Frequently Asked Questions

Agent Robbi is a Personal AI Agent that you install on your own PC. One installer sets up everything you need to run a local Large Language Model (LLM) entirely on your machine, with up to 50 built-in integrations to the apps you already use — email, calendar, web search, code editors, project management, and more.

You buy it once. There’s no monthly fee for the software itself, and no per-token API charges when you run a local model.

Three big differences:

• It runs on your machine. Your conversations, files and models stay local. No prompt is sent to a third party unless you explicitly choose a cloud model.
• It connects to your tools. ChatGPT and friends are chatbots; Agent Robbi is an Agent that can act inside your email, calendar, code repository, project board and more.
• It’s a one-time purchase. $49 buys a lifetime licence with unlimited private AI usage. Cloud subscriptions usually run $20–$240+ per month per service. See the AI Costs page for a full comparison.

The Agent Robbi software is a one-time payment of $49 USD for a lifetime licence. That includes unlimited use of local LLMs on your machine and the full set of built-in integrations.

You can optionally subscribe to Updates for $5 / month for ongoing model catalogue refreshes, new features, and community membership. Updates are optional — the software keeps working without them.

Local model inference is free of charge. If you choose to also use a paid cloud model (e.g. GPT-5, Claude Sonnet, Gemini Pro), you pay that provider directly — Agent Robbi takes no cut.

Agent Robbi runs on Windows, Mac and Linux. The minimum to install is 8 GB RAM and 5 GB free disk; for a comfortable local-LLM experience plan on 16–32 GB RAM and a recent GPU. The installer auto-detects your hardware and recommends a model that’ll run smoothly on it.

For platform-specific details see the Windows, Mac and Linux entries below.

Windows 11 64-bit (recommended) and Windows 10 64-bit version 22H2 or later. ARM-based Windows on Snapdragon X is on the roadmap; for now use the x64 build under emulation if you need it on ARM.

The installer is shipped as AgentRobbi-Setup-x.y.z.exe and runs per-user, so no admin rights or UAC prompts are required.

Minimum (CPU only): 8 GB RAM, 5 GB free SSD, any 64-bit x86 CPU from the last ~6 years (AVX2 support).

Recommended (GPU acceleration): 16–32 GB RAM, NVIDIA RTX with 8 GB+ VRAM (driver 555+) or AMD RX 7000-series with ROCm 24.10+, 10 GB+ free SSD.

Power users: 32 GB+ RAM and a 16 GB+ VRAM GPU lets you run the larger 13B / 30B local models comfortably.

The binaries land in %LOCALAPPDATA%\Programs\AgentRobbi. Models, workspace files, secrets vault and memory database live in %LOCALAPPDATA%\AgentRobbi. Both are per-user, never in Program Files, so a normal Windows account can install, update and uninstall without admin rights.

macOS 13 Ventura or later, on Apple Silicon (M1 / M2 / M3 / M4) or Intel x86_64. Apple Silicon is recommended — the unified memory architecture makes it an excellent local-LLM machine.

The installer is a signed and notarised .pkg file; double-click to install. No admin password is required for the per-user install path.

Minimum: 8 GB unified memory (M1 / M2 base), 5 GB free disk.

Recommended: 16 GB+ unified memory (M2 Pro / M3 / M3 Pro), 10 GB+ free disk — enough headroom for 7B–13B models with comfortable context windows.

Power users: 32 GB+ unified memory (M3 Max / M4 Max / Ultra). Apple’s Metal Performance Shaders backend means even larger 30B models run at usable speeds without a discrete GPU.

The app bundle lives at ~/Applications/AgentRobbi.app (per-user). Models, workspace, secrets vault and memory live under ~/Library/Application Support/AgentRobbi. Uninstall is a drag-to-trash on the app plus removing that support folder.

Any modern x86_64 distribution with glibc 2.35+. Tested on Ubuntu 22.04 / 24.04 LTS, Debian 12, Fedora 39+, Arch (rolling), Pop!_OS, Linux Mint 21+. ARM64 / aarch64 builds are on the roadmap.

Agent Robbi ships as a self-contained .tar.gz archive, an AppImage, and a .deb for Debian/Ubuntu. Pick whichever matches your workflow — all three are equivalent.

Minimum (CPU only): 8 GB RAM, 5 GB free disk, any 64-bit x86 CPU from the last ~6 years (AVX2 support).

Recommended (NVIDIA GPU): 16–32 GB RAM, an NVIDIA card with 8 GB+ VRAM, the proprietary 555+ driver and CUDA 12.4+ runtime.

Recommended (AMD GPU): 16–32 GB RAM, an AMD RX 7000-series with ROCm 6.2+ installed.

Agent Robbi will fall back to CPU inference automatically if the GPU runtime isn’t found, so your install will always work even if the driver stack isn’t perfect yet.

The .deb places binaries under /opt/agentrobbi with a launcher in /usr/local/bin. The tarball / AppImage variants are fully portable — drop the directory anywhere under your home folder and run ./agentrobbi. User data lives in ~/.local/share/agentrobbi and config in ~/.config/agentrobbi following the XDG Base Directory spec.

Yes. When you use a local model, every prompt, every document and every memory stays on your machine. There is no telemetry of conversation content, and there is no “training run” on your data.

If you opt in to a cloud model, the prompt for that specific request is sent to the provider you chose (OpenAI, Anthropic, Google, etc.) under their privacy terms. Agent Robbi never relays your requests through our servers — the connection is direct, browser-to-provider style.

All workspace files, project documents, semantic-memory embeddings and chat history are stored under %LOCALAPPDATA%\AgentRobbi on Windows (and the equivalent paths on Mac and Linux when those ship). They never leave your machine.

Back them up like any other local files — Time Machine, OneDrive, your usual backup tool. Agent Robbi works fine when restored to a new machine; just point the installer at the existing data folder.

You need internet for the initial install and to download your first model. After that, you can work fully offline with local models. Cloud models obviously need internet.

Agent Robbi ships with a curated catalogue of open-source models that work well on real consumer hardware — including Gemma, Qwen, Mistral, Llama and others. The catalogue is refreshed weekly for Updates Pass holders, with picks tailored to your machine’s CPU, RAM and GPU.

Power users can also drop in any GGUF model file and Agent Robbi will run it.

Yes. You bring your own API key, paste it into the local Secrets Vault, and Agent Robbi can route any chat to the provider of your choice. Local and cloud are first-class citizens — you can even mix them within the same conversation.

Agent Robbi never sees your API keys in plain text after you store them, and they never leave the machine.

A Runner is a focused workspace mode — Research, Code, Content Creator, and others — that activates a tailored set of integrations, starter prompts and tool permissions. It’s a one-click way to switch the agent into the right shape for the task without re-explaining context every time.

You can also build your own Runners.

Up to 50 integrations including Email, Web Search, Calendar, Slack, Project Management (Asana, Jira, Linear, Trello), CRM, Customer Support, Code Editors, GitHub / GitLab, Social Platforms and more. The full list ships in-app and is updated each week for Updates Pass subscribers.

One purchase, one machine, indefinitely. The licence covers the version you bought plus every minor patch. You can move the licence to a new machine when you upgrade hardware — just deactivate the old install first.

Major new feature releases roll out under the optional Updates Pass; without it, the version you bought keeps working forever.

• Curated model catalogue updated weekly with picks tailored to your machine
• New features as they ship
• Agent Robbi community membership
• No lock-in — cancel anytime, the software keeps working

Yes — 30-day money-back, no questions asked. Email hello@agentrobbi.com from the address you bought with and we’ll process the refund.

No. The installer ships everything Agent Robbi needs as native binaries. No runtimes, no virtual machines, nothing else to install on the host. It just works.

Yes, very little is locked away. You can change the active LLM, adjust system prompts during onboarding, choose which integrations are enabled, build your own Runners, and edit Agent Robbi’s memory directly. Or just use it as shipped — it’s up to you.

Agent Robbi is commercial software, not open source. The product statically links permissively-licensed open-source components (full attributions on the Third-party page), but the Agent Robbi source is proprietary.

Email hello@agentrobbi.com. Updates Pass subscribers can also use the in-app feedback channel for priority routing.