BRAIN / SUGGESTIONS, NOT SILENT ACTIONS01 / KEEP CONTEXT
A Brain that keeps the next move visible.
Keep goals, ideas, and recent work in one reviewable workspace. Suggestions stay suggestions until you choose what to do next.
IDLE HOPS · HOVER FOR HAPPY · CLICK TO GREET
Robot Orchestra is ready.
EDGE | AGENCY / AI AUTOMATION SPECIALISTS
A whole AI team.
Right on your desktop.
Free, privacy-conscious research and drafting for Windows, orchestrated by a Coordinator that directs Scout, Builder and Reviewer. Use Ollama or OpenRouter directly; Hermes is optional for tool-enabled work. The app never downloads models or sends prompts automatically.
See all 0.7.2 downloads01 Auto-connect to an installed local Ollama model02 No model download or prompt is automatic03 Windows x64 + ARM64 installers04 Browser extension is a separate ZIP
01 LOCAL OLLAMA AUTO-CONNECT · V0.7.2
Use a model you already have.
Keep inference on your PC.
On desktop, the 0.7.2 checks for installed Ollama models at startup and selects an available model when auto-connect is on. You can switch the option off. It never installs a model or submits a prompt on its own. If no local model is installed, choose another direct provider such as OpenRouter; existing model choices are preserved when available. Hermes is not a fallback.
Direct local inference. Select an installed Ollama model under Task Model. Ollama runs at
Already use Hermes? See the optional setup
127.0.0.1:11434; this route bypasses the separate Hermes bridge.LOCAL OLLAMA SETUP
ON DEVICEThree steps to your first model.
- 01Install and open Ollama
Use the official Windows installer, then leave Ollama running on this PC.
Install Ollama ↗ - 02Install one model sized for your PC
Choose a model from the Ollama library. In PowerShell, use
Browse models ↗ollama pull <model-name>and wait for the download to finish. - 03Let the desktop app find it
With auto-connect enabled, the 0.7.2 checks for installed models on startup and selects one that is available. Turn the setting off whenever you prefer to choose manually.
No Hermes adapter or model API key is needed for this direct local route. Model requirements depend on your hardware and the model you choose.
QUICK RECOVERY
Model not showing up?
- Check Ollama is open. Install it for Windows and start it before detecting models.
- Check that a model is installed. Run
ollama list; if the list is empty, choose a model from the Ollama library and install it. - Detect again. Click Detect local Ollama. If no model appears, confirm it is installed and Ollama is running, then retry.
02 A QUIETER WAY TO WORK
Stay in the loop.
Keep the final say.
Research notes, drafts, and suggestions stay visible and reviewable. Start with local Ollama; choose the optional Hermes route only when you want delegation.
BRAIN / SUGGESTIONS, NOT SILENT ACTIONS HOVER TO WAKE · CLICK TO GREET
02 / STAY CLOSE
Your companion, one click away.
Open the draggable desktop head and compact chat when you need them. Keep it out of the way when you don’t.
YOUChoose the task
HHermesdelegated route
LOllamadirect local route
03 / CHOOSE THE ROUTE
Direct model routes. Hermes when you choose.
Run inference through Ollama on your PC or OpenRouter directly with your configured key. Choose Hermes only when you explicitly need its optional tool-enabled route. These are separate routes.
03 INSIDE THE DESKTOP APP
A real look at
the workbench.
Authentic Robot Orchestra screens—not concept-only UI. New in 0.7.2: the approval pop-up — a plan big enough to need your go-ahead asks on screen, with its item count and call budget, and its Accept / Decline lands in Runs. New in 0.7.1: the Runs workspace, where every composed run shows its results and files, and the Brain Graph, a live map of how your Brain connects. Local inference is the recommended first route, with Hermes available when you choose delegation.
Overview / Connect HermesSee the team, connections and task activity at a glance.
DASHBOARD
Compact companionChat, tasks and route selection.
Brain workspaceGoals and suggestions for you to review.
Runs & resultsEvery fan-out and batch run — results, files, budget. Copy anything, export the whole run.
NEW IN 0.7.1
Brain GraphHow the Brain connects — agents, documents and goals on one live map.
NEW IN 0.7.1
Approval pop-upA plan big enough to need your go-ahead asks on screen first. Nothing runs until you accept.
NEW IN 0.7.2
The teamOne coordinator and focused specialists, ready to assign.
Runs, completedA finished fan-out: every result, its file, and the budget it used.
Brain Graph, currentHow the Brain connects — the live map in 0.7.2.
04 0.7.2 · FREE WINDOWS DOWNLOAD
Test the refreshed
desktop companion.
0.7.2 teaches the team to take big jobs — and to ask, visibly, before a big one runs. Composer routes one instruction into a single task, a parallel fan-out across agents, or a batch file job. The Run Board shows every run live — who is working, who is blocked, and what needs your call — with hard budget caps enforced in code, so a run pauses at its cap instead of overspending. Bring your own model key (Anthropic, OpenAI, OpenRouter, xAI, Groq, Mistral, DeepSeek, Together) or run fully local with Ollama. Choose x64 for Intel/AMD PCs or ARM64 for Windows on ARM. Setup is per-user and needs no administrator access. The browser extension is a separate ZIP. New in 0.7.2: a plan big enough to need approval now pops up on screen — the instruction, the item count, and the call budget — and nothing runs until you Accept; Decline, and the job is recorded as declined with zero items run. 0.7.2 also adds an in-app update notice, so the app itself tells you when a newer version is out — a notice only; it never downloads or installs anything by itself.
RELEASEv0.7.2 · Free
PLATFORMWindows 10/11 · x64 + ARM64
INSTALLER128.3 MB x64 · 122.1 MB ARM64
05 ON THE ROADMAP
V2 · IN DEVELOPMENT
More voice.
More environments.
V2 is in development now. It turns the floating head into a conversational voice companion — talk to Roboto, he talks back, and you can hand him work by voice. Delegated agents work in pop-up sandbox environments you can watch live, and gated actions surface a visible approval pop-up, so nothing sensitive happens off-screen. Scout goes further: name the sites, and an agent reads them inside hard limits it shows you up front — same site only, page caps, robots rules honored — then brings back a table and a brief you can check against the source pages. Watchers re-check pages you pick on a schedule and tell you what changed; they run while the app is open, and say when they last checked. Packs turn any finished job into a client-ready pack — a summary document plus the results table, in Markdown and CSV, in one click. A Scores screen lets you save a job that worked and re-run it in one action, and Council Mode lets the agents deliberate a big call in the open, against a consensus bar, before anything moves. Underneath it all, an eval and tracing harness keeps every run measurable. The approval pop-up and the in-app update notice shipped sooner, in the free 0.7.2 patch. Mobile remains a planned direction only: Android for true floating-head support; iOS would use system surfaces such as Dynamic Island / Live Activities and wallpaper experiences, not a permanent overlay.
- ✳ Conversational voice companion — talk to Roboto, he talks back, delegate by voice
- ✳ Pop-up sandbox environments you can watch live, plus a visible approval pop-up for gated actions
- ✳ Scout — bounded research runs that read the sites you name and bring back a brief
- ✳ Watchers — scheduled re-checks that report what changed, while the app is open
- ✳ Packs — any finished job becomes a client-ready Markdown + CSV pack
- ✳ Scores screen — save a job that worked, re-run it in one action
- ✳ Council Mode — agents deliberate big calls in the open, against a consensus bar
- ✳ Android floating head; iOS Dynamic Island / Live Activities and wallpaper direction
TALKAsk RobotoVoice or text, in the floating head
DELEGATERoboto routes itScout, Builder and Reviewer take the work
WATCHSee it happenPop-up sandboxes, live on your desktop
APPROVEYou make the callGated actions ask before anything runs
PLANNED EXPANSION
$8.99
ONE-TIME · V2 EXPANSION · NOT AVAILABLE YETThe current Windows desktop download is free. The V2 expansion will be $8.99 one time; it is not available yet and this page has no checkout.
Roadmap preview
No checkout on this page
No checkout on this page
This is a roadmap preview, not an offer to purchase. Features, timing and final terms may change before release.
06 GOOD TO KNOW
A few things
before you start.
Can’t find the answer? Ask the Robot Orchestra head in the corner, or email EDGE | AGENCY.
Do I need Hermes to use Robot Orchestra?
No. Auto mode uses direct Ollama or OpenRouter model routes; Hermes is a separate, optional mode for tool-enabled work. Install an Ollama model first if you want local inference. The app never downloads models or submits prompts automatically.
What does “local-first” mean here?
Robot Orchestra stores its Brain data on your PC, while each task uses the model route you choose. Ollama and OpenRouter are direct routes; Hermes is an explicitly selected optional bridge. Browser and Knowledge context are off until you opt in.
Can I scan a browser page?
The Chrome/Edge companion supports user-triggered, bounded site scanning. Review the captured text first; page text is not sent to a model until you explicitly include it. The browser side panel is optional to the desktop download.
Why might Windows show a warning?
The 0.7.2 installers are unsigned, so SmartScreen may show a warning. Compare the architecture-specific SHA-256 on this page or in SHA256SUMS.txt. The extension ZIP is a separate download.
Is the V2 expansion available now?
No. V2 is in development and is not a current offer; it will be $8.99 one time when it launches. The current Windows desktop app and 0.7.2 are free; this page has no checkout. Mobile directions remain planned only.