Valo FAQ — Trust & Transparency

Valo is the workspace for AI—a native desktop app for thinking, creating, researching, organizing, and building with AI. It works with Claude and Codex; local-model support through Ollama is in active development.

You bring the intent. Valo handles the machinery.

Valo has a free plan. Hosted models need access to Claude Code or Codex through your own account or API credits. Local-model support is being built around Ollama on your computer. Valo does not resell AI access.

Valo Pro is $12.99/mo and unlocks context control, codebase intelligence, voice, and the nerve map. See the pricing page for the full breakdown.

The 14-day Pro trial starts inside Valo with no account, credit card, or signup. Paddle is only used if you choose to buy Pro.

Valo is in early access. Builds for Mac, Windows, and Linux download directly from getvalo.dev/download. They are not code-signed yet, so your system shows a first-launch warning that the download page walks you through.

macOS (Apple Silicon and Intel), Windows, and Linux.

Valo itself makes two outgoing connections:

• updates.getvalo.dev — checks for app updates (our Cloudflare R2 bucket)
• licensing.getvalo.dev — license validation when activating a Pro key, plus an anonymous ping at launch (app version, platform, CPU architecture, a random install ID, the furthest first-run setup step you reached, and the provider setup route taken)

No names, no email, no file paths, no message content, no IP address stored. The setup fields are words from fixed lists — steps such as needs-node, wizard-done, first-message and so on — sent again when you reach a later step. It tells us where new users get stuck; it says nothing about what you do in the app.

When you choose Claude, the Claude CLI running underneath connects to Anthropic's servers:

• api.anthropic.com — all AI inference (this is the point)
• claude.ai — OAuth authentication
• statsig.anthropic.com — Anthropic's feature flags and operational metrics
• sentry.io (Anthropic's project) — crash reporting

The last two are Anthropic's telemetry, not ours. You can disable them by setting CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC=1 in your environment.

When you choose Codex, the Codex CLI connects to api.openai.com when using API credentials, or chatgpt.com and auth.openai.com when using a ChatGPT account. Provider traffic belongs to the AI engine you choose, not to Valo.

When you choose GLM, its isolated runtime connects to api.z.ai using the Z.AI key stored in your encrypted Keyguard vault. Those requests go to Z.AI, not through Valo's servers.

Local-model support is in active development. When enabled, inference runs through Ollama on your computer and no request goes to an AI provider.

When you use research agents or web tools, those connections will also appear — they're user-initiated and visible in your chat.

You don't have to trust us — you can verify. Here are three ways to confirm Valo only connects to the domains listed above:

Quick check (30 seconds)

Open a terminal while Valo is running:

• Mac: Run nettop -p $(pgrep -f Valo) -J bytes_in,bytes_out to see every active connection.
• Windows: Open Resource Monitor → Network tab, or run netstat -b in an admin terminal.

Visual monitor (install once)

Use a trusted, well-known network monitor:

• Mac: Little Snitch
• Windows: GlassWire (free tier)

These show every connection any app makes, with graphs and alerts. Point them at Valo and see for yourself.

Firewall lockdown

Allow updates.getvalo.dev, licensing.getvalo.dev, and the domains required by the AI provider you choose. Claude, Codex, and local models use different network paths; a local model needs no provider connection. Your monitor will show anything outside that expected path.

Now — getvalo.dev/download has builds for Mac, Windows, and Linux.

The builds are not code-signed yet, so Windows says it protected your PC and macOS may call the app damaged. Neither is a virus report; both mean the operating system cannot read a certificate. The download page explains each warning, shows which button to press, and publishes a checksum so you can verify the file yourself. If you get stuck, say where in a YouTube comment, and the homepage waitlist is for the signed public release.

Questions go to YouTube. Stuck on setup, wondering whether Valo does a thing, want to argue about a decision — leave a comment on the channel. Valo is built by one person, in public, on that channel. Every comment is read and answered there, and the answer stays up for the next person with the same question.

Bug reports go to Valo Mail. If something is actually broken, report it from inside Valo — the assistant reads the logs, writes the report, and sends it encrypted after you review it. That carries detail a comment can't.

Valo Mail is a drop box for bug reports, not a conversation. Send it and forget it; you will not get a reply, and you don't need to wait for one — reports are read and the fix shows up in an update. If you want an actual answer from a human, that is the YouTube comments.

Projects and conversations are stored on your machine. Valo doesn't upload your code, chat history, or project data to our servers.

The only data that leaves your machine goes to the AI provider you choose (for inference) and, optionally, to our license server (for Pro key validation).