FAQ

Short answers to the questions that come up first. Each one describes something that is live today.

Where do my Bloomy Earth credentials go?

They go into the OAuth browser flow only. You type them on the server's consent page, the server verifies them live against the Bloomy API before issuing a token, and they are sealed encrypted inside that token. They are never written to a host configuration file, a chat message, a log or a tool result.

Which AI hosts have been validated?

Seven surfaces are covered in the installation section: the ChatGPT custom connector, the ChatGPT agent plugin, the Claude app on claude.ai, Claude Code, opencode, Kimi Code CLI and Codex CLI. Six were connected and exercised on 2026-11-04; the Codex page is documentation-verified only and says so plainly.

Does the estimate measure my real emissions?

No. Footprint figures are illustrative estimates, not measurements of electricity consumption or of actual emissions from AI infrastructure, and they are not verified carbon-accounting data. With token counters the estimate uses token-energy coefficients; without them it uses flat-rate per-request constants — either way it reports its mode, the size bucket each entry routed to and client-asserted provenance. The methodology section documents the model in full.

Why do the write tools need a confirmation code?

Because a write spends the signed-in user's own Tree Credits or publishes publicly, and neither can be undone. The first call executes nothing and returns a one-time code plus a summary of exactly what the second call would do; the second call repeats the identical arguments with that code. The floor is enforced by the server, so it holds whatever a host's own approval UI does or does not add on top. The capabilities page lists which tools are write tools.

What happens if a confirmation code is wrong, already used, or expired?

Nothing is spent. A wrong or already-used code is refused with "not executed — the confirmation code was never issued or has already been used. Nothing was spent." An expired code (about five minutes) likewise executes nothing, and a fresh first call mints a new one. Declining works the same way: the declined outcome confirms nothing was spent.

Does my conversation content leave the host?

No. Estimation receives only counters — response counts, or token counts when the host exposes them — and actions receive only the confirmed parameters of that action. The assistant's messages, files and code are never sent to Bloomy Earth.

How do I revoke access?

Disconnect the connector in the host, and revoke or rotate the application password in your Bloomy Earth Dashboard → Integration API settings. Both are effective: the server re-verifies grants against the Bloomy API at refresh time and revokes the token immediately when the credentials are rejected, so a revoked application password cannot keep minting access tokens.

Why does a host keep asking me to confirm, and then the server asks too?

Hosts add their own approval steps on top of the server's floor — a plugin-wide setting on ChatGPT, a per-tool policy on claude.ai, a per-call prompt on Kimi. Those are the host's guardrails, not a second implementation of the server's confirmation. The server's two-call confirmation is the one that actually decides whether anything is spent, and it cannot be removed by a host.