Pongu for macOS · Private preview
Install Pongu, connect a supported AI account you already use, and iMessage your personal agent—without Docker, Terminal, or server wrangling in the normal setup path.
You bring the Mac and the AI. Pongu keeps them connected.
Apple-silicon Mac and macOS 14+ required. Public download remains in qualification.
You bringa spare Mac
You connecta supported AI account
You reach it throughiMessage
Pongu keepsthe connection running
The normal setup path
Pongu turns the usual chain of runtimes, macOS permissions, and messaging helpers into one guided Mac setup.
Put Pongu on an Apple-silicon Mac that can stay on when you want the agent available.
Use a supported AI subscription or API key you already have. Pongu checks the account and Mac permissions.
Choose an allowed Messages conversation and verify the full round trip before setup is considered ready.
The actual macOS app
Pongu shows whether the AI runtime, Messages access, allowed conversations, and local services are ready—and gives you one place to pause or restart them.
Guided readiness
Connect the supported runtime, verify Messages access, test sending, and choose exactly which conversation can reach the agent.
Built for the long conversation
Pongu keeps concise, conversation-scoped memory on the Mac and selects relevant entries with on-device semantic matching.
When the agent answers, the chosen AI provider receives the current request and relevant recalled context.
Already running an agent?
Keep the framework, model, tools, and skills you chose. Pongu adds the native Mac control, iMessage routing, durable state, memory, and diagnostics around them.
Terminal commands, runtimes, paths
Guided framework setup
Helpers, permissions, bridge config
One readiness checklist
Read logs and test each layer
End-to-end delivery proof
Monitor and recover it yourself
Visible status, diagnostics, pause
Edit IDs and configuration files
Conversation allowlists
Compatibility, stated precisely
The native foundation already has adapters for Codex, Claude Code, and Pi. OpenClaw and Hermes Agent are launch targets still being qualified.
Installed Codex harness and existing ChatGPT sign-in
Installed Claude Code harness and its existing authentication
Installed Pi runtime and configured provider access
Exact versions and guided setup path still to be published
Exact versions and transport path still to be published
Allowlisted direct and group conversations, commands, durable queues, deduplication, and safe retry contracts.
Images and files, plus optional local transcription for supported voice messages.
Persistent runtime sessions, chat-scoped memory, local diagnostics, pause, and restart controls.
Clean-host permissions, broad replay and crash recovery, updates, backup, and export remain launch gates.
Seven days, then one clear license
Start with a seven-day trial—no card required. One purchase, no recurring Pongu fee, and product updates included. AI-provider subscriptions, API usage, and connected services remain separate.
No-card trial
Founding access
$20USD · one time
Pongu remains in private preview. Trial and purchase both grant license keys now; the public app download is not yet available, and preview onboarding is handled separately while release qualification continues.
Trust is part of setup
Pongu’s current architecture is a local control plane—not a hosted agent cloud. We are keeping the boundaries explicit while the launch integrations are qualified.
On your Mac
Your provider
Model + toolsUnder its own policiesThe current design runs on an Apple-silicon Mac you control. Pongu does not host the agent runtime.
Messages uses the account signed in on that Mac through a local helper—not an official Apple iMessage API.
Model authentication stays with the selected framework and provider. Pongu does not bundle a model account.
The memory store remains on the Mac. Relevant recalled context is sent to the chosen model when it answers.
Sender allowlists, group controls, redacted diagnostics, and a pause switch are part of the native foundation.
OpenClaw/Hermes adapters, clean-host setup, notarized distribution, updates, recovery, backup, and export.
A useful first text
These launch recipes shape onboarding around a concrete outcome. Choose one to tell us what you want first.
01 · DAILY
02 · MONITOR
03 · RESEARCH
04 · BUILD
Straight answers
We would rather explain a boundary now than hide it behind a launch button later.
The current architecture runs the agent control plane on a Mac you control. It is not a Pongu-hosted, multi-tenant runtime. A packaged public release is still in development.
Yes for the local iMessage path we are building first. The current target is an Apple-silicon Mac on macOS 14 or newer, signed into Messages and available while you want the agent online.
It uses the Messages account signed in on the host Mac. A dedicated macOS user and separate Apple ID may be preferable for agent identity and isolation. The final pairing flow is still being validated.
Yes. You connect the model provider supported by the selected agent framework. Provider usage, policies, and charges are separate from Pongu.
They are the launch integrations we are qualifying, not public compatibility claims yet. Private-preview users will help us validate exact versions, setup paths, and recovery behavior before release.
No universal approval layer is promised. Agent and tool actions follow the selected framework’s policies. Pongu’s native controls include sender allowlists and pause; app-owned recurring work can ask for explicit confirmation.
The agent cannot answer through the local iMessage path while the Mac or Messages is unavailable. Durable delivery and recovery are part of the design, with final real-world qualification still underway.
The frameworks keep their own data, and Pongu’s memory is local. A supported Pongu backup, restore, and export experience is a launch gate—not something we claim is shipped today.
Private preview
Tell us which Mac, AI account, and first workflow you want to use. That helps us qualify the launch around real setups—not a perfect demo machine.