What Rose, Ada and Vi Are Each Good At
realvco pre-installs OpenClaw in the Rose container (OC-1) and Hermes-Agent in the Ada (HM-2) and Vi (HM-3) containers. All three containers are wrapped in realvco’s custom management system and interface (admin-panel), so you never touch installation, configuration, or upgrades.
These two AI agent tools are good at different things:
- OpenClaw (Rose) — deep reasoning, cross-tool chaining, brand voice
- Hermes-Agent (Ada / Vi) — fast, low cost, broad messaging platforms, self-maintaining skill library
Below: what Rose, Ada and Vi are each best suited to handle.
One-Line Summary
OpenClaw is like a senior consultant — thinks deeply, gets it right, but each answer is more expensive and slightly slower.
Hermes is like an efficient front-desk agent — fast and cheap, perfect for handling lots of repetitive questions.
Full Comparison
| Dimension | OpenClaw (Rose) | Hermes-Agent (Ada / Vi) |
|---|---|---|
| Pre-installed tool | OpenClaw (on Claude Agent SDK) | Open-source Hermes-Agent |
| Reasoning depth | Deep reasoning | Solid for everyday |
| Response speed | Medium | Fast |
| Token cost | Medium-high | Low |
| Cross-tool chaining | Strong (many tools in one go) | Medium |
| Messaging platform breadth | Many | Widest (22 platforms) |
| Skill-library upkeep | Manual | Self-curating (Curator) |
| Chinese ability | Strong | Medium-strong |
| English ability | Strong | Strong |
| Best for | Deep reasoning, long conversations, cross-tool, brand voice | High-volume, platform breadth, cost efficiency, English |
| Default models | 4 models + aliases (gm / ds / hk / gf) — details | See the pre-installed model list |
Rose comes with OpenClaw pre-installed, Ada / Vi with Hermes-Agent, all managed uniformly by realvco’s management layer. OpenClaw is strong at deep reasoning and cross-tool chaining; Hermes-Agent is broad on platforms and low-cost. The sections below cover what each is best suited to handle.
What’s New in Hermes (updated through v0.14.0)
Hermes-Agent (pre-installed in Ada / Vi) keeps shipping customer-visible capabilities on top of the existing “cheap, fast” baseline (realvco upgrades you automatically; the examples below highlight Vi, but Ada gets the same capabilities):
1. Self-improving skill library (Curator)
Hermes now maintains its own skill library in the background — by default once every 7 days:
- Grades which skills are useful and which aren’t
- Merges duplicates
- Prunes skills that haven’t been used in a long time
What you feel: you don’t have to manually clean up Vi’s settings — it gets sharper on its own.
2. 22 messaging platforms (now including Microsoft Teams, Google Chat, LINE)
Vi’s messaging platform list keeps growing — now 22 platforms. Added progressively since v0.12:
- Microsoft Teams — the natural fit for enterprise customers’ internal collaboration
- Tencent Yuanbao — the collaboration platform for serving customers in Greater China (mainland / HK / Macau)
- Google Chat — internal collaboration for Google Workspace enterprise customers (added in v0.13)
- LINE — the dominant messenger in Taiwan / Japan / Korea; especially valuable for customers in those markets (added in v0.14)
- SimpleX Chat — privacy-first, decentralized messaging (added in v0.14)
Setup: Messaging — Microsoft Teams and Tencent Yuanbao; the rest are connectable in platform connection settings.
3. Native Spotify integration
Vi can now drive Spotify directly with 7 built-in tools:
- Play / pause / skip
- Search songs, artists, albums
- Queue tracks
- Switch the active playback device (phone / computer / speaker)
- Manage playlists
First-time setup runs an interactive wizard. Once you log into your Spotify account, it just works.
4. Google Meet integration
Vi can join Google Meet meetings and do four things:
- Join the meeting (no need to invite manually)
- Live transcription (full meeting transcript)
- Speak on your behalf (with your authorization)
- Post-meeting follow-up (key points + action items)
Use cases: internal meeting assistant, sales call follow-ups, training session notes.
5. 57% faster cold start
Bringing Vi up into interactive mode is roughly twice as fast as the previous version. The improvement is most noticeable when entering Vi through the admin panel’s built-in web terminal.
6. Major TUI upgrade
Inside admin panel → hm-3 container → Terminal button, you’ll find:
- LaTeX math rendering (integrals, fractions, etc. display nicely)
/resumeto switch history (arrow keys to pick, pressdto delete)/reloadenv hot-reload (changed .env? no need to quit)- Mouse-wheel scrolling (scroll TUI content directly)
- Auto-resume last session (TUI picks up where you left off)
Full guide: Hermes TUI via web terminal.
Breaking-ish: starting in v0.12, secret redaction is off by default (it used to be on). That means logs may contain raw secret-like strings (API keys, token-shaped values). Don’t share raw logs with outsiders as a precaution.
What Rose Is Best At
These needs run more smoothly on Rose (OpenClaw):
- Complex reasoning: a customer asks about returns, but order status, membership tier, and active promos all interact — you need synthesis, not lookup
- Cross-tool workflows: one request that has to query the database, send a LINE message, and log to Notion all at once
- Long conversations: 8 back-and-forth turns where everything earlier in the thread matters
- Brand voice: support replies need to match your company’s tone, every sentence polished
- Deep copywriting: ad copy, product descriptions, social posts where the AI actually has to think before writing
- System management: scripting, running system commands, touching host files (Rose ships with Host Root)
If your business is e-commerce support that needs deep judgment, knowledge services, consulting, or the deep parts of training — let Rose take it.
What Ada and Vi Are Best At
Ada and Vi both come with Hermes-Agent pre-installed. These needs are a better fit for them:
- High volume, repetitive questions: thousands of FAQ, order lookup, or address confirmation queries per day where the answers barely change
- Cost-sensitive: you want to keep the monthly token bill below a target — Hermes-Agent is noticeably cheaper per call than OpenClaw
- Fast response required: users expect a reply in 1-2 seconds (live chat, voice-driven backends)
- English-first audience: most of your users speak English, so Hermes’s English ability is plenty
- Many messaging platforms: Hermes supports 22 platforms (incl. Microsoft Teams, Tencent Yuanbao, Google Chat, LINE, SimpleX Chat) — that’s Vi’s signature strength
- Want a self-maintaining skill library: Hermes’s Curator tidies the skill library automatically, one less thing to manage by hand
Ada vs Vi — how to split: both come with Hermes-Agent and have similar capabilities. A common split is “Ada handles daily support and high-volume conversation; Vi focuses on multi-platform reach and skill-library curation” — or you can run the same kind of work in parallel on both (they don’t interfere).
Rule of thumb: the more of “high volume, fast, cost-sensitive, English” applies, the better the fit for Ada / Vi; “needs many messaging platforms” → Vi first.
What’s Pre-installed on Each Companion
realvco sets this up for you — nothing to install or pick (this is the default for new installs and reinstalls on or after 2026-05-26):
- Rose (
oc-1) container — OpenClaw pre-installed - Ada (
hm-2) container — Hermes-Agent pre-installed - Vi (
hm-3) container — Hermes-Agent pre-installed
All three containers are wrapped in realvco’s custom management system and interface (admin-panel). All three companions run at once, each doing what it’s best at.
Switching the underlying model (e.g. Rose from Claude Sonnet to Opus) is a different setting and is adjustable. See Token Pricing & Model Selection. The pre-installed tool and the model are two separate layers.
Mixing All Three Companions
The most common pattern is “Ada / Vi up front, Rose in the back for deep tasks” — keep costs down without giving up quality.
Pattern 1: English FAQ With Auto-Escalation
English customer message
↓
Vi (Hermes) handles first — about 80% of FAQ (shipping, returns, payment methods)
↓
Vi can't answer / customer is upset / refund amount involved
↓
Hands off to Rose (OpenClaw) for deeper reasoningWhy split it this way: FAQ answers are basically the same every time, so Vi is enough — paying for OpenClaw on every one is wasteful. But for cases that actually need judgment (emotion, money, policy exceptions), let OpenClaw take over.
Pattern 2: Daily Volume to Ada / Vi, Deep Tasks Escalate to Rose
Daily support traffic → Ada (Hermes) — high volume, fast replies, low cost
Multi-platform traffic → Vi (Hermes) — Telegram / Teams / LINE in parallel
Deep reasoning / cross-tool / system management → Rose (OpenClaw)Why split it this way: Ada and Vi both come with Hermes-Agent — great for volume. Rose stays for cases that really need deep judgment.
Pattern 3: Vi Pre-Filters, Rose Closes
Let Vi (or Ada) handle simple price checks and order lookups. Only escalate to Rose when a decision is needed (give a free gift? approve an out-of-policy refund?).
How Do I Configure These Splits?
The routing logic isn’t set in admin-panel directly — it lives in each companion’s conversation rules. In each companion’s Settings sub-tab, you tell them “if you see X, hand off to Y”.
Setup details: Agent Settings and Meet Your Three Companions.
How Do I Know Tasks Are Going to the Right Companion?
Some signs to watch for:
| Signal | What it might mean |
|---|---|
| Token bill is much higher than expected | Too many simple questions are going to Rose (OpenClaw) — move some to Ada / Vi |
| Answer quality is unstable, off-topic | Complex questions are being handled by Ada / Vi — escalate to Rose |
| Customers complain about slow replies | Speed matters here — consider putting Ada / Vi on the front line |
| Multi-turn chats keep “forgetting” earlier context | Long conversations on Ada / Vi — switch to Rose |
Every so often, check the cost trend in the Usage sub-tab and combine it with customer feedback to keep tuning who handles what.