v2026.5.16 — The Foundation Release — Laying the foundation
Released: May 16, 2026 Theme: Slim and solidify the foundation — faster startup, smaller install, sturdier skeleton Breaking-ish change: Heavy backends now install on first use (see Upgrade Recommendations)
This release rolls up 808 commits across 1,393 files. The headline for you: Vi got lighter and starts faster, while also gaining a fresh batch of useful AI models and platforms.
Core Highlights
1. xAI Grok, Directly Wired In (SuperGrok OAuth)
New xAI Grok provider, authenticated via SuperGrok OAuth (browser login — no manual API key pasting). grok-4.3 is bumped to a 1M context window — far more content fits in one pass, great for handing it whole long documents or entire conversation histories.
2. Local OpenAI-Compatible Proxy (hermes proxy)
This release adds hermes proxy: it wraps your Claude Pro / ChatGPT Pro subscription as a local OpenAI-compatible endpoint, so tools that only speak OpenAI format — Codex, Aider, Cline — can run on the subscription you already pay for.
In plain terms: you don’t have to buy separate OpenAI API credits just to use Codex — lend it your existing Pro subscription through hermes proxy.
3. x_search — X / Twitter Search Tool
New x_search tool — Vi can search X (Twitter) content directly.
Use case: “Check what people on X have been saying about our competitor the last two days and pull three worth flagging.”
4. Microsoft Teams — End-to-End
Teams graduates from the earlier preliminary plugin to a full end-to-end integration — receiving, replying, media, and pipeline are all complete. Teams running Microsoft 365 can now treat Vi as a proper team member.
5. ⚠️ Breaking-ish: Debloating
This release slims Vi down significantly:
| Change | In plain terms |
|---|---|
| On PyPI | You can now pip install hermes-agent — more standard |
| Heavy backends are lazy-install | Bulky backend packages no longer all install upfront — they install the first time you actually use them |
| Smaller image | The container image is noticeably smaller — faster to download and deploy |
Heads up for realvco customers: heavy backends now install “on first use.” That means the first time you use a given heavy feature, there’s a one-time wait while it downloads and installs (first time only). After that it’s as fast as before. It’s a “wait once, lighter and faster every day” trade-off.
6. ~19s Faster Cold Start
Building on the previous slim-down, cold start is ~19 seconds shorter again. Customers entering Vi from the admin panel web terminal will feel it directly.
7. Browser CDP — Up to 180x Faster
Browser automation moved to CDP (Chrome DevTools Protocol) under the hood — related operations are up to ~180x faster. A big lift for workflows where Vi drives the web heavily (scraping, auto-filling forms).
8. LINE + SimpleX Chat (22 Platforms Total)
| New platform | Notes |
|---|---|
| LINE | Especially relevant for Taiwan / Japan / Korea customers — support and notifications can run over LINE |
| SimpleX Chat | Privacy-focused, decentralized messaging protocol |
9. Other Notable Items
- 1-hour cross-session Claude prompt cache: the cache now persists across conversations for 1 hour, saving token cost on multi-turn / multi-session work
/handofftruly transfers a session: hands off the entire session with its context to another agent / person — not just a pasted summary- LSP semantic diagnostics: for coding workflows, Vi uses a language server to understand code structure, giving sharper feedback
- 9 new optional skills: more out-of-the-box capabilities
Upgrade Recommendations
- Understand the first-use install behavior: after upgrading, the first use of certain heavy features adds a one-time download wait (first time only) — this is expected, in exchange for being lighter and faster day to day
- Grok users: run setup via SuperGrok OAuth and try
grok-4.3’s 1M context with long documents - Codex / Aider / Cline customers: try
hermes proxyto point those tools at your existing Claude Pro / ChatGPT Pro subscription and save a set of credits - LINE customers (Taiwan / Japan / Korea): wire up LINE so support and notifications live in the tool customers use most
- realvco customers: enter Vi from the admin panel web terminal to feel the ~19s faster cold start
- Long-conversation / multi-session workflows: lean on the 1-hour cross-session prompt cache for cost savings