v3.2.3

Release v3.2.3

• recover with partial answers or clarifying follow-ups after upstream stream failures when tool evidence already exists
• classify gene, transgene, and expression requests as gene-expression analytics before connectivity
• bias broad transgene-expression requests toward short representative lists instead of long tool loops

v3.2.2

Release v3.2.2

• restore production compatibility when APPROVED_ELM_* values are not explicitly configured
• add partial-summary and clarification fallbacks for broad or tool-heavy queries instead of returning an empty failure
• raise the tool-round budget to 10
• add a generator for named example conversations and screenshots under lib/examples/generated

VFB Chat v3.2.1

VFB Chat 3.2.1

This release expands approved site coverage so the assistant can answer a wider range of VFB- and NeuroFly-related questions without falling back to general web search.

Highlights

• Expands the approved site search path beyond a hand-maintained page list.
• Adds a wider server-side, domain-restricted discovery flow using sitemap and robots sources for approved sites.
• Introduces approved page extraction so the assistant can read the page it found when it needs details such as dates, programme information, venue, or registration guidance.
• Adds neurofly.org to the approved search and outbound allow-lists.
• Keeps the governance model intact by preserving hard domain restrictions and avoiding unrestricted browsing.

User-facing improvements

• Better coverage for VFB news, blog, workshop, conference, and event questions.
• Better support for NeuroFly queries.
• More useful answers when a page needs to be read after search, rather than only linked.

Operational notes

• Default discovery and allow-list settings now include NeuroFly.
• The release version is now 3.2.1 across the application metadata.

VFB Chat v3.2.0

VFB Chat 3.2.0

This release prepares VFB Chat for a governance-conscious production launch with stronger privacy controls, safer search behavior, and clearer user feedback options.

Highlights

• Introduces a governance-ready logging model with /logs volume support for security, analytics, and feedback data.
• Removes routine free-text analytics logging and strengthens retention boundaries for security and service metrics.
• Replaces unrestricted browsing with reviewed-domain search controls and outbound link allow-listing.
• Adds structured thumbs up/down feedback with reason codes, plus optional conversation attachment for problem investigation when a user explicitly chooses it.
• Publishes updated privacy messaging in the app and a dedicated VFB Chat privacy notice page.
• Enforces production fail-closed checks so deployments only run against the approved ELM endpoint and model configuration.

User-facing improvements

• Clearer privacy messaging in the chat experience.
• More consistent, safer links in assistant responses.
• Better issue reporting through structured response feedback.

Operational updates

• Docker deployments now mount a dedicated /logs volume.
• Security logging, blocked-domain auditing, and aggregated analytics are separated more cleanly for governance and retention control.
• The release version is now 3.2.0 across the application metadata.

v3.1.1

Full Changelog: v3.1.0...v3.1.1

VFBchat v3.0.4

What's Changed

Link Display Fixes

• Fixed dark blue links — all links now consistently use the readable cyan #66d9ff colour against the dark background. Root cause was an override in ChatMessage that replaced the custom renderLink component with an unstyled <a> tag.
• FBrf IDs auto-link to FlyBase — bare FBrf0262545 references in responses are now automatically converted to clickable FlyBase links, alongside the existing VFB_ and FBbt_ auto-linking.
• All external links open in new tabs — verified target="_blank" on all non-query links.

Citation Artifact Cleanup

• Stripped OpenAI Responses API citation artifacts — removes citeturn0search0, citeturn0vfbsomething, and Private Use Area Unicode bracket delimiters (U+E200–E2FF) that were leaking into displayed text.

Prompt Improvements

• Simplified image/link formatting instructions — replaced overly prescriptive DISPLAYING IMAGES prompt (which caused the LLM to output literal text like "Thumbnail:" and "link to VFB page") with concise guidance.
• Better citation formatting — FlyBase references now use "Author et al., 2025" as clickable link text instead of showing bare FBrf IDs. Same for DOI and PubMed links.

Full Changelog: v3.0.3...v3.0.4

VFBchat v3.0.3

Fix thumbnail handling by avoiding linkification of IDs inside existing URLs. See RELEASE_NOTES.md for full history.

VFBchat v3.0.2

Improve linked ID display by using preferred labels (e.g., 'ME on JRC2018Unisex' instead of raw IDs), plus bugfixes.

VFBchat v3.0.1

Add VFB report link generation and improved term linking (auto-link VFB IDs/terms to https://virtualflybrain.org/reports/). Updated to show preferred labels for linked IDs (e.g., ME on JRC2018Unisex).

VFBchat v3.0.0

Improved MCP tool handling, parallel tool calls, and default GPT model selection. See RELEASE_NOTES.md for full history.