Data, storage and integrations
Data and privacy
Section titled “Data and privacy”Indexing — what reaches search and the agent’s context: notes, meetings and mail as separate switches. A source switched off is invisible to the agent.
Terminal and file system — permission for the agent to run commands and work with files outside the workspace. Enable only if you understand the risks.
Storage
Section titled “Storage”Disk usage — what takes space, cleanup, backups.
Data transfer — export and import. The export shows a summary: notes, tasks, projects, documents, chats, subscriptions, database size, attachments and the total.
Options: encrypt the archive (AES-256, a password of at least 8 characters) and include API keys — with a warning that these are sensitive. Exporting keys is impossible without encryption.
The import checks the archive first: date, version, contents, whether a schema migration is needed, whether the archive is too old. Then you pick a strategy:
| Strategy | What it does |
|---|---|
| Replace everything | existing data is deleted |
| Merge | conflicts are resolved by modification date |
| Add new only | only missing records are imported |
There’s a separate option to back up current data before importing.
Accounts
Section titled “Accounts”IMAP/SMTP, Exchange (with autodiscover or an explicit URL), Google via an app password, Jira and Confluence. Connection errors are specific.
The same page holds mail signatures and calendar subscriptions (.ics) per account: the last sync or the error, “Refresh now”, and deletion that names how many events go with it.
Plugins
Section titled “Plugins”Extensions run in an isolated sandbox with explicit permissions. Installation from a file, a URL or a developer folder.
Before installing you see: the source, the requested access per domain (read, or read and write), what the plugin adds (agent tools, MCP domains, account connections, background jobs, commands) and the signature — verified, unsigned (with a warning) or invalid.
An update lists new permissions, no-longer-needed ones and unchanged ones separately — access is never widened silently. After an update a rollback is available.
Plugin health: working, disabled, error, disabled after repeated crashes.
MCP Server
Section titled “MCP Server”Yttri hands its tools to external AI agents: enable, autostart, transport, host and port, status, connected clients and uptime.
Project context limits access to one project’s data.
API keys: created with a name, copied once — the key is shown a single time — with creation and last-use dates and deletion.
Domain permissions — per domain, whether it’s enabled and the level: read, or read and write, with the tool count. Plus ready-made configurations for popular clients.
Web search
Section titled “Web search”Maximum results per query (1–10) and parallel search across providers.
| Provider | Notes |
|---|---|
| DuckDuckGo | no registration or key, free |
| Jina | returns page content as Markdown |
| Tavily | tuned for AI and RAG |
| Brave | independent search focused on privacy |
| Ollama | web search through the Ollama service |
A key is kept in secure storage, never written into the settings, and after saving it cannot be viewed — only replaced.
Remote access
Section titled “Remote access”Access from a phone through remote.yttri.online (marked as in development).
The order: set up an authenticator (TOTP) → enable access (a local server and an outbound tunnel start; no inbound ports are opened) → scan the QR code → enter the code, which is verified locally.
The Yttri server acts as a router and stores none of your notes, mail, prompts or assistant answers.