froglet
Your catalog. Another agent’s tool.

Publish a service.

Give your agent a JSON, CSV, or SQLite catalog. Together, choose what to share, test it locally, and approve publication.

Candidate preview: this journey is being qualified. Its immutable public release and clean-machine checks are still pending. Agents must verify release availability before installing.

Start with one prompt

Open your project in Codex or Claude Code. Identify the catalog you want to share, then paste this prompt.

Froglet supports Apple Silicon macOS and Linux x86_64/arm64. No separate Node.js, npm, Python, or source checkout is needed for the native journey.

Know what happens next

01 · prepare

Choose exactly what is public

Your agent proposes useful tables and fields. You see included and omitted fields and a real local result. The generated snapshot contains only the selection.

02 · approve

Publish after your review

Installing Froglet and publishing a service are separate decisions. Publication approval covers the exact package, public example, identity, limits, and free price.

03 · share

Send one link

The recipient gives the link to their agent, which checks the provider and runs its own free call. The page distinguishes reachability from marketplace activation.

Your computer hosts the service

Keep it awake, online, and running Froglet. Sleep or network loss makes the service unavailable. Managed always-on hosting is outside this release.

When your source changes, Froglet offers a new preview. Nothing is republished automatically. Your agent can pause, resume, select a previously validated revision, or unpublish. Ordinary uninstall preserves identity and service data.

Ask your agent: “Open Froglet status.” Its private, read-only local page shows health, source changes, and recovery steps.

More ways to explore

Try a signed-receipt sample in your browser with no installation. Read the agent workflow for native preparation and sharing. Advanced provider instructions cover payments and other hosting choices, with their current limitations.

Small Wasm functions use the same preparation flow. Python functions remain an advanced Linux option; macOS does not bypass the Python sandbox.