One of the core ideas behind Xeref is that agents should be able to work together — not just respond to prompts in isolation. v3.7 introduces the infrastructure that makes that possible.
v3.7 — XerefHermes Inter-Agent Message Bus
XerefHermes is a typed message bus for routing envelopes between system agents. Any server-side code can publish to the Hermes queue; registered handlers process messages asynchronously via a Vercel cron that runs every minute.
The first live handler is memory:save — it publishes a payload to the Pinecone xeref_user_memory namespace, making any information Xeref processes searchable in long-term memory.
A Hermes Inspector is available in the Code tab → AI Agents → XerefHermes sidebar. It shows queued/processing/done/error message counts and a live log of the last 50 envelopes. The XerefHermes System Agent is also now selectable in the chat input for building on top of the envelope schema.
This release also fixed a chat infinite-loop bug: an errant activeChat dependency in a sync effect was causing repeated setActiveChat calls on every message submit. Fixed by reading through a ref.
v3.8 — Navigation Performance
- Single round-trip view switches: Dashboard navigation now triggers one server call instead of one per data source. Auth verification happens once per navigation — cuts Supabase Auth API calls by ~75%.
- View data cache: Switching back to a previously visited view (Tasks, Stats) is now instant. Cached data renders immediately while a background refresh keeps it current.
- Shared public page layout: About, Changelog, Pricing, FAQ, Privacy, Terms, and Docs now share a sticky header and footer via a shared route group — per-page redundant nav removed.
v3.9 — Claude Opus 4.8
Anthropic's newest flagship is now the top Ultra model. Opus 4.8 brings better judgment and code review (~4x less likely to miss flaws), stronger agentic decision-making, and improved alignment at the same price as 4.7. Opus Plan Mode plans with 4.8 too.
These three releases together make Xeref faster to use, more capable in reasoning, and more powerful under the hood as a multi-agent platform.