Weave 0.9.16 Release Notes
New Features
MCP Tool Integration
Weave now supports the Model Context Protocol. Connect external tool servers (Slack, Notion, Linear, Monday, Atlassian, Asana, Discord, and more) and the AI can use their tools mid-conversation. A built-in catalog lets you set up servers with one click, and OAuth-based authentication is handled automatically. Each server's tools can be enabled or disabled individually from the new AI Tools settings page. The activity timeline shows which tools the AI used, with the actual server logo and name rather than a generic icon.
Redesigned Settings
The old single-page settings dialog has been replaced with a two-panel layout: a collapsible navigation sidebar on the left and the settings content on the right, opened from your avatar in the sidebar. Sections are organized into General, Account, AI Setup, Capabilities, Privacy, Explore, and Maintenance.
Parallel Web Search
When you ask a multi-topic question, the search planner can now fire 2–4 queries simultaneously instead of running them one at a time. Results are merged and deduplicated automatically. This significantly reduces wait times for complex searches.
Provider Fallback on Rate Limits
When a cloud provider returns a rate limit error after exhausting retries, Weave now automatically falls back to the next enabled provider in your configuration. Messages are re-truncated to fit the fallback model's context window. At most one fallback attempt is made, keeping total API calls bounded.
Improvements
Sustainable Chat Mode. "Eco Mode" has been renamed to "Sustainable Chat Mode" across the entire app, with a new Trees icon.
Cloud streaming resilience. Cloud model streaming now retries automatically (up to 2 times) on transient errors like 502, 503, and 429 with exponential backoff. Errors that previously failed silently are now logged and the response is properly terminated.
Sidebar new-item buttons. Every sidebar history section (Chat, Images, Transcriptions, Documents, Creations, and Analyze) now has a quick-create button. Unsaved changes are guarded before navigating away.
Creation search. Creations are now included in sidebar search results.
Sidebar auto-collapse. The sidebar collapses automatically when you open a file, URL, or attachment viewer in chat, and restores when you close it.
Active mode indicators. The extras selector shows lit-up icons when Sustainable Chat, Deep Research, or skills are active. A count badge appears when multiple skills are attached.
Unified file access. User-attached files and AI-created files now share a single workspace directory, so the AI can list and re-read your attachments mid-conversation.
AI SDK upgraded. The Vercel AI SDK has been upgraded from v5 to v6, along with React 19.2 for peer dependency compatibility.
Bug Fixes
Fixed prompt input paste handling. Resolved pill cursor positioning and text overflow issues when pasting into the prompt bar.
Fixed error messages for non-admin users. Local model fetch errors and an error boundary reload loop have been addressed.
Fixed skill editor. Website folder targeting and reference ID stability issues resolved.
Fixed updater blocking. Child processes no longer prevent the updater from proceeding.
Fixed Ollama model detection. Ollama-backed main models are now correctly detected during the readiness scan.
Sanitized file-not-found errors. Error messages no longer leak internal file paths.
MCP data cleared on sign-out. MCP server data is wiped when you sign out or delete your account.
