Workavera v0.0.7 — MCP access, tool approvals, and inbox management

Highlights

🔌 Connect any MCP client to your workspace

Workavera now speaks MCP. Point Claude Code, Cursor, or any Streamable
HTTP MCP client at /api/mcp and it gets the full set of assistant
tools — boards, calendar, docs, reading list, and contacts — scoped to
your account.

• Create sk-wa- API keys from Settings → API Keys, with optional
  expiration. The full key is shown exactly once, together with a
  copy-paste-ready MCP client configuration; only a SHA-256 hash is
  stored on the server. Revoke a key any time by deleting it.
• Destructive operations (deleting tasks and calendar events) are off
  by default. A key only exposes those tools if you explicitly opt in
  at creation — and the UI makes very clear what that means.

✅ Approval cards for destructive chat actions

When the assistant wants to delete a task or calendar event in chat, it
now pauses and streams an approval card showing exactly what will be
removed (with a warning for recurring series). The run resumes after
you approve, and rejections are recorded without executing the tool.

🔔 Notifications & Reading, managed properly

Both inboxes gain search, filters, pinning, mark-all-read, paginated
active/archived lists, restore, and permanent delete.

Improvements

• Document lists fetch metadata only and load full content on open.
• Model settings use shared token-size controls with compact k/M
  values, presets, and custom sizes.
• The conversation list shows a realtime Responding indicator for
  every active assistant stream, including background conversations.
• Notifications and Reading standard operations now use PocketBase
  collection CRUD APIs directly.

Fixes

• HTML document anchor links stay inside the rendered document.
• Bell-dropdown items no longer show the browser focus outline.
• Internal one-shot text generation uses the streaming transport,
  avoiding provider rejections on large output limits.

⚠️ Breaking change

The deprecated ai_micro_apps collection and all remaining AI Micro
Apps records and files are permanently deleted by migration,
completing the removal announced in 0.0.6. Export anything you still
need from the PocketBase admin UI before upgrading — after the
migration runs there is no way to recover it.

Full changelog: CHANGELOG.md

Workavera v0.0.6 — Docs absorb AI Micro Apps, context compaction & usage ring

⚠️ Workavera is early-stage software (0.0.x). Features and data schemas are still changing quickly, and releases may include breaking changes. Not recommended for production use yet.

⚠️ Breaking: AI Micro Apps module removed

The AI Micro Apps module has been absorbed into Docs. Its UI, API routes (/api/ai-micro-apps/*), and AI tools (microapps_*) no longer exist.

Your data is NOT deleted in this release. The ai_micro_apps collection and its stored HTML files remain in pb_data/ as a recovery buffer, but they are no longer reachable from the app.

⏰ The collection and all remaining Micro Apps data will be PERMANENTLY DELETED in v0.0.7. To keep an app, before upgrading past this version:

1. Open the PocketBase admin UI (/_/) and browse the ai_micro_apps collection.
2. Download each record's html_file.
3. Re-create it as an HTML document in Docs (paste the source in the HTML source view), or keep the file.

---

v0.0.5 was tagged but not published as a release, so this page covers both versions. If you are upgrading from v0.0.4, you get everything below.

v0.0.6 highlights

Docs: interactive HTML documents

• Docs now have a kind: markdown (default) or html. HTML documents hold a self-contained interactive page rendered in a sandboxed preview (scripts run in an opaque origin with no access to your session), with source editing in CodeMirror, raw .html export, versions, project sharing, pins, and conflict detection.
• New docs_write_chunk AI tool writes long content in pieces; a whole chunked session records a single version.
• The AI assistant now registers 24 tools instead of 30 — the seven microapps_* tools were replaced by the extended docs tools.

Onboarding

• Fresh installations seed a demo user (demo@workavera.local / workavera) when the users collection is empty, so you can sign in without touching the PocketBase admin first. Change the password before exposing the server.

Context ring fixes

• When a provider reports no input usage (e.g. GLM's Anthropic-compatible endpoint always returns input_tokens: 0), the context size falls back to a character-based estimate — the ring stays truthful and auto-compaction still triggers. Estimated values are marked with a ~ prefix.
• The ring hover was reworked: cache hit/write, the compaction threshold ("Compacts at"), and the conversation's accumulated input/output totals.

v0.0.5 highlights

• Model context window setting. Each model configuration has a "Max context" size (presets 32k–2M or a custom value); existing models migrate to 256k.
• Automatic context compaction. When a conversation exceeds 75% of the model's window, older turns are compacted into a running summary (the newest four user turns stay verbatim). Stored chat history is never modified, and the transcript marks where compaction happened. The fixed 30-message history window was removed, so provider prompt caches stay warm across turns.
• Context-usage ring. The chat input shows a ring next to the model selector with live token and cache details.
• Upgraded Fantasy to 0.37.3 and Go to 1.26.5.

Install / upgrade

docker run -p 8090:8090 -v workavera-data:/app/pb_data ghcr.io/xusenlin/workavera:0.0.6

Or download the archive for your platform below and verify it against SHA256SUMS.txt. Migrations run automatically on first start.

Full changelog: CHANGELOG.md · v0.0.4...v0.0.6

Workavera v0.0.4 — Improved Docs and Board Experience

Workavera v0.0.4

This release introduces a significantly improved document editing experience and adds personalized Board project state management.

✨ New Features

• Added slash commands, drag handles, a “Turn into” menu, and a floating formatting toolbar to the document editor
• Added access-controlled image and file uploads, duplicate reuse, and persistent attachment cards
• Uploaded images are now embedded in HTML exports
• Documents can now be exported as Markdown or standalone HTML files
• Board projects can be reordered per user without affecting other members
• Expanded and collapsed Board project states are now saved independently for each user, with multiple projects allowed to remain open

🔄 Improvements

• Migrated the document editor to BlockNote while preserving Markdown storage and versioning
• Replaced document header actions with streamlined icon buttons
• Code blocks now follow the application’s light or dark theme
• Added an Intel macOS (darwin/amd64) archive to task release

📦 Downloads

Choose the archive for your operating system and architecture:

• Linux x64: workavera_0.0.4_linux_amd64.tar.gz
• macOS Intel: workavera_0.0.4_darwin_amd64.tar.gz
• macOS Apple Silicon: workavera_0.0.4_darwin_arm64.tar.gz
• Windows x64: workavera_0.0.4_windows_amd64.zip

Use SHA256SUMS.txt to verify the integrity of downloaded files.

Full Changelog

v0.0.3...v0.0.4

v0.0.3

Highlights

Self-contained single binary. The compiled frontend is now embedded into the
binary via go:embed — download one file, run it, done. No separate
frontend/dist directory is needed at runtime.

73% smaller downloads. Release assets are now compressed archives (~19 MB,
down from a ~70 MB raw binary). Under the hood the frontend bundle went from
28 MB to 11 MB by deduplicating shiki (it was bundled twice at v3 and v4) and
shipping a curated syntax-highlighting set: 28 common languages with GitHub
light/dark themes. Code blocks in other languages render as plain text.

Apache License 2.0. The project is now formally licensed.

Downloads

Platform	File
Linux (x86_64)	workavera_0.0.3_linux_amd64.tar.gz
macOS (Apple Silicon)	workavera_0.0.3_darwin_arm64.tar.gz
Windows (x86_64)	workavera_0.0.3_windows_amd64.zip

Verify with SHA256SUMS.txt:

shasum -a 256 -c SHA256SUMS.txt --ignore-missing

## Quick start

tar -xzf workavera_0.0.3_linux_amd64.tar.gz
./workavera serve --http=0.0.0.0:8090

Open http://127.0.0.1:8090/, create the first superuser via the Admin UI at
/_/, then add a model configuration in Settings before using Chat.
Data is stored in ./pb_data. Docker images remain available at
ghcr.io/xusenlin/workavera.

## Other changes
fix(board): expose state sortOrder to AI tools
docs: license badge, NOTICE file, chat screenshot