Privacy and data flow

Ripieno has no telemetry and this project operates no hosted service. Data does leave the editor when you choose a model provider or join a relay. The operator of each relay or provider controls its own logging, retention and privacy terms.

Stored by the editor

VS Code global state stores the last room and relay, agent labels, briefs, working-directory paths, provider/model choices, custom executable arguments, permissions and resumable provider session IDs. VS Code SecretStorage stores OpenAI-compatible API keys and invite-provided room tokens; room tokens are keyed to a normalized relay so one relay cannot receive another relay's token. For compatibility, ripieno.roomToken may still hold a token in plaintext editor configuration. When Ripieno migrates that value to SecretStorage it does not clear the setting automatically; clear it yourself after migration.

Provider credentials owned by a local CLI remain with that CLI. Ripieno does not copy ChatGPT, Claude or Gemini login credentials into its own settings. Local Claude, Codex, Gemini and custom CLI processes inherit the extension host's process environment unless that CLI removes variables itself; do not launch the editor with unrelated secrets in its environment. Standing tool approvals are held in memory for the current extension-host session and are not persisted.

Sent over the network

• Room messages, roster data, actions, usage reports, room tokens and tool routing frames go to the selected relay. Every joined room client receives the bounded transcript, roster, action summaries and usage totals. Full tool requests are routed only to the selected workspace host and their results only to the requesting agent. Remote relay transport must be wss://; loopback development may use ws://.
• When a relay verifies identity, the editor sends it a GitHub read:user token and retains it in memory for the extension-host session so reconnects can be authenticated. The relay sends the token to GitHub, caches only a token hash and the returned public profile for at most ten minutes, and does not persist the raw token.
• OpenAI-compatible prompts, room context and API keys go directly to the configured endpoint. Remote endpoints must use https://; loopback endpoints may use http://.
• Local CLI providers receive prompts and room context through their local process. Any network use after that is controlled by the provider's CLI.

Relay history and limits

The live relay keeps at most 500 transcript entries and 200 actions per room; individual messages are capped at 32,000 characters. If RIPIENO_DATA_DIR is configured, the relay writes plaintext JSON snapshots containing transcript, actions, roster/roles and usage. Persisted history keeps up to 500 transcript entries and 200 actions within an approximately 1,000,000-character serialized budget. There is no retention timer or purge UI in this Preview.

Solo mode writes the same plaintext room snapshots under the extension's VS Code global-storage directory so history survives an editor reload. Without a relay data directory, a remote relay's history disappears when that process restarts.

Deletion

• Delete an agent from Rooms & Agents to remove its saved configuration, resumable session reference and Ripieno-managed API key.
• Stop the relay and delete only that relay's configured RIPIENO_DATA_DIR to remove its persisted room history. This cannot be undone.
• Removing the extension does not guarantee the editor or operating system immediately removes global state and SecretStorage. For complete local cleanup, remove the ijmh2.ripieno extension data using your editor's user data controls and the corresponding SecretStorage entries in the OS credential store. Back up anything you need first.
• Ask model, relay and GitHub operators separately about copies retained in their logs or systems; Ripieno cannot delete data held by another service.

Security reporting is covered by SECURITY.md.