appendix A (Custom
  Renderers) and Appendix B (Model Roles with utility/embedding/generation slots)
- [x] ROADMAP.md restructured: extension foundation pulled to v0.11.0,
  model roles to v0.10.0, dependency graph, TBD replaces post-1.0,
  removed v0.8→v0.9 migration (OBE — no public release, no test path)

### Added
- **Heuristic patterns**: Updated to detect qwen3, gpt-5, grok, kimi, minimax,
  glm-5, gemma-3 model families. Vision expanded to claude-opus/sonnet (not
  just claude-3). Reasoning expanded for thinking, grok, glm patterns.

### Fixed
- **Preset capability pills**: Presets with auto-resolve (no `provider_config_id`)
  now inherit base model capabilities via `GetByModelIDAny` catalog fallback.
- **Venice `optimizedForCode`**: Added mapping to `CodeOptimized` capability.
- **CI test stability**: BYOK journey tests use unreachable endpoints so auto-fetch
  doesn't race with simulated data injection.

## [0.9.0] — 2026-02-22

### Added
- **Schema consolidation**: 21 migrations collapsed to single `001_initial.sql`
- **Store layer**: All database access through typed interfaces (no raw SQL in handlers)
- **Persona model**: Trust-boundary model replacing old presets; scoped global/team/personal
- **Capabilities resolver**: Three-tier chain — catalog → known table → heuristic inference
- **Three-state model visibility**: enabled / disabled / team-only
- **BYOK auto-fetch**: Creating a personal provider triggers model discovery from provider API
- **User model refresh**: `POST /api-configs/:id/models/fetch` endpoint + UI button
- **Composite model IDs**: `configId:modelId` format prevents cross-provider collisions
- **Audit log foundation**: All admin operations logged with actor, action, resource
- **Journey integration tests**: API-driven test suite replacing fake-data tests
- **Frontend test suite**: 107 tests, 27 suites validating model processing pipeline
- **Live Venice API test**: Proves real BYOK → auto-fetch → models visible flow

### Fixed
- **API key storage**: `json:"-"` tag on `ProviderConfig.APIKeyEnc` silently dropped
  keys during admin create/update. Fixed with wrapper structs that bypass the tag.
- **NULL model_default scan**: `scanProviders()` crashed on NULL `model_default` column,
  silently hiding all team and BYOK models. Fixed with `sql.NullString`.
- **Nil slice serialization**: Go nil slices serialized as JSON `null` instead of `[]`,
  breaking frontend fallback chains. Fixed with `make([]T, 0)`.
- **Frontend error swallowing**: API responses with `errors` field were silently ignored.

### Changed
- Backward-compatible API routes with v0.8 field name aliases
- User model preferences table (`user_model_settings`)